WO2025068263A1

WO2025068263A1 - A system and method for determining a physiological parameter of a body comprising blood perfused tissue

Info

Publication number: WO2025068263A1
Application number: PCT/EP2024/076899
Authority: WO
Inventors: Adrianus Maria Lafort; Alwin Fransen; Peter Christiaan Post
Original assignee: Sonion Nederland BV
Current assignee: Sonion Nederland BV
Priority date: 2023-09-28
Filing date: 2024-09-25
Publication date: 2025-04-03
Anticipated expiration: 2026-03-28

Abstract

A system, method, device and computer program are provided for determining a physiological parameter of a body comprising blood perfused tissue. The system comprises a sensor, or a plurality of sensors. The sensor(s) receive first and second radiation from the blood perfused tissue. The sensor(s) generate a first signal relating to an amount of blood (x(t)), and a second signal relating to a speed of blood (v(t)). An evaluation unit of the system defines first and second signal fractions in the first and second signals, respectively. The evaluation unit performs an intrinsic check whether to accept or reject the signal fractions and optionally uses a cross-check algorithm that has as its input both signal fractions and the result of the intrinsic check. The physiological parameter is determined using, if the cross-check algorithm was performed, the first and/or second signal fraction if accepted by the cross-check algorithm.

Description

A SYSTEM AND METHOD FOR DETERMINING A PHYSIOLOGICAL PARAMETER OF A BODY

COMPRISING BLOOD PERFUSED TISSUE

Field of the invention

The present invention relates to a system and method for determining a parameter of blood flowing in blood perfused tissue, such as blood flowing below the skin of a person or animal. The present invention further relates to a device for use in the system or method and to a computer program for carrying out the method. In particular, the invention relates to the determination of the parameter based on information both from a photoplethysmography (PPG) sensor and a laser Doppler velocimetry (LDV) sensor, whereby the parameter may be determined based on information relating to a blood speed and information related to a blood volume.

Background of the invention

It has been found that PPG information may vary in quality depending on e.g. the distensibility of the blood vessels. In addition, the output of an LDV sensor will also depend on e.g. the distensibility, as a higher speed may be expected in blood vessels which are less flexible. By basing the determination both on information from a PPG sensor and an LDV sensor, a more robust method is obtained and thus a better quality of the determined parameter.

Laser Doppler velocimetry (LDV) is also commonly referred to as laser Doppler flowmetry (LDF) when used for measuring blood flow in a body. The term 'velocimetry' suggests that velocity is measured but the blood flow signal obtained by LDV is in fact a scalar, and contains no direction information. The blood flow signal is therefore in fact related to speed of the blood.

US20220133165 describes the determination of a physiological parameter that is related to the blood pressure from a waveform that is acquired by means of a PPG sensor by using peak detection in the 1st and 2nd time-derivative of the PPG waveform and extracting features from the PPG waveform using the 1st and 2nd time-derivative of the waveform based on the wave reflection theory of the arterial system.

It is known that PPG for obtaining biometrics has limitations:

• It is sensitive to movement of the device/sensor relative to skin/tissue, e.g. during motion/exercise;

• It is sensitive to ambient light and modulated ambient light; and

• Its sensitivity deteriorates when vascular distensibility is decreased, mainly for elderly people and people with prevalence of hypertension. Garrett, A. et al. [2023] Biomedical Optics Express, Vol. 14, No 1, pages 1594-1607 propose simultaneous photoplethysmography (PPG) and blood flow index (BFI) measurement to estimate blood pressure using speckle contrast optical spectroscopy. They correlated various features from PPG or BFI alone or a mathematical combination of BFI and PPG, such as a multiplication of diastolic time as determined from BFI pulse wave forms and systolic time as determined from PPG pulse wave forms. They found that for systolic blood pressure a mathematical combination of BFI and PPG pulse waveform correlated more strongly with blood pressure than PPG alone or BFI alone. For diastolic blood pressure, both BFI alone or PPG alone correlated more strongly with diastolic blood pressure as measured with an arm cuff. Although the study group mainly comprised young healthy subjects with medium skin tones, the correlation leaves much to be desired and is insufficient to reliably determine (systolic and diastolic) blood pressure instead of the known arm cuff method.

WO2023/031927 relates to a wearable physiological monitoring system comprising emitting coherent light in one or more wavelengths onto tissue of an examined subject at a measurement point; simultaneously detecting and determining with at least one light detector and a control unit a pulsating blood flow signal and a pulsating blood volume signal; and determining from both signals a local pulse wave velocity and determine from the local pulse wave velocity a blood pressure measure of the examined subject at the measurement point as a function of time. From a measured PPG signal, first-time derivatives of the PPG signal at two separate time points are determined and combined with an amplitude of a blood flow related pulse wave (PBFv(t)) in a mathematical model to arrive at a blood pressure measure. The model assumes a linear relationship between blood volume and blood flow velocity which according to WO2023/031927 only applies during the first phase of increase of the PPG signal and the PBFv signal.

It would be desirable to be able to accurately measure both systolic and diastolic blood pressure. It would further be desirable to be able to accurately measure blood pressure in subjects with high blood pressure and/or with a low distensibility of blood vessels.

It would further be desirable to be able to be able to measure both systolic and diastolic blood pressure or another physiological parameter of a body comprising blood perfused tissue whilst the body is in motion. of the invention

It is an object of the invention to improve the measurement of one or more physiological parameters, in particular blood pressure, systolic and diastolic blood pressure. Since PPG measures a parameter related to the perturbation of the volume of blood in the tissue, and LDV measures a parameter related to the blood flow speed in, typically the arteries of, the blood perfused tissue, LDV cannot be a 1 : 1 replacement of PPG. Also, LDV may according to one embodiment be determined using radiation penetrating deeper into the tissue, to gain access to relatively larger blood vessels, relative to radiation used for the PPG, which may according to the one embodiment be determined at a shallower depth at the capillaries of the tissue.

Additionally, the blood flow speed is less affected by reduced vascular distensibility, and may be even increasing with reduced distensibility, so that the LDV signal does not deteriorate the same way as the sensitivity of PPG when vascular distensibility is decreased.

Thus, by combining PPG and LDV, or other information representing similar information, in e.g. a wearable device, such device may cover a wider range of users including elderly people and people with prevalence of hypertension where PPG may not function accurately.

According to a first aspect, the present invention relates to a system for determining a physiological parameter of a body comprising blood perfused tissue in accordance with claim 1. According to a second aspect, the present invention relates to a computer-implemented method for determining a physiological parameter of a body comprising blood perfused tissue in accordance with claim 14. According to a third aspect, the present invention relates to a device for use in a system for determining a physiological parameter of a body comprising blood perfused tissue; and/or for carrying out a method for determining a physiological parameter of a body comprising blood perfused tissue. According to a fourth aspect, the present invention relates to a computer program comprising instructions which, when the program is executed by a computer or computing system, cause the computer or computing system to carry out a method for determining a physiological parameter of a body comprising blood perfused tissue, in accordance with claim 15.

Accordingly, according to a first aspect, the present invention provides a system for determining a physiological parameter of a body comprising blood perfused tissue, the system comprising : a sensor, or a plurality of sensors, configured to: receive first radiation from the blood perfused tissue and generate a corresponding first signal comprising first information relating to an amount of blood in the tissue as a function of time (x(t)); and receive second radiation from the blood perfused tissue and generate a corresponding second signal comprising second information relating to a speed of blood in the tissue as a function of time (v(t)); an evaluation unit, or a plurality of evaluation units, configured for evaluating the first and second signals by:

(a) determining at least two triggers (t_x,i and t_x,_i+i) in the first signal and defining the first signal between the two triggers as a first signal fraction; (b) determining at least two triggers (t_v,i and t_v,i+i) in the second signal and defining the second signal between the two triggers as a second signal fraction;

(c) determining in an intrinsic check whether to accept or reject the first signal fraction by comparing the first signal fraction with a first reference; and whether to accept or reject the second signal fraction by comparing the second signal fraction with a second reference;

(d) if in the intrinsic check one or both of the first signal fraction or the second signal fraction have been rejected, and, optionally, if both the first and second signal fractions have been accepted, determining whether to accept or reject both signal fractions, or to reject one signal fraction while accepting the other signal fraction, using a cross-check algorithm that has as its input both signal fractions and the result of the intrinsic check; and a controller configured to determine the physiological parameter using, if the cross-check algorithm was performed, the first signal fraction if accepted by the cross-check algorithm and/or using the second signal fraction if accepted by the cross-check algorithm.

The first and second information relate to different parameters of the body, so that the physiological parameter is based on additional information and not solely information relating to the amount of blood in the tissue or information relating solely to the speed of the blood in the tissue.

When the heart pumps blood through the blood vessels of a person, the blood vessels will expand and contract by the blood pressure pulses of the blood. Thus, the pulse generated by the heart will be extended in time and along the direction of the blood vessels. This will depend also on other parameters, such as the distensibility of the blood vessels. If these are not very flexible, the pulse will remain more concentrated in time so that the pulsed behaviour will be more prevalent also at a distance from the heart compared to a person with more flexible blood vessels. Also, reflections will be seen when the blood passes a bifurcation in the blood vessels, so the behaviour of the blood away from the heart may be quite complex but may, on the other hand, reveal a number of parameters of the blood vessels or cardiac system of the body.

Then, the physiological parameter may in principle relate to any parameter of the blood, the blood vessels or the cardiac system, such as (but not limited to) blood biomarkers, blood analytes, such as glucose level, (peripheral) oxygen saturation (SpO2); heart rate, heart rate variability, breathing rate, blood perfusion of the tissue, blood flow speed, blood pressure, arterial distensibility, user motion, and user activity. The body may be a body of a person or an animal. In principle not only vertebrates such as mammals and avians may be relevant users, but at least larger mammals, such as horses, dogs, hounds or the like may benefit from the use of the present invention.

Blood perfused tissue is tissue which has a supply of blood, often via blood vessels provided in the tissue. The tissue may in principle be any tissue but will normally be tissue not too far from a surface or skin of the user, such as the tissue at or forming at least part of an ear canal, a portion of an ear, a finger, a limb, an arm, a leg, or the like of the body. The tissue may be tissue within 10mm from a surface, such as the outer surface of skin, of the user, such as within 8 mm from the surface.

A sensor is an element which is configured to detect or sense a parameter and output a signal relating to the sensed parameter. Often, the output is correlated with the parameter so that when the parameter varies, the output of the sensor varies. A calibration may be provided from which a quantification of the parameter may be determined from a quantified output of the sensor. The output of the sensor may be analogue or digital and may be a simple signal, such as a voltage, a current, a sine signal or a pulsed signal, or a parameter, such as a number, a frequency of a signal or the like, or may be a more complex signal, such as a series of numbers, a TDM signal, a value determined from a detector or the like.

In principle, the first and second sensors may share a radiation source and/or a detector. Embodiments exist wherein each sensor has a source and a detector. In other embodiments, the sensors have separate sources but share a detector, and it is also possible for the sensors to share a source and have separate detectors or share a detector. It is noted that, as mentioned below, the first sensor may not require a source at all, if the radiation is generated in the blood.

The first sensor is configured to receive the first radiation from the blood perfused tissue and, based on the received first radiation, generate a first signal comprising first information relating to an amount of blood in the tissue as a function of time. In other words, the first sensor generates a first signal indicative of the amount of blood in the tissue as a function of time.

The amount of blood in the tissue may be quantified in a number of ways. In one method, a known amount of radiation is launched into the tissue, where some of the radiation is absorbed by the blood. The absorption may be caused by natural contents of the blood, such as the red blood cells, or may be absorbed by a component, such as a drug, added to the blood. Alternatively, the first radiation may be generated in or by the blood, such as when the blood comprises a radioactive, fluorescent or emitting component. When the absorbing, fluorescing or emitting component is evenly distributed in the blood, the intensity of received radiation correlates to the amount of blood in the tissue. The first radiation may have any desired wavelength or combination of wavelengths depending on the way it is generated or interacts with the blood. If the first radiation is desirably absorbed by the blood, the wavelength may be selected to be one which a component of natural blood, or a component added to the blood, may absorb. Red blood cells typically absorb radiation in the interval 200-1000nm.

When radiating, emitting or fluorescent agents are used, these are preferably biocompatible but may in principle be selected from a wide array of materials, where each material will define the wavelength(s) of the first radiation.

The second sensor is configured to receive the second radiation from the blood perfused tissue. The output from this sensor comprises second information relating to a speed of blood in the tissue. In other words, the sensor generates a second signal indicative of the speed of blood in the tissue as a function of time. A widely used method of determining a speed of biological elements or components is based on the Doppler shift generated when the radiation such as light (or sound or ultrasound) is scattered by an element moving in relation to the sensor. Another manner could be the use of speckles, spots or structured light. The movement, creation or disappearance of a speckle depends on the movement of the surface vis-a-vis a detector and/or an emitter. When structured light, such as a grid or spots, is fed on to an element which moves, the pattern on the element will alter in a way in which the movement may be determined.

A controller is provided. A controller may be a single element or a combination of elements in communication with each other. A controller may be or may comprise a processor, an ASIC, an FPGA, a DSP or the like. The controller may be hardwired, or software controlled, or a combination thereof, and is configured to determine the physiological parameter from the first information and the second information. Clearly, the controller may be configured to perform additional tasks if desired.

As an example, the present system may form part of a hearing aid or hearable for positioning in or at an ear canal and for determining e.g., blood pressure, pulse, activity, or the like of a user. The controller may then also perform other controlling, such as of components of the hearable/hearing aid.

According to another example, the present system forms part of a wearable device positioned around the wrist for determining e.g., blood pressure, pulse, activity or the like of a user. The controller may then also perform other controlling, such as of components of the wearable device. In one embodiment of this example, the present system forms part of a smart device such as a smart watch.

According to yet another example, the present system is to be (removably) attached to the skin by means of a patch, an adhesive plaster. According to yet another example, the present system forms part of a patch, such as a plaster to be (removably) attached to the skin of a body that incorporates another function such as a transdermal patch or a diabetes patch, such as a diabetes patch incorporating a sensor, preferably an optical sensor, adapted for determining insulin or glucose levels in blood.

An evaluation unit is provided. The evaluation unit may be a component separate of the controller, or the evaluation unit and the controller may be embodied as a single module, i.e. a single processing unit, such as a processor, an ASIC, FPGA or DSP.

In one embodiment, the system comprises a first sensor for generating the first signal, wherein the first sensor comprises a radiation source configured to emit radiation into the tissue, and generate the first signal that is indicative of an amount of radiation absorbed and/or scattered in the tissue. It is preferred that the radiation emitted by the first radiation source is incoherent. By using incoherent radiation, the sensor output is less sensitive to relative movement between the sensor and the tissue. Incoherent radiation may be radiation of a single wavelength or of multiple wavelengths.

A preferred type of first sensor is a PPG sensor. PPG sensors are known for determining parameters of blood and they may be made quite compact, which often is an advantage in sensors. In this context, a "PPG" measurement may relate to the amount of blood in the tissue by way of an amount of radiation absorbed thereby - by determining an amount of radiation received from the tissue and, from that radiation, an amount of radiation absorbed in the tissue.

In one embodiment, the system comprises a second sensor for generating the second signal, wherein the second sensor comprises a source of coherent radiation and is configured to generate the second signal based on a Doppler shift determination based on the second radiation received by the second sensor. Doppler shifting is seen also by blood moving inside the tissue, such as when the coherent source emits radiation which is capable of travelling sufficiently far in blood perfused tissue. In this context, infrared or near infrared radiation may be preferred, such as radiation in the wavelength interval of 700-1000nm.

Then, it may be desired that the second sensor comprises a source of coherent radiation. Alternatively, or additionally, it may be desired that the first sensor comprises a source of incoherent radiation. When both an incoherent and a coherent source are employed, a number of advantages are obtained, such as when the source of incoherent radiation does not emit radiation, or at least not a large amount of radiation, at the wavelength at which the source of coherent radiation emits. In this situation, the two sensors may perform their operations simultaneously even when so closely spaced that their radiation enters the same volume of the tissue and/or when the sensors are exposed to radiation from each other. In general, both sensors may emit radiation into the same volume of tissue, but the radiation from one sensor may travel further into the tissue than that from the other sensor. It may, e.g. be desired to obtain PPG/absorption information from the outermost layers of the tissue but the speed/LDV information from deeper lying portions where larger blood vessels lie.

When the second sensor comprises a source of coherent radiation, the second sensor may comprise a VCSEL with integrated photodetector. Then, if the first sensor employs a separate radiation source, such as a source of incoherent radiation, the first sensor may comprise a radiation detector separate from the source of incoherent radiation. A VCSEL (Vertical Cavity Surface-Emitting Laser) has an internal cavity in which the radiation is generated. A photodetector may be integrated into the VCSEL and which may determine a radiation intensity inside that cavity. This sensor is highly suitable for measuring a speed of blood flow but less suitable for measuring a blood volume. Thus, a separate radiation detector may be provided for measuring blood volume.

Preferably, the source of coherent radiation and the source of incoherent radiation may be operated at the same time. In this situation, the first and second radiation may be received at the same time, so that the output of the sensors will relate to the tissue in the same state, such as where the blood flow is high due to a heartbeat, where the blood vessel(s) is/are expanded, or low, such as between heart beats, where the blood vessel(s) is/are more contracted. During the time of a heartbeat period, the blood flow speed will change and the amount of blood in the tissue will change.

It may be desired to determine both the first and second information at the same time, or at least receive the first and second radiation at the same time. Alternatively, it may be desired to receive the first and second radiation at different points of time, such as points of time where the particular type of information is determined the easiest or with the highest precision.

On the other hand, such as if the radiation from one sensor disturbs the other sensor, it may be desired that the source of coherent radiation is operated within a first number of time intervals and the source of incoherent radiation is operated within a second number of time intervals, where no second time interval overlaps with any first time interval. In this manner, the sensors may be operated one at the time, so that one sensor is not disturbed by the radiation from the other. Also, this allows using a single detector for both sensors.

It may, in fact, be possible to determine which source is operating from the output of the sensors or detector(s), so that no timing is required between the operation of the sources and the determining step. For example, it may be possible to use a DC level of the output of a detector to determine which radiation source is on. In situations where two sources of radiation, such as a source of coherent radiation and a source of incoherent radiation, are used, the output intensity of these may differ to a degree where the total intensity detected from the sources differs sufficiently for the received intensity to in itself reveal which source is operating. Often, the output intensity of coherent radiation is lower than that of incoherent radiation.

Intermittent use of the radiation sources may cause a large fluctuation of the amount of light on the sensors, such as on one or more photodetectors, thus requiring a large dynamic range of any preamp(s) used. Alternatively, a preamp may be designed to be self-adaptive to the received DC signal of the photodetector, or adaptable for example by means of a DC compensation circuit comprising memory elements that preserve the required level of compensation for each cycle of the intermittent system. Additionally, the gain setting of the preamp can be designed to be self-adaptive to the received AC signal of the photodetector, or adaptable by means of a control signal that is synchronized with the intermittent drive of the light sources. Further alternatively, dual multiple preamps may be connected in parallel to the same photodetector output, each of which can be optimized for the dynamic range of the photodetector signal during operation of a corresponding light source.

It may then be desired to use the same pre-amp and/or analysis hardware, such as a processor, one or more filters, integrators, differentiators or the like, for the output of both sensors or the photodetector(s) and then provide a selector for selecting the desired sensor/detector output and the corresponding setting for DC compensation or guiding that output to the analysis hardware in sequential time intervals, such as the time intervals of operation of the respective sources.

As mentioned, it may be desired that the first sensor and the second sensor comprise and share a single detector configured to detect the first and second radiation. Alternatively, the first and second sensors may each comprise at least one detector.

Situations also exist where multiple detectors are provided and where a subset of one or more of the detectors provides an output indicative of either blood speed or blood volume and the other detector(s) provide an output indicative of both the blood speed and blood volume.

In one embodiment, the first sensor is configured to generate a first signal based on the received first radiation, by low pass filtering the first signal and providing a low pass filtered first signal. Typically, a low pass filtering will be configured with a corner frequency of 15-250 Hz. An alternative to the low pass filtering is to band pass filter the first signal. Typically, the bandpass filter will have a lower corner frequency from 0.01Hz to 0.2 Hz, and a higher corner frequency of 15-250 Hz.

In the above situation or in another situation, the second sensor is configured to generate a second signal based on the received second radiation, high pass filter the second signal and provide a high pass filtered second signal. Typically, a high pass filtering will be configured with a corner frequency of 100 Hz - 20 kHz. An alternative to the high pass filtering is to band pass filter the second signal. Typically, the bandpass filter will have a lower corner frequency from 100 Hz to 20 kHz, and a suitable higher corner frequency of at most 100 kHz.

An advantage of signals relating to a speed of the blood, such as LDV, over signals relating to the amount of blood in the tissue, such as PPG, is that the speed signal may be derived from interference of Doppler shifted radiation with radiation from the same coherent radiation source. For a speed signal, the blood flow speed related output signal of a photodetector is a wide-band signal, ranging from DC to more than 50 kHz. This is a much broader range than what will be experienced in a signal relating to the variation in the amount of blood in the tissue. Even artefacts caused by relative motion between the tissue and the sensor, or ambient light, will normally be in the low frequency range. Such artefacts may be created by relative motion between the sensor and the tissue, such as due to motion of the user.

Preferably, these lower frequencies of the second signal are rejected. This also has the advantage that the second signal then is less sensitive to motion and ambient light. Also, the higher frequencies of the first signal may be rejected.

In a preferred embodiment, the system comprises a source of coherent radiation and a source of incoherent radiation.

In the preferred embodiment, the system includes a second sensor configured to have the second radiation comprise radiation from, via the tissue, the source of coherent radiation and to no substantial degree from the source of incoherent radiation.

Incoherent radiation may be suppressed or removed at the second sensor by e.g. an optical filter removing wavelengths outside of the wavelength of the coherent radiation.

In the preferred embodiment, the source of coherent radiation and the source of incoherent radiation can be configured to direct the coherent and incoherent radiation to same volume of blood perfused tissue, wherein a first sensor and a second sensor comprise one or more radiation detectors positioned to receive radiation from the volume.

According to an alternative preferred embodiment, both the first and second sensors may be configured to receive radiation from, via the tissue, a source of coherent radiation.

When the first and second signal fraction are derived from radiation from the same tissue region/volume, the first and second signal fraction can be related to each other and can be used to improve the quality of the biometric parameter.

In addition, the first and second signal fraction may be derived for the same point in time or within the same time interval again allowing a relation to exist between the first and second signal fraction.

In another embodiment of the preferred embodiment: the source of coherent radiation may be configured to emit the coherent radiation into a first volume of the blood perfused tissue, the source of the incoherent radiation may be configured to emit the incoherent radiation into a second volume of the blood perfused tissue, the first and second volumes being non-overlapping, and the first sensor and the second sensor comprise one or more radiation detectors positioned to receive radiation from the first and second volumes.

When the volumes are non-overlapping, overlap or interference of the radiation from one sensor on the other may be reduced, which may simplify signal management in the system.

It may be desired that: the first sensor comprises a first radiation source and is configured to receive the first radiation from a first volume of the tissue, the second sensor comprises a second radiation source and is configured to receive the second radiation from a second volume of the tissue, and the first and second sensors are configured to have: no more than 50%, such as no more than 35%, of the first radiation stem from the second radiation source and no more than 50%, such as no more than 35%, of the second radiation stem from the first radiation source.

This is in order to prevent cross-talk or interference of the radiation from one sensor on the other.

In this context, it may be desired that the percentage of radiation (intensity) received by one or both of the first and second sensors from the source of the other sensor is less than 25%, such as less than 20%, such as less than 10%, such as less than 5%, such as less than 1%, of the intensity received from the source of the pertaining source.

In one embodiment, the system further comprises a housing, wherein the first sensor is configured to receive the first radiation travelling in a first direction, where the second sensor is configured to receive the second radiation travelling in a second direction, and wherein the first direction is at least 90 degrees to the second direction. This housing may be configured to be provided in an ear, such as in an ear canal. Alternatively, the housing may be configured to fit around a body part, such as by providing a channel or hole in which the body part fits. In this context, a direction may be a central direction of a field of view of the sensor. The field of view may be defined by a detector and/or one or more optical elements in front of the detector, such as one or more windows, filters, radiation guides, and/or lenses or the like.

When at least 90 degrees exist between the two directions, the volume of the tissue from which radiation may be received may be optimized. It may, however, still be possible to receive radiation from both sources, if two sources are used, even by the same detector, as radiation may travel quite far in blood perfused tissue, depending clearly on the wavelength.

The angle between the directions may be determined by projecting both directions on to a common plane, such as a plane in which at least one direction exists. The angle may be desired more than 90 degrees, such as at least 100 degrees, at least 120 degrees, at least 150 degrees or around 180 degrees. When the angle is 180 degrees, the directions may be directly opposite to each other so that if one direction is up, the other may be down.

As the detectors may reside inside the housing, the directions may penetrate the housing and thus penetrate an optical element, such as a window or a lens, forming part of the housing.

It may be desired that the first and second directions penetrate the housing at opposite surfaces of the housing.

In another embodiment, the first sensor is configured to receive the first radiation in a first field of view, and the second sensor is configured to receive the second radiation in a second field of view, wherein the first field of view and the second field of view do not substantially overlap.

According to an aspect of the invention, the evaluation unit is configured to determine triggers in the first and second signals to define fractions thereof. Preferably, the triggers define a signal fraction that corresponds with a full period of a heartbeat. Many signal processing methods are known for extracting a single period from a periodic signal, and the invention is not limited to a specific one of these methods. A non-exhaustive list of examples for determining the two triggers for the first and/or second signal include: determining time points of rising edges of the signal; determining time points where the signal has a maximum; determining time points where a derivative of the signal has a maximum, determining a zero-crossing of the signal, determining a zero-crossing of the derivative of the signal, determining the time points that an integral of the signal crosses a threshold.

Optionally, the at least two triggers for the first signal (t_x,i and t_x,i+i) are also used as triggers for the second signal, i.e. t_v,i = t_X;i and t_v,i+i = t_x,_i+i, or vice versa. Preferably however, the at least two triggers for the first signal are determined independently of the at least two triggers of the second signal.

As described above, the cross-check algorithm is optional if both signal fractions are accepted in the intrinsic check. In a first configuration of the system, the evaluation unit is configured to bypass the cross-check algorithm if both signal fractions are accepted in the intrinsic check, i.e. the cross-check is not performed in that case. In this first configuration, the controller is further configured to determine the physiological parameter using the accepted first signal fraction and the second signal fraction if both were accepted in the intrinsic check. In a second configuration of the system, the cross-check algorithm is also performed if both signal fractions are accepted in the intrinsic check, i.e. the cross-check algorithm is always performed after the intrinsic check. In this second configuration, the controller is configured to determine the physiological parameter using the first signal fraction if accepted by the cross-check algorithm and/or the second signal fraction if accepted by the cross-check algorithm.

In an embodiment, the evaluation unit is configured to bypass the cross-check algorithm if both signal fractions are accepted in the intrinsic check, and the controller is further configured to determine the physiological parameter using the first signal fraction and/or the second signal fraction if both were accepted in the intrinsic check.

Further embodiments are defined in the dependent claims.

Brief description of the drawings

In the following, preferred embodiments will be described with reference to the drawings, wherein:

Figure 1 illustrates a first embodiment of the device or system according to the invention,

Figure 2 illustrates an example of a front-end signal processing for a device according to the invention,

Figure 3 illustrates another example of a front-end signal processing for a device according to the invention,

Figure 4 illustrates yet another example of a front-end signal processing for a device according to the invention,

Figure 5 illustrates determining two triggers in a first signal; determining a first signal fraction; determining two triggers in a second signal; determining a second signal fraction; and determining a time difference between the first and second signal fraction,

Figure 6 illustrates a part of the evaluation unit part of the system according to the invention, in particular the steps of determining whether to accept to reject a first or second signal fraction in a first intrinsic check, determining whether to accept or reject the first and second signal fraction for determining a physiological parameter in a cross-check unit; updating a reference first and second signal fraction; and determining the physiological parameter,

Figure 7 illustrates a part of the evaluation unit part of the system according to the invention, in particular the steps of determining features of a first or second signal fraction; and determining whether to accept to reject the first or second signal fraction in a first intrinsic check,

Figure 8 illustrates a Windkessel cross-check,

Figure 9 illustrates a Features cross-check,

Figure 10 illustrates generating a synthetic first and second signal fraction from a model,

Figure 11 illustrates different Windkessel approximations, and

Figure 12 illustrates an alternative Windkessel cross-check.

Detailed description of the drawings

A first embodiment of the invention is depicted in Figure 1. The figure shows a cross section view of a body-worn device 10 and a cross-sectional view of a region of blood perfused tissue 30. The device 10 comprises a coherent radiation source 12 (e.g. laser, VCSEL) and an incoherent radiation source 14 (e.g. LED), radiation (122 and 142) from both exposing and penetrating the skin 300 and other parts of the tissue 30 into exposed tissue regions 32 and 34, respectively. In this embodiment, the exposed tissue regions 32 and 34 do not coincide, but this is not essential, and for some biometric parameters it can be advantageous that the exposed regions coincide as much as possible.

Discontinuities of optical properties in the tissue 30 can scatter the radiation in other directions than that of the incident direction, and moving discontinuities 36, e.g. blood cells, moving in blood vessels 301 can Doppler-shift as well as scatter the radiation. A photodetector 16 receives (162) correlated Doppler-shifted and scattered radiation and uncorrelated scattered radiation (and also Doppler-shifted incoherent radiation).

The general amount of radiation absorbed in the tissue 30 may relate to the amount of blood therein, as the absorption of the blood reduces the amount of scattered light, and may be represented by first information contained in the first signal, and the Doppler shifting will relate to a blood speed and may be represented by second information contained in the second signal. All this information may be derived from the output signal 164 of the detector 16.

Thus, the device 10 may be seen as comprising two sensors, a sensor measuring the radiation from the source 12 and sensor measuring the radiation from the source 14, where the two sensors share the detector 16. The deriving of the signals from these sensors will be described further below.

For example, the low-frequency component of the output signal of the photodetector 16 is proportional to the total amount of radiation, while the high-frequency component contains contributions caused by the interference of the Doppler-shifted and non-shifted coherent radiation. Figure 2 shows a block diagram of the front-end signal processing for the device of Figure 1. The main purpose of the front-end signal processing is to derive a first signal 24, containing first information relating to the blood amount or volume, and a second signal 22, containing second information relating to the blood speed, from the photodetector output 164. Different approaches are possible for this. The first is depicted in Figure 2 and can be applied if both radiation sources 12 and 14 are operating continuously, and the front-end signal processing uses known frequency ranges of both signals. The first signal 24, is extracted from the photodetector output 164 by means of a low-pass filter 184, which may be embodied as a bandpass when, as preferred, DC is not desired and thus blocked or filtered-out. The second signal 22, is extracted from the photodetector output 164 by means of a high-pass filter 182, which may be embodied as a bandpass if very high frequencies, often containing only noise, are undesired and thus suppressed.

From the second signal 22, the average Doppler-shift may be obtained by means of calculating the 1st moment, in the element 186, of the spectrum:

where: is the first moment of the spectral density at time t, and f₂ are the limits of the band pass filter, and is the spectral amplitude density of the photodetector output.

The magnitude of the first moment is proportional to the average Doppler shift of the radiation received by the photodetector and proportional to the intensity of the radiation. The first moment may be normalized in order to remove or reduce the dependency of the intensity by dividing it by the average determined over the same frequency band.

wherein: v(t) is the average Doppler shift at time t,

M₀(t) is the average spectral density at time t.

Figure 3 shows another example of a block diagram for a signal processing of the set-up seen in figure 1 and which can be applied if at least one of the sources is operating intermittently.

The extraction of the first signal 24 and second signal 22 from the photodetector output 164 can be performed by time-windowing of the signal processing after the preamp 18. If the output signals of the preamp 18 are digitized, this digitization should also be performed with the same timing.

In figure 3, a selector 181 is seen which is capable of routing the output of the preamp 18 to either of the filters 182 and 184. The selector 181 is controlled by a timing circuit 183 which also operates the sources 12/14 to achieve the intermittent operation. The timing circuit 183 operates source 12 while instructing the selector 181 to forward the output of the preamp 18 to the filter 182 for generation of the second signal 22. At a later point in time, the timing circuit 183 will instead operate source 14 while instructing the selector 181 to forward the output of the preamp 18 to the filter 184 for the generation of the first signal 24.

As is known, intermittent use of the radiation sources 12/14 may cause a large fluctuation of the amount of radiation on the photodetector. This may be managed in a number of ways. In one embodiment, a preamp 18 is used having a large dynamic range. Alternatively, the preamp 18 may be designed to be self-adaptive to the received DC signal of the photodetector 16, or adaptable for example by means of a DC compensation circuit comprising memory elements that preserve the required level of compensation for each cycle of the intermittent system.

Additionally, the gain setting of the preamp 18 can be designed to be self-adaptive to the received AC signal of the photodetector 16, or adaptable by means of a control signal that is synchronized with the intermittent drive of the radiation sources 12/14. Another alternative can be to have multiple preamps 18 connected in parallel to the same photodetector output, each of which can be optimized for the dynamic range of the photodetector signal during operation of a corresponding radiation source.

Instead of using the timing circuit 183, a DC level of the photodetector 16 may be used in order to detect which radiation source 12/14 is active (at least in case the DC is different between operation of the two radiation sources 12/14) and can be used to select the signal processing channel 184 or 182+186, such as using the selector 181, and/or select the setting for DC compensation and/or select the gain setting of the preamp 18 corresponding to the active radiation source 12/14.

Intermittent operation of the radiation sources 12/14 may be advantageous in case a continuous operation thereof would consume too much current from e.g. a battery of the body-worn device 10.

Figure 4 depicts another block diagram for the signal processing, which can be applied if the coherent radiation source 12 comprises a separate detector, such as an integrated detector, forming element 121, such as if it is a VCSEL with integrated photodetector. In general, the coherent radiation from the emitter will be fed to the detector both via and not via the tissue in order for the radiation to interfere (self-mix) and thus create the Doppler information.

When the element 121 is a VCSEL with Integrated Photodetector, the detector is positioned to detect radiation inside the actual laser cavity. Self-mixing will thus take place in the cavity of the VCSEL. When positioned inside the cavity, any incoherent radiation entering the cavity does not interfere with the coherent radiation therein. This incoherent radiation then will only affect the DC level of the photodetector. Thus, the integrated photodetector has a low sensitivity to uncorrelated radiation so that the output of the integrated photodetector and thus element 121 contains mainly the Doppler-shifted signal.

The second signal 22 is obtained by means of the photodetector and the high-pass or bandpass filtering of filter 182 and the 1st moment (of the spectrum) integrator 186.

The first signal 24 is obtained from the now separate detector 16 and the low-pass filter or band-pass filter 184 as described above.

Pre-amps 18 are illustrated. These are preferred, as the output signals from the detectors 121/16 are usually quite weak.

Figure 5 illustrates determining two triggers in a first signal 24, obtainable for example as depicted in figures 2-4; determining a first signal fraction; determining two triggers in a second signal 22, for example obtainable as depicted in figures 2-4; determining a second signal fraction; and the optional step of determining a time difference between the first and second signal fraction. Figure 5 illustrates an embodiment where the triggers define a signal fraction that corresponds with a full period of a heartbeat. In another embodiment, it may be preferred to define a signal fraction between two triggers that relate to only part of the heartbeat period, for example with t_x,r and t_x,i +i defining a characteristic fraction of the first signal that is best suited for reliably determining the physiological parameter.

The first signal 24 comprises first information relating to an amount of blood in the tissue as a function of time (x(t)); and the second signal 22 comprises second information relating to a speed of blood in the tissue as a function of time (v(t));

In an evaluation unit configured for evaluating the first and second signal, at least two triggers (t_X;i and t_x,_i+i) in the first signal 24 are determined. The part of first signal 24 between the two triggers is defined in the evaluation unit as a first signal fraction 50. In other words, the first signal fraction 50 includes all data points of the first signal 24 within the time window between the two triggers (t_X;i and t_x,_i+i).

Further, in the evaluation unit (not shown) at least two triggers (t_v,i and t_v,i+i) are determined in the second signal 22. The part of second signal 22 between the two triggers is defined in the evaluation unit as a second signal fraction 60.

In order to time-correlate the first signal fraction 50 and the second signal fraction 60, a time difference 27 at the first and second signal fraction start i, that is Ati, is calculated as the difference between trigger t_v,i, corresponding with fraction start i in second signal 22 and trigger t_X;i, corresponding with fraction start i in first signal 24. Further, figure 5 depicts determining a time difference 28 at the first and second signal fraction start (i+1), that is At(i+1), by determining the difference between trigger t_v,i+i, corresponding with fraction start (i+1) in second signal 22 and trigger t_x,_i+i, corresponding with fraction start (i+1) in first signal 24. As shown in Figure 5, trigger t_v,i+i defines second signal fraction 60 end in second signal 22 and trigger t_x,i+i defines first signal fraction 50 end in first signal 24. Where in the evaluation of the first signal fraction 50 and the second signal fraction 60, both signal fractions 50, 60 are accepted, the corresponding time differences Ati and At(i+ 1) are added to a plurality of time differences of previously accepted first and second signal fractions for calculating an average time difference. In other words, if both signal fractions 50, 60 are accepted, an ensemble average time difference is updated with the time differences Ati and At(i+1) of the accepted signal fractions.

Figure 6 illustrates a part of the evaluation unit part of the system according to the invention, in particular the steps of determining whether to accept to reject a first signal fraction 50 or second signal fraction 60 in a first intrinsic check 55, 65; determining whether to accept or reject the first and second signal fraction for determining a physiological parameter in a cross-check unit 70; updating a reference first and second signal fraction in ensemble average calculators 58, 68, 71, 72; and determining the physiological parameter 81 in controller 80.

First signal fraction 50 is passed via line 51 to an intrinsic check evaluation 55. In this intrinsic check 55, quality and relevant features of the first signal fraction 50 are determined and compared to a first reference. Embodiments for carrying out the intrinsic check 55 are set out in more detail with reference to Figure 7. The first reference may for example comprise one or more reference values for features of the first signal fraction or a reference signal fraction, such as an ensemble average first signal fraction or a template signal fraction. In the embodiment depicted in Figure 6, the first reference is an ensemble average first signal fraction. The ensemble average first signal fraction may be determined from a plurality of previously accepted first signal fractions, for example at least 3, such as at least 6, at least 10 or at least 15 previously accepted first signal fractions and, typically, at most 200, such as at most 100, at most 50 or at most 30 previously accepted first signal fractions. As will be discussed in more detail below, the previously accepted first signal fractions may be first signal fractions that were accepted in intrinsic check 55 or first signal fractions that were rejected in intrinsic check 55 but subsequently accepted in cross-check 70.

The ensemble average first signal fraction to be used for comparing against a first signal fraction 50, starting at time trigger t_X;i, as depicted in Figure 5, may be calculated according to the following general formula:

In which: n is an integer corresponding with the number of previously accepted first signal fractions included in the ensemble average first signal fraction;

F_a,tx refer to first signal fractions accepted in intrinsic check 55, that started at a time trigger t_x, where t_x may be a time trigger in the first signal 24 as depicted in Figure 5, that is 1 to y time triggers prior to first signal fraction 50, starting at time trigger t_x,i.

F_ra,tx refer to first signal fractions rejected in intrinsic check 55 and subsequently accepted in cross-check 70, that started at a time trigger t_x, where t_x may be a time trigger in the first signal 24 as depicted in Figure 5, that is 1 to y time triggers prior to first signal fraction 50, starting at time trigger t_x,i.

In one embodiment, F_a,_tx may be replaced, fully or in part, by F_aa,_tx. F_aa,tx refers to first signal fractions accepted in intrinsic check 55 and subsequently accepted in cross-check 70, that started at a time trigger t_x, where t_x may be a time trigger in the first signal 24 as depicted in Figure 5, that is 1 to y time triggers prior to first signal fraction 50, starting at time trigger tx,i .

Where one or more prior first signal fractions starting at time triggers t_x within the range of from (/-l) to (/-y) are rejected in intrinsic check 55 and in cross-check 70 (i.e. F_rr,tx), the number, n, of previously accepted first signal fractions included in the ensemble average first signal fraction is less than the theoretical maximum number of previously accepted first signal fractions, y, that may be included in the ensemble average first signal fraction. Thus, y > n. In stable, steady state operation of the system and device according to the invention, where no change in physiological parameter is encountered and no first signal fractions are rejected due to e.g. motion artefacts, y may be equal to n, but usually, y will be greater than n. If the number n of previously accepted first signal fractions is too small as compared to the theoretical maximum y, the ensemble average first signal fraction may itself be rejected as a suitable reference first signal fraction and an alternative reference first signal fraction, such as a template, may be used instead. Typically, y/2 < n < y for the ensemble average first signal fraction to be accepted. Thus, where, by way of example, y is selected to be 20, n may range from 10 to 20.

The above may apply mutatis mutandis to second signal fraction 60. Thus, second signal fraction 60 is passed via line 61 to an intrinsic check evaluation 65. In this intrinsic check 65, quality and relevant features of the second signal fraction 60 are determined and compared to a second reference. Embodiments for carrying out the intrinsic check 65 are set out in more detail with reference to Figure 7. The second reference may for example comprise a reference value for a feature of the second signal fraction or a reference signal fraction, such as an ensemble average second signal fraction or a template signal fraction. In the embodiment depicted in Figure 6, the second reference is an ensemble average second signal fraction. The ensemble average second signal fraction may be determined from a plurality of previously accepted second signal fractions, for example at least 3, such as at least 6, at least 10 or at least 15 previously accepted second signal fractions and, typically, at most 200, such as at most 100, at most 50 or at most 30 previously accepted second signal fractions. As will be discussed in more detail below, the previously accepted second signal fractions may be second signal fractions that were accepted in intrinsic check 65 or second signal fractions that were rejected in intrinsic check 65 but subsequently accepted in crosscheck 70.

The ensemble average second signal fraction to be used for comparing against a second signal fraction 60, starting at time trigger t_v,i, as depicted in Figure 5, may be calculated according to the following general formula:

In which: m is an integer corresponding with the number of previously accepted second signal fractions included in the ensemble average second signal fraction;

G_a,tv refer to second signal fractions accepted in intrinsic check 65, that started at a time trigger t_v, where t_v may be a time trigger in the second signal 22 as depicted in Figure 5, that is 1 to z time triggers prior to second signal fraction 60, starting at time trigger t_v,i.

G_ra,tv refer to second signal fractions rejected in intrinsic check 65 and subsequently accepted in cross-check 70, that started at a time trigger t_v, where t_v may be a time trigger in the second signal 22 as depicted in Figure 5, that is 1 to z time triggers prior to second signal fraction 60, starting at time trigger t_v,i.

In one embodiment, G_a,_tv may be replaced, fully or in part, by G_aa,tv. G_aa,tv refer to second signal fractions accepted in intrinsic check 65 and subsequently accepted in cross-check 70, that started at a time trigger t_v, where t_v may be a time trigger in the second signal 22 as depicted in Figure 5, that is 1 to z time triggers prior to second signal fraction 60, starting at time trigger t_v,i.

Where one or more prior second signal fractions starting at time triggers t_v within the range of from (/-l) to (/-z) are rejected in intrinsic check 65 and in cross-check 70 (i.e. G_rr,tv), the number, m, of previously accepted second signal fractions included in the ensemble average second signal fraction is less than the theoretical maximum number of previously accepted second signal fractions, z, that may be included in the ensemble average second signal fraction. Thus, z > m. In stable, steady state operation of the system and device according to the invention, where no change in physiological parameter is encountered and no second signal fractions are rejected due to e.g. motion artefacts, z may be equal to m, but usually, z will be greater than m. If the number m of previously accepted second signal fractions is too small as compared to the theoretical maximum z, the ensemble average second signal fraction may itself be rejected as a suitable reference second signal fraction and an alternative reference second signal fraction, such as a template, may be used instead. Typically, z/2 < m < z for the ensemble average second signal fraction to be accepted. Thus, where, by way of example, z is selected to be 20, m may range from 10 to 20.

The ensemble average of the first signal fraction and/or the second signal fraction may alternatively be computed as a weighted average, a weighted moving average or an exponential moving average. In particular, more recent signal fractions may be assigned a greater weight than older signal fractions.

Referring again to figure 6, if the first signal fraction 50 is rejected in intrinsic check 55, the rejected first signal fraction is passed via line 56 to cross-check 70. Similarly, if the second signal fraction 60 is rejected in intrinsic check 65, the rejected second signal fraction is passed via line 66 to cross-check 70.

If only the first signal fraction 50 is rejected, the accepted second signal fraction 60 is likewise passed to cross-check 70, via line 64. If only the second signal fraction 60 is rejected, the accepted first signal fraction 50 is likewise passed to cross-check 70 via line 54.

If both the first signal fraction 50 and the second signal fraction 60 are accepted in the intrinsic check 55, respectively 65, the cross-check 70 may be avoided (bypassed) and the accepted first and second signals are passed via line 52,53 and 62,63, respectively, to controller 80 for determining a physiological parameter, output via line 81. Any accepted first signal fraction is then also passed via line 57 to ensemble average calculator 58 to determine an, updated, reference first signal fraction output via line 59 to intrinsic check 55 for a next first signal fraction 50. Similarly, any accepted second signal fraction is then also passed via line 67 to ensemble average calculator 68 to determine an, updated, reference second signal fraction output via line 69 to intrinsic check 65 for a next second signal fraction 60.

Alternatively, also in case both signals are accepted in the intrinsic check 55,65, the accepted signals are passed via lines 54,64 to cross-check 70. The latter may be preferred for practical reasons because of a time difference At normally occurring between time triggers t_x,i and t_v,i, as shown in Figure 5. In addition, performing the cross-check 70 also on accepted signals may be preferred to double check for false positive, type I errors.

The reasons for rejecting one or both signal factions 50,60 in intrinsic check 55,65, may vary widely and may for example be related to motion artefacts, causing the first and/or second signal fraction to be distorted. Determining a physiological parameter from such a signal or signals could produce erroneous results. However, a changed signal fraction relative to its reference signal fraction, may also be rejected for the wrong reasons, as it could be representative of a change in a physiological parameter of interest. It is of paramount importance to avoid overlooking false negative, type II errors where a critical change in a physiological parameter of interest is missed due to wrongly rejecting first and/or second signal fractions. Cross-check 70 will be discussed in more detail with reference to figures 8 and 9, but cross-check 70 aims to reduce the number of false positive and false negative errors. Any first signal fraction 50 that is accepted following cross-check 70 is passed via line 75 to controller 80 and any second signal fraction 60 that is accepted following cross-check 70 is passed via line 76 to controller 80.

Following cross-check 70, any accepted first signal fraction is then also passed via line 77 to ensemble average calculator 71 to determine an, updated, reference first signal fraction output via line 73 to intrinsic check 55 for a next first signal fraction 50. Similarly, any accepted second signal fraction is then also passed via line 78 to ensemble average calculator 72 to determine an, updated, reference second signal fraction output via line 74 to intrinsic check 65 for a next second signal fraction 60.

In controller 80, a physiological parameter is preferably determined using both accepted signal fractions, input via lines 75,76. Where, as set out in more detail below with reference to figures 10 and 11, for a given time fraction t only the first signal fraction or the second signal fraction is accepted, a physiological parameter may be determined using only the accepted signal fraction or the accepted signal fraction and a synthetic parameter representing the rejected signal fraction, derived from a reference signal fraction or from the accepted signal fraction or both.

Figure 7 illustrates a part of the evaluation unit part of the system according to the invention, in particular the steps of determining features of a first or second signal fraction; and determining whether to accept to reject the first or second signal fraction in a first intrinsic check. The first intrinsic check is carried out on the signal fractions themselves, i.e. the intrinsic check of the first signal fraction does not use the second signal fraction and vice versa. Thus, the features intrinsic in the first signal fraction are compared against one another and/or against a reference first signal fraction and the features intrinsic in the second signal fraction are compared against one another and/or against a reference second signal fraction. In other words, the first (respectively second) signal fraction is compared to a first (respectively second) reference, wherein this first (resp. second) reference may comprise: a different feature of the first (resp. second) signal fraction or a predetermined reference value for said feature or a reference first (respectively second) signal fraction. In practice, it may be that the process is continuous, where timing triggers for determining a period (i.e. fraction) in the first or second signal are identified and the first or second signal plus timing triggers is passed to a first intrinsic check. It may be that the first or second signal requires further processing to allow filtering, improvement in signal/noise ratio and the like prior to analysis in a first intrinsic check. Alternatively, a processed discrete first or second signal fraction is analysed in a first intrinsic check.

As used in this specification, the terms 'first signal fraction' or 'second signal fraction' include a continuous first signal or second signal with timing triggers; a separate first signal fraction extracted from the continuous first signal by using the timing triggers or a second signal fraction extracted from the continuous second signal by using timing triggers; and further processed first signal or second signal with timing triggers or discrete fractions thereof.

A first or second signal fraction 90, depicted as 50 or 60 in Figure 6, is passed to analyser 91 in which characteristic features of the first or second signal fraction are determined.

Examples of characteristic features for the first signal fraction include features derived from the signal itself, its first and/or second time derivative and/or third time derivative. Examples of characteristic features for the second signal fraction include features derived from the signal itself, its first and/or second time derivative and/or third time derivative. Characteristic features may include: (i) minimum and maximum amplitude of the first and/or second signal fraction and/or first and/or second derivative of the first and/or second signal fraction; (ii) timing of minimum and maximum amplitude of the first signal and/or second signal fraction and/or first and/or second derivative of the first and/or second signal fraction; and (iii) ratio of AC and DC in a time period, i.e. (maximum amplitude - minimum amplitude)/average amplitude in the time period.

The values of the features determined from the signals, signal fractions or derivatives of the signal fractions may be stored and processed. Statistical parameters may then be generated from these values, which can be used to determine whether to accept or reject the signal fraction (e.g. signal fraction is accepted if its waveshape (amplitude at all samples within the signal fraction) does not differ more than 2 times the standard deviation from the average waveshape), or can be used as reference values for intrinsic or cross checks.

Examples of these statistical parameters are e.g. average amplitude or time, standard deviation or RMS amplitude or time, skewness etc.

Characteristic features of the first or second signal fraction 90 may then be analysed in internal check 92,97,93 and/or in reference check 94. In the internal check 104, characteristic features of the signal fraction 90 may be compared against one another in timing check 92, amplitude check 93 or a combination thereof, depicted in Figure 7 as 97.

By way of example, a first signal fraction may comprise features 1 and 2, such as a feature related to the dicrotic notch or the diastolic peak. Where the first signal fraction only shows feature 1 and not 2, or determination of feature 1 and/or 2 is difficult due to a low signal to noise ratio or movement artefacts, or the signal of features 1 and 2 substantially overlap, or otherwise, it may be that the first and/or second signal fraction is rejected following this internal check 104.

Other examples of features that can be extracted from the first signal are the ratio between the amplitudes of the diastolic peak and the systolic peak, the ratio between the amplitude of the dicrotic notch to the systolic peak, the inflection point area ratio, crest time or other features known in the field of PPG analysis.

In addition to or instead of internal check 104, the first or second signal fraction 90 may be analysed in reference check 94. According to one embodiment, either the internal check 104 is first carried out or the reference check is first carried out and any rejected signal fractions are then re-checked in the other, i.e., reference or internal, check.

In reference check 94, one or more features in the first or second signal fraction determined in analyser 91 may be compared against one or more features in a reference first or second signal fraction. The comparison may for example include a timing check, an amplitude check or a combination thereof. It will be appreciated that a feature of a first or second fraction may also be the characteristic shape of the entire signal fraction or a substantial part thereof. In the comparison, for example, the variance relative to the reference first or second signal fraction may be determined. Where the variance is within set limits, the signal fraction may be accepted and if the variance is outside set limits, the signal fraction may be rejected.

According to one embodiment, the reference first or second signal fraction may be an ensemble average of a number, p, of previously accepted first or second signal fractions or a number, q, of characteristic feature(s) of previously accepted first or second signal fractions. According to another embodiment, the reference first or second signal fraction may be a template. In Figure 7, an ensemble average is depicted as 100 and a template is depicted as 101. According to a preferred embodiment an ensemble average is used. It may however be that during a defined time period, the ensemble average 100 is not sufficiently reliable and it is preferable to use an alternative reference first or second signal fraction such as template 101. It is also an embodiment of the present invention to use both the ensemble average 100 and the template 101 or combinations thereof such as an ensemble average 100 in respect of certain features of the first or second signal fraction and an alternative reference such as template 101 in respect of other features of the first or second signal. The analysis and decision to use the ensemble average 100 or the template 101 or a combination thereof is made in reference signal controller 102, the selected ensemble average and/or template information input into reference check 94 via line 98. The reference signal controller 102 may also receive input from reference check 94 via line 98 as input into the selection of the appropriate reference signal fraction. For example, controller 102 is configured to select the template during an startup phase, in which no accepted signal fractions 90 are yet available. The controller 102 switches to the ensemble average once a predetermined number of accepted signal fractions 90 is available.

In addition to calculating an ensemble average of the waveform, the system may calculate an ensemble average of one or more specific features over the ensemble of accepted signal fractions. For example, the system may calculate an average of the dicrotic notch or diastolic peak of a predetermined number of accepted signal fractions. Alternatively, or additionally, to calculating the average, the system may calculate one more other statistics of said one or more specific features of a predetermined number of previously accepted signal fractions, such as the standard deviation of the specific feature, or the variation, maximum, minimum, median, mode, quantile, kurtosis, skew. In an embodiment, both the average and the standard deviation of a certain feature are tracked by the system. In such an embodiment, the corresponding feature of a newly obtained signal fraction is determined and the difference with the tracked average is calculated. The system then compares said difference with the standard deviation (or variation) to determine whether the difference exceeds the standard deviation. For example, if the difference is more than X times the standard deviation, the system rejects the signal fraction, with X being at least 1, preferably at least 1.5, more preferably at least 2.

Using output from the reference check 94 and/or the internal check 104, in internal check controller 95 the decision is made whether to accept or reject the first or second signal fraction. Any accepted signal fraction may be passed via line 99 to a controller (not shown) for determining a physiological parameter or may be passed to a cross-check (not shown). Any rejected signal fraction may be passed via line 96 to a reject table (not shown) and/or to a cross-check (not shown). Any accepted signal fraction may further be passed to an ensemble average calculator 103 to update ensemble average 100. According to another embodiment, the accepted signal fraction is first passed to a cross-check (not shown) and any accepted signal fraction from that cross-check output to ensemble average calculator 103 to update ensemble average 100. Ways to calculate ensemble averages have been described hereinbefore, with reference to figure 6.

Figures 8, 9 and 12 illustrate embodiments of cross-check 70 in Figure 6. Figure 8 illustrates a Windkessel cross-check and Figure 9 illustrates a features cross-check. Figure 12 illustrates another, alternative, Windkessel cross-check.

With reference to Figure 8, a continuous first signal 105 and timing triggers 107, together defining a first signal fraction, and a continuous second signal 106 and timing triggers 107, together defining a second signal fraction, are input into Windkessel parameter analyser 108. Windkessel parameters determined in Windkessel parameter analyser 108 may be input into statistical analyser 109 and compared against Windkessel statistics 111. Alternatively, or in addition, a variance may be determined in variance analyser 110 as compared with a template and Windkessel limits 112. Output from the variance analyser 110 and I or statistical analyser 109 is input into Windkessel cross-check controller 113 for deciding whether to accept or reject the first and/or second signal fraction.

In the context of the invention, Windkessel statistics may comprise predetermined statistics and/or statistics gathered during operation, as will be described below.

In the Windkessel cross-check controller 113, output from the variance analyser 110 and/or statistical analyser 109, is used in combination with output from internal check controller 95 (figure 7), 55, 65 (figure 6), to decide whether to accept or reject the first and/or second signal fraction. If in the intrinsic check both signal fractions are rejected, for example by comparison with an ensemble average, and in the Windkessel cross-check the relation between the first and second signal is accepted, this may be an indication of a physiological change such as a blood pressure increase or decrease and in the Windkessel cross-check controller 113, both the first and the second signal fractions are accepted.

If in the intrinsic check one signal fraction, for example the first signal fraction, is rejected and the other accepted, then a rejection of the relation between the two signals, is a further confirmation that the rejected signal is rejected for a reason not associated with a physiological change. In the Windkessel cross-check controller 113, the signal fraction that was rejected in the intrinsic check is again rejected and the signal fraction accepted in the intrinsic check is again accepted in the Windkessel cross-check controller 113.

If on the other hand, in the intrinsic check one signal fraction, for example the first signal fraction, is rejected and the other accepted, then an acceptance of the Windkessel relation between the two signals may be an indication that either the first signal fraction was wrongly rejected (false negative) in the intrinsic check, or may indicate a physiological change, or may be an indication that the second signal fraction was wrongly accepted (false positive) in the intrinsic check. In this situation, both signal fractions are rejected in the Windkessel cross-check controller 113 as the output may be unreliable. The controller 80 (figure 6) may receive a signal that the signal fractions cannot be used for determining a physiological parameter with sufficient accuracy.

Optionally, the system is configured to track occurrences of the intrinsic check rejecting at least one signal fraction (e.g. both signal fractions) while the relation between the signal fraction passes the cross-check. For example, the system is configured to count these occurrences and if the count within a predetermined time windows exceeds a predetermined threshold, determines that a physiological change probably occurred. Additionally or alternatively, the system determines that a physiological change probably occurred if the intrinsic check rejects at least one signal fraction while the cross-check is passed for a predetermined number of consecutive signal fractions, e.g. for at least 5-15 consecutive signal fractions. If it is determined that a physiological change probably occurred, based on tracking said reject/pass occurrences, one or more additional check may be performed to determine that the physiological change indeed occurred. These additional checks may avoid mistaking e.g. a sensor error for a physiological change. For example, an additional check determines whether the respective signal fraction(s) are within predetermined limits, as described above with respect to the intrinsic check. If the signal fraction(s) are within the predetermined limits, it is likely that a physiological change occurred (instead of e.g. a sensor error). Another example of an additional check is comparing a number (e.g. 5 - 15) of consecutive rejected signal fractions and determining whether they are similar to each other within certain limits (e.g. within 1-2 standard deviations). If they are determined to be similar within the predetermined limits, it is likely that a physiological change occurred (instead of e.g. a sensor error).

In response to determining that a physiological change occurred (detected by said counting or otherwise), the system may reset the ensemble average calculators (58, 68, 71, 72), such that newly incoming signal fractions are checked against a more representative ensemble average. Optionally, further analysis is performed to determine the nature of the physiological change and/or the system generate an output indicating e.g. in the form of a visual or audible message to the user of the system.

Any accepted signal fraction may be passed via line 114 to a controller (not shown) for determining a physiological parameter. Any accepted signal fraction may further be passed to an ensemble average calculator (not shown) to update an ensemble average signal fraction for use as reference signal fraction. Any rejected signal fraction may be discarded or passed via line 115 to another analyser I controller (not shown), for example for root cause analysis.

Figure 9 illustrates a features cross-check. A continuous first signal 105 and timing triggers 107, together defining a first signal fraction, and a continuous second signal 106 and timing triggers 107, together defining a second signal fraction, are input into feature relation analyser 116 and compared against feature relation limits 117. Output from the feature relation analyser 116 is input into feature relation controller 118 for deciding whether to accept or reject the first and/or second signal fraction. Any accepted signal fraction may be passed via line 119 to a controller (not shown) for determining a physiological parameter. Any accepted signal fraction may further be passed to an ensemble average calculator (not shown) to update an ensemble average signal fraction for use as reference signal fraction. Any rejected signal fraction may be discarded or passed via line 129 to another analyser I controller (not shown), for example for root cause analysis.

In an embodiment, the feature relation analyser 116 determines the period of the first signal 105 as the difference between consecutive triggers for the first signal, t_v+i - t_v, and, likewise, determines the period of the second signal 106 as the difference between consecutive triggers for the second signal. The determination of the period may be based on the two triggers that define a single signal fraction, or the feature relation analyser 116 may be configured to determine the difference of the two triggers for a number of consecutive signal fractions, e.g. 5-10 signal fractions, and determine the average, mode or median of the differences as the period. Subsequently, the feature relation analyser 116 calculates the difference between the period determined for the first signal and the period determined for the second signal, and determines whether the difference exceeds a predetermined threshold. If the difference exceeds the predetermined threshold, the relation between the first and second signal is rejected, otherwise the relation is accepted (although further feature checks may be performed before accepting).

Alternatively or additionally, the analyser 116 determines whether the rising edge of the first signal fraction Xi arrives later than the rising edge of the second signal fraction Vi. If not, the relation between the first and second signal fractions is rejected. In an embodiment, the timing of the rising edges of the first and second signal fractions are determined by the analyser 116. For example, the set of triggers 107 received by the analyser 116 includes triggers for the rising edge of the first and second signal fraction. For example, triggers 107 includes timing triggers t_x,i and t_v,i, for the first and second signal fraction, respectively, and the timing of the rising edge corresponds to the triggers with lowest /', typically 0. For example, the rising edges of the first and second signals are received in timing data 107 as t_x,o and t_v,o. The analyser 116 determines whether t_x,o is greater than t_v,o, optionally the difference between the two rising edges is at least a predetermined minimum value and does not exceed a predetermined maximum value.

Alternatively or additionally, analyser 116 determines and compares the timing and/or amplitude of one or more of the following features: signal foot (a.k.a. diastolic foot, start of upslope of signal), maximum slope point (point of maximum slope towards the systolic peak), anacrotic notch, dicrotic notch, systolic peak, diastolic peak.

In the context of the invention, checks based on features (whether in the intrinsic check or in the cross-check) may use the same one or more features for checking the first signal fraction as for checking the second signal fraction, or may use different one or more features for checking the first signal fraction than for checking the second signal fraction. Specifically, the signal fractions typically originate from a different physical signal (e.g. coherent radiation and non-coherent radiation and/or different wavelengths), such that it may be expedient that the intrinsic check uses a different set of features for determining whether to accept the first signal fraction than for determining whether to accept the second signal fraction.

Moreover, the intrinsic check may use the same or a different feature set than the crosscheck algorithm. Furthermore, the determination of the physiological parameter may be based on one or more features of the accepted signal fraction(s), which are not necessarily the same one or more features used in the intrinsic check and/or cross-check.

Preferably, if both signal fractions are accepted in the cross-check (or in the intrinsic check if both signal fraction are accepted in the intrinsic check and the cross-check is bypassed), the controller uses both signal fractions for determining the physiological parameter. In cases where the cross-check algorithm accepts one signal fraction but rejects the other signal fraction, the physiological parameter may be determined based on said accepted signal fraction and a synthetic version of the other signal fraction. The synthetic version of the other signal fraction may be the respective reference signal fraction, such as the ensemble average calculated for that signal fraction. For example, the reference signal fraction is used instead of the rejected signal fraction for a predetermined period, for example for the next 1-15 signal fractions or during the next 1-2 seconds. Alternatively, according to a preferred embodiment, the synthetic version of the other (rejected) signal fraction is computed based on a model that relates the first signal to the second signal, as will be explained in the following with reference to figures 10 and lla-c.

The first signal is proportional to the blood volume in the perfused tissue, which in turn depends on the distensibility of the arteries or arterioles. When distensibility degrades with age also the first signal deteriorates. On the other hand, the second signal is proportional to the blood flow speed which is less affected by distensibility.

The relation between the blood volume and the blood flow speed is determined by the (local) blood vessel impedance. This relation may be used to derive a synthetic version of the first signal fraction from the second signal fraction, or vice versa.

Both the first signal (e.g. PPG) and the second signal (e.g. LDV) acquired in the blood perfused tissue relate to the arterial blood pressure wave. Both relations are determined by vascular parameters such as diameter, length and distensibility of the connecting blood vessels. These relations can be expressed as transfer functions, that may be non-linear and may vary over time, both on a short term basis due to regulation of physiological mechanisms and on a long term basis such as due to aging or pathological changes.

Referring now to figure 10, the direct relationship between the first and second signals (24 and 22, respectively) can be modelled by means of a more or less complex mathematical model 120. Such a model may be linear or nonlinear, deterministic or non-deterministic etc. An example of a linear model is a differential equation and another example is a non- deterministic model, such as one based on a deep learning model.

The simplest possible model that may be applicable is a first order linear differential equation (may be known as Windkessel model). v t) ~ c₀ + c, • x t) + c₂ • x'(t) wherein: v(t) is the second signal fraction, x(t) is the first signal fraction, x'(t) is time derivative of the first signal fraction, and c_n are the model parameters.

The model parameters may be used to create a reconstructed or synthetic second signal fraction from the first signal fraction by: iw(t) = C_o + . xft) + c₂. x'(t)

The residue from the model fitting procedure may be determined by: r(t) = v(t) — iw(t)

In one embodiment, the residue r(t) may be used as a quality indicator for the Windkessel fit. For example, the root mean square of the residue (RMS(r)) should be much smaller than (such as smaller than 20%, 10% or 5% of) the root mean square of the second signal (RMS(v)).

The model parameters may be time dependent but do not vary within the timeframe of at least one heartbeat or that of a predetermined number of heartbeats or within a predetermined time span.

The model parameters may be calculated continuously over a shifting time window with a duration of at least one heartbeat, for example using a least mean square method. Furthermore, statistics of the model parameters may be calculated continuously over the shifting time window. For example, the average, median, mode, maximum, minimum, standard deviation, variance, moment or other statistic. Particularly, for the Windkessel model these statistics include statistics of the coefficients of the Windkessel model, such as Co, Ci and c₂ in the example above. For higher order models, statistics of further coefficients may be determined on a continuous basis.

Similarly, a reconstructed or synthetic version of the time derivative of the first signal fraction can now be calculated as a linear function of the first signal fraction and the second signal fraction according to:

Wherein xx'(t) is a synthetic version of the time derivative of the first signal fraction. One advantage of using the second signal fraction in addition to the first signal fraction in a model such as a Windkessel model is that the time derivative of the first signal fraction can be derived without a need to calculate a time derivative of the second and first signal fraction. This is particularly advantageous where the SNR of the time derivative of the first signal fraction is undesirably low, creating unreliable or unclear information from that time derivative, which in turn affects the accuracy of the determination of a physiological parameter.

According to one embodiment of the system and method, a Windkessel model is used to calculate the time derivative of the first signal fraction from the second signal fraction and the first signal fraction.

Depending on the accuracy needed a more complex model may be used, e.g. a higher order differential equation, or a non-linear differential equation, or a non-deterministic model, and a larger number of model parameters is determined.

In general, from the model, parameters may be generated for a first transfer function 122 and a second transfer function 124, which parameters may be performed in steps 134 and 132, respectively, for arriving at synthetic first and second signal fraction 24' and 22', respectively. Thus, the signal fraction 24' may be generated solely from the signal fraction 22 or partly therefrom, based on the model 120 and vice versa.

A Windkessel model can be represented as an electrical network, see Figure 11. In this analogy, the charge in the capacitor represents the amount of blood in the tissue, and the input current represents the volume velocity of the blood, which is proportional to the speed. Since the charge is proportional to the capacitance and the voltage, the voltage across the capacitor is proportional to the blood pressure.

Since good results can already be obtained with a first order model, i.e. a two parameter Windkessel model, this will be elaborated in the following, although use of higher order models may improve the accuracy of the method, such as a second order model, i.e. a three parameter Windkessel model. Also, in the following, we will relate to the first signal fraction, which is proportional to the amount of blood in the blood perfused tissue and thus to the volumetric expansion, as PPG information and the second signal fraction as LDV information, where it is clear that this is an example only of how this information may be obtained.

The second order Windkessel model is depicted in Figure 11a wherein: q_;(t) is the volumetric inflow,

Cfi,i ■ P'(t) is the volumetric expansion rate,

P(t)/R_fliO is the volumetric outflow,

P(t) is the (local) blood pressure, P'(t) is the time derivative of the blood pressure,

C_{fl l} is the effective fluidic compliance of the blood vessels, and

R_{fl 0} is the fluidic resistance of the downflow blood vessels.

The volumetric inflow is proportional to the average blood flow speed, which is proportional to the average Doppler shift measured by LDV. The average Doppler shift can be determined by taking the first moment of the spectral density of the photodetector output. v(t) oc M₁(t) where the second signal fraction is normalized with respect to the amplitude of the photodetector output.

wherein: v(t) is the LDV signal, is the first moment of the spectral density, and f₂ are the limits of the frequency band over which the first moment is calculated, and is the spectral amplitude density of the photodetector output.

Figure 11a illustrates a rather simple model, a two-element model. More complex models are seen in figures lib, illustrating a three-element model, and figure 11c illustrating a four- element model. The more elements, the better may the correlation with the blood flow be emulated. The skilled person will know how to alter the above formulas to include the additional elements.

Examples of measurement systems are seen in relation to figures 1-4 illustrating combined sensors for measuring the PPG signal x(t) or first signal 24 and for measuring the LDV signal v(t) or second signal 22.

As described above, the model parameters may be calculated continuously over a shifting time window with a duration of at least one heartbeat, for example using a least mean square method.

Depending on the accuracy needed a more complex model 120 may be used, e.g. a higher order differential equation, or a non-linear differential equation, or a non-deterministic model, and a larger number of model parameters is determined.

After the model parameters are determined, the parameters can be applied to a first transfer function 132 that calculates a synthetic version 22' of the second signal 22 from the first signal 24, and/or to a second transfer function 134 that calculates a synthetic version 24' of the first signal 24 from the second signal 22. This may in one embodiment be useful in case of temporary loss of one sensor, e.g. due to movement of the body comprising blood perfused tissue relative to the sensor, or temporary loss of signal quality, a poor signal to noise ratio, of either the first or the second signal. In a particular embodiment, if only one of the signal fractions is accepted in the cross-check algorithm, the synthetic version of the other signal fraction is computed and the controller uses the single accepted signal fraction and the synthetic version of the other signal fraction for determining the physiological parameter.

Using the same example of a first order linear differential equation, the first and second transfer functions may be written in the frequency domain as (ignoring DC levels) :

wherein: j is the imaginary unit, w = 2nf is the angular frequency, xx(w) is synthetic version of first signal x(w), and is synthetic version of second signal v(w).

Depending on the model that was used to relate the first and second signals, the transfer function may have more parameters, may be non-linear, or non-deterministic.

Figure 12 illustrates another, alternative, Windkessel cross-check. In this Windkessel crosscheck, the quality of the relation between the first and second signal fraction is assessed by creating a synthetic second signal fraction from the first signal fraction; and determining the residue r(t) by subtracting the synthetic second signal fraction from the second signal fraction. With reference to Figure 12, a continuous first signal 205 and timing triggers 207, together defining a first signal fraction, and Windkessel statistics 211, are input into synthetic second signal fraction calculator 220. In calculator 220, the synthetic second signal fraction is determined by using equation: iw(t) = C_o + . xft) + c₂. x'(t)

The model parameters c₀, Ci and c₂ are taken from Windkessel statistics 211 and x(t) and x'(t) from the first signal fraction. The synthetic second signal fraction vv(t) is output to residue calculator 230. Continuous second signal 206 and timing triggers 207, together defining a second signal fraction are also input into residue calculator 230. In residue calculator 230, the residue r(t) is determined by using equation: r(t) = v(t) — vv(t)

The calculated residue r(t) is output to comparator 240 and/or to comparator 250. In comparator 240, r(t) is compared against limits, and/or a template and/or a variance analysis, input via 210. In comparator 250, the root mean square of the residue (RMS(r)) is compared against the root mean square of the second signal (RMS(v)). The second signal is input via 206 and timing triggers via 207. If RMS(r) << RMS(v), the synthetically created second signal fraction from the first signal fraction fits well, is an adequate approximation of the second signal fraction. The result of the comparison in comparator 240 and/or the comparison of RMS(r) and RMS(v) is output to Windkessel cross-check controller 260 for deciding whether to accept or reject the second and/or the first signal fraction. For example, the cross-check controller 260 determines that the relation between the first and second signal fraction is "accepted" if the residue r(t) fulfils the conditions of both the comparator 240 the comparator 250, and is "rejected" otherwise. The cross-check controller 260 then combines this result with the result of the intrinsic check.

If in the intrinsic check both signal fractions are rejected, for example by comparison with an ensemble average, and in the Windkessel cross-check the relation between the first and second signal is accepted, this may be an indication of a physiological change such as a blood pressure increase or decrease and in the Windkessel cross-check controller 260, both the first and the second signal fractions are accepted.

If in the intrinsic check one signal fraction, for example the first signal fraction, is rejected and the other accepted, then a rejection of the relation between the two signals, is a further confirmation that the rejected signal is rejected for a reason not associated with a physiological change. In the Windkessel cross-check controller 260, the signal fraction that was rejected in the intrinsic check is again rejected and the signal fraction accepted in the intrinsic check is again accepted in the Windkessel cross-check controller 260.

If on the other hand, in the intrinsic check one signal fraction, for example the first signal fraction, is rejected and the other accepted, then an acceptance of the Windkessel relation between the two signals may be an indication that either the first signal fraction was wrongly rejected (false negative) in the intrinsic check, or may indicate a physiological change, or may be an indication that the second signal fraction was wrongly accepted (false positive) in the intrinsic check. In this situation, both signal fractions are rejected in the Windkessel cross-check controller 260 as the output may be unreliable. The controller 80 (figure 6) may receive a signal that the signal fractions cannot be used for determining a physiological parameter with sufficient accuracy. Any accepted signal fraction may be passed via line 270 to a controller (not shown) for determining a physiological parameter. Any accepted signal fraction may further be passed to an ensemble average calculator (not shown) to update an ensemble average signal fraction for use as reference signal fraction. Any rejected signal fraction may be discarded or passed via line 280 to another analyser I controller (not shown), for example for root cause analysis.

Optionally, if both signal fractions are rejected by the cross-check algorithm, the controller may be configured to determine the physiological parameter from the reference first signal fraction and the reference second signal fraction for a predetermined period. Alternatively, the controller may be configured to output the previous value of the physiological parameter if both signal fractions are rejected by the cross-check algorithm.

The physiological parameter, such as a systolic blood pressure or diastolic blood pressure, may be determined based on the first signal fraction (xi(t), related to blood volume) and/or second signal fraction (vi(t), related to blood speed) in many different ways. The physiological parameter may be determined using a single first signal fraction and/or a single second signal fraction as described above, or the physiological parameter may be determined based on a set of first signal fractions and a set of corresponding second signal fractions, for example in the form of an ensemble average first signal fraction and an ensemble average second signal fraction. For example, determining the physiological parameter comprises determining timing and/or amplitude of representative features of the ensemble average first signal fraction, and determining timing and/or amplitude of representative features of the ensemble average second signal fraction. Herein, the representative features may be different for the first and second ensemble. Optionally, in the case of determining a timing of a feature of an ensemble average signal fraction, the controller may perform a timecorrection of said timing, based on an ensemble average time difference as described above, e.g. by adding or subtracting said ensemble average time difference from the determined timing.

Moreover, the physiological parameter can be determined even using only one of the first and second signal fraction, i.e. either the first or the second signal fraction. Particularly, if one of the signal fractions is rejected, the physiological parameter may be determined using only the other signal fraction. As described above, this may be done using a synthetic version of the other (rejected) signal fraction or a reference for the other (rejected) signal fraction. However, the invention is not limited thereto. Various techniques for determining a physiological parameter from a single signal are known in the art, e.g. determining a physiological parameter from a single PPG signal. Embodiments of the invention may also use these known techniques to determine the physiological parameter, particularly in case one of the signal fractions is accepted but the other is rejected. It is stressed that such embodiments also provide advantages over the prior art, as the intrinsic check and the cross- check algorithm improve the accuracy of the determined physiological parameter. For example, false positives and false negatives are reduced.

In a first example, a physiological parameter is determined using the first signal fraction and a first derivative of the first signal fraction. The first derivative may be derived from the first signal fraction or from the second signal fraction using the Windkessel model as described above (synthetic version of the first derivative of the first signal fraction). In an example, the first derivative is used for determining timing of the systolic and/or diastolic peak and/or anacrotic notch, whereas the first signal fraction is used for determining amplitudes or areas of the peaks and/or other features. Additionally or alternatively, a second derivative is computed from the (synthetic) first derivative and one or more physiological parameters is derived therefrom.

In a second example, determining a physiological parameter comprises performing a numerical integration of the synthetic version of the first derivative of the first signal fraction, as obtained through the Windkessel model described above, to obtain a synthetic version of the first signal fraction. The numerical integration may for example comprise integration using the trapezoid rule. A weighted sum of the first signal fraction and the synthetic version of the first signal fraction may be calculated, and the physiological parameter, such as blood pressure, is extracted from the result using known methods for PPG signals.

In a third example, a physiological parameter is determined using a machine learning model, such as a neural network. In a specific example, the machine learning model comprises a computer-implemented neural network, such as a deep learning neural network with an input layer, and output layer and at least one hidden layer. The input layer receives the datapoints of each of the signal fractions (or a reference signal fraction when replacing a rejected signal fraction). For example, each datapoint of the first signal fraction and each datapoint of the second signal fraction is fed to a single input node of the input layer. In another example, in a pre-processing step a synthetic version of the first derivative of the first signal fraction is obtained by transforming the second signal fraction using the Windkessel model as described above, and the datapoints of the first signal fraction and the synthetic first derivative of the first signal fraction are used as an inputs to the input layer of the neural network. In addition or alternatively, the first and second signal fractions may be pre-processed to extract characteristic features of the first and second signal fractions, and said features are fed to the input layer. The features may for example relate to a dicrotic notch and/or diastolic peak and/or anacrotic notch of the respective signal fraction. The output layer of the neural network comprises at least one output node, the output of which is indicative of the physiological parameter. For example, the output layer comprises a first output node that outputs a value indicative of the systolic blood pressure and a second output node that output a value indicative of the diastolic blood pressure. The neural network may be trained using supervised, unsupervised or reinforced learning. For example, the neural network is trained with a dataset comprising as input values: first signal fractions, second signal fractions and/or synthetic derivatives, optionally features extracted from the first and second signal fractions, and optionally corresponding physiological parameters, such as systolic and/or diastolic blood pressure.

Embodiments of the invention

1. System for determining a physiological parameter of a body comprising blood perfused tissue, the system comprising : a sensor, or a plurality of sensors, configured to: receive first radiation from the blood perfused tissue and generate a corresponding first signal comprising first information relating to an amount of blood in the tissue as a function of time (x(t)); and receive second radiation from the blood perfused tissue and generate a corresponding second signal comprising second information relating to a speed of blood in the tissue as a function of time (v(t)); an evaluation unit, or a plurality of evaluation units, configured for evaluating the first and second signals by:

(a) determining at least two triggers (t_x,i and t_x,_i+i) in the first signal and defining the first signal between the two triggers as a first signal fraction;

(b) determining at least two triggers (t_v,i and t_v,i+i) in the second signal and defining the second signal between the two triggers as a second signal fraction;

(d) if in the intrinsic check one or both of the first signal fraction or the second signal fraction have been rejected, and, optionally, if both the first and second signal fractions have been accepted, determining whether to accept or reject both signal fractions, or to reject one signal fraction while accepting the other signal fraction, using a cross-check algorithm that has as its input both signal fractions and the result of the intrinsic check; and a controller configured to determine the physiological parameter using, if the cross-check algorithm was performed, the first signal fraction if accepted by the cross-check algorithm and/or using the second signal fraction if accepted by the cross-check algorithm. 2. System according to embodiment 1, wherein the evaluation unit is configured to bypass the cross-check algorithm if both signal fractions are accepted in the intrinsic check, and the controller is further configured to determine the physiological parameter using the first signal fraction and/or the second signal fraction if both were accepted in the intrinsic check.

3. System according to embodiment 1 or embodiment 2, wherein the intrinsic check of step (c) comprises comparing :

• the timing and/or amplitude of representative features in the first signal fraction against the timing and/or amplitude of corresponding representative features of a reference first signal fraction; and/or

• the waveform of the first signal fraction or a portion thereof against the waveform of the reference first signal fraction or a portion thereof; and comparing :

• the timing and/or amplitude of representative features in the second signal fraction against the timing and/or amplitude of corresponding representative features of a reference second signal fraction; and/or

• the waveform of the second signal fraction or a portion thereof against the waveform of the reference second signal fraction or a portion thereof.

4. System according to embodiment 3, wherein the reference first signal fraction is an ensemble average of previously accepted first signal fractions; and the reference second signal fraction is an ensemble average of previously accepted second signal fractions.

5. System according to embodiment 4, wherein the ensemble average first signal fraction is determined from at least 5 accepted first signal fractions and the ensemble average second signal fraction is determined from at least 5 accepted second signal fractions.

6. System according to any one of the preceding embodiments 3-5, wherein the first signal fraction is rejected if one or more features Ci,i-_n of the first signal fraction deviates from one or more corresponding features of the reference first signal fraction by a difference Di,i-_n.

7. System according to embodiment 6, wherein the difference Di,i-_n is at least one standard deviation, preferably at least two standard deviations, from the average of the one or more features of the reference first signal fraction.

8. System according to any one of the preceding embodiments 3-7, wherein the second signal fraction is rejected if one or more features C₂,i-_n of the second signal fraction deviates from one or more corresponding features of the reference second signal fraction by a difference D₂,i-_n. 9. System according to embodiment 8, wherein the difference D₂,i-_n. is at least one standard deviation, and preferably at most two standard deviations, from the average of the one or more features of the reference second signal fraction.

10. System according to any one of the preceding embodiments, wherein the evaluation unit is further configured to time-correlate the first signal fraction and the second signal fraction using at least one trigger from the first signal fraction and at least one trigger from the second signal fraction.

11. System according to embodiment 10, wherein time-correlating the first and the second signal fraction comprises subtracting t_x,i from t_v,i ,or vice versa, to determine a time difference between the first signal fraction and the second signal fraction at trigger i (At); and subtracting t_x,i+i from t_v,i+i ,or vice versa, to determine a time difference between the first signal fraction and the second signal fraction at trigger i+1 (At_i+i) and, where in the evaluation of the first signal fraction and the second signal fraction defined by triggers i and i+1 both are accepted, adding time differences (At) and (At+i) to a plurality of time differences At between previously accepted first and second signal fractions and calculating an ensemble average time difference, and wherein determining the physiological parameter comprises using the ensemble average time difference.

12. System according to embodiment 11, as far as depending on embodiment 4, wherein the controller is configured to use the ensemble average time difference to correct timing of representative features of the ensemble average of previously accepted first signal fractions and/or timing of representative features of the ensemble average of previously accepted second signal fractions, preferably by adding or subtracting the ensemble average time difference.

13. System according to any one of the preceding embodiments, wherein in case of a rejected first signal fraction and an accepted second signal fraction, the controller is configured to determine the physiological parameter from the second signal fraction and the reference first signal fraction for a predetermined period; or, in case of an accepted first signal fraction and a rejected second signal fraction, the controller is configured to determine the physiological parameter from the first signal fraction and the reference second signal fraction for a predetermined period.

14. System according to any one of the preceding embodiments, wherein in case of a rejected first signal fraction and a rejected second signal fraction, the controller is configured to determine the physiological parameter from the reference first signal fraction and the reference second signal fraction for a predetermined period.

15. System according to embodiment 13 or 14, as far as dependent on embodiment 4, wherein during the predetermined period a plurality of rejected first signal fractions and/or rejected second signal fractions are compared to determine whether to calculate a new ensemble average first signal and/or a new ensemble average second signal.

16. System according to any one of the preceding embodiments in which the cross-check algorithm comprises the steps of:

(i) inputting the first signal fraction and the second signal fraction, together with the result of the intrinsic check;

(ii) identifying the timing and/or amplitude of representative features in the first signal fraction and/or the timing of the first signal fraction or a portion thereof; the timing and/or amplitude of corresponding representative features of the second signal fraction and/or the timing of the second signal fraction or a portion thereof;

(iii) determining a difference selected from a timing and/or amplitude difference between representative features in the first signal fraction and the second signal fraction; and/or a timing difference between the timing of the first signal fraction or a portion thereof and the timing of the second signal fraction or a portion thereof;

(iv) comparing the difference against a reference and calculating a deviation from the reference;

(v) determining whether to accept or reject both signal fractions, or to reject one signal fraction while accepting the other signal fraction, based on (1) the output of step (iv) and (2) the result of the intrinsic check, wherein: if the output of step (iv) is that the deviation from the reference exceeds a predetermined threshold; and the result of the intrinsic check is that the first and second signal fractions were either both accepted or both rejected, then both signal fractions are rejected; if the output of step (iv) is that the deviation from the reference exceeds the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then only the signal fraction rejected in the intrinsic check is rejected and the other signal fraction accepted; if the output of step (iv) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that both the first and second signal fractions were either both accepted or both rejected, then both signal fractions are accepted; and if the output of step (iv) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then both signal fractions are rejected. 17. System according to any one of the preceding embodiments 1-15, in which the crosscheck algorithm comprises the steps of:

(ii) fitting the first signal fraction onto the second signal fraction or the second signal fraction onto the first signal fraction with a Windkessel model and determining Windkessel model parameters;

(iii) comparing the Windkessel model parameters against a reference and calculating a deviation from the reference;

(iv) determining whether to accept or reject both signal fractions, or to reject one signal fraction, based on (1) the output of step (iii) and (2) the result of the intrinsic check, wherein: if the output of step (iii) is that the deviation from the reference exceeds a predetermined threshold; and the result of the intrinsic check is that the first and second signal fractions were either both accepted or both rejected, then both signal fractions are rejected; if the output of step (iii) is that the deviation from the reference exceeds the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then only the signal fraction rejected in the intrinsic check is rejected and the other signal fraction accepted; if the output of step (iii) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that both the first and second signal fractions were either both accepted or both rejected, then both signal fractions are accepted; and if the output of step (iii) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then both signal fractions are rejected.

18. System according to embodiment 17, wherein the reference comprises: (i) statistics of Windkessel model parameters of previously accepted relations between the first and second signal fractions; and/or (ii) a template and Windkessel model parameter limits.

19. System according to any one of the preceding embodiments 1-15, in which the crosscheck algorithm comprises the steps of:

(i) inputting the first signal fraction and the second signal fraction, together with the result of the intrinsic check; (ii) calculating a synthetic second signal fraction from the first signal fraction or calculating a synthetic first signal fraction from the second signal fraction with a Windkessel model and template Windkessel model parameters;

(iii) subtracting the synthetic first or second signal fraction from the first of second signal fraction and determining a residue r(t) comparing the residue r(t) against a reference and calculating a deviation from the reference;

20. System according to embodiment 19, wherein the template Windkessel model parameters are averages of previously accepted Windkessel model parameters.

21. System according to embodiments 19 or 20, wherein step (iii), comparing the residue r(t) against a reference and calculating a deviation from the reference comprises: (a) comparing residue r(t') against an ensemble average of previously accepted residues and calculating a deviation; and/or (b) calculating a root mean square of the residue (RMS(r)) and subtracting RMS(r) from the root mean square of the second signal fraction (RMS(v)).

22. Computer-implemented method for determining a physiological parameter of a body comprising blood perfused tissue, comprising : receiving a first signal from a sensor, or a plurality of sensors, for measuring radiation from a blood perfused tissue, the first signal comprising first information relating to an amount of blood in a blood perfused tissue as a function of time (x(t)); and receiving a second signal from the sensor, or the plurality of sensors, the second signal comprising second information relating to a speed of blood in the tissue as a function of time (v(t)); evaluating the first and second signals by:

(d) if in the intrinsic check one or both of the first signal fraction or the second signal fraction have been rejected, and, optionally, if both the first and second signal fractions have been accepted, determining whether to accept or reject both signal fractions, or to reject one signal fraction while accepting the other signal fraction, using a cross-check algorithm that has as its input both signal fractions and the result of the intrinsic check; and determining the physiological parameter using, if the cross-check algorithm was performed, the first signal fraction if accepted by the cross-check algorithm and/or using the second signal fraction if accepted by the cross-check algorithm.

23. Method according to embodiment 22, comprising bypassing the cross-check algorithm if both signal fractions are accepted in the intrinsic check, and determining the physiological parameter using the first signal fraction and/or the second signal fraction if both were accepted in the intrinsic check.

24. Method according to embodiment 22 or embodiment 23, wherein the intrinsic check of step (c) comprises comparing :

25. Method according to embodiment 24, wherein the reference first signal fraction is an ensemble average of previously accepted first signal fractions; and the reference second signal fraction is an ensemble average of previously accepted second signal fractions.

26. Method according to embodiment 25, wherein the ensemble average first signal fraction is determined from at least 5 accepted first signal fractions and the ensemble average second signal fraction is determined from at least 5 accepted second signal fractions.

27. Method according to any one of the preceding embodiments 24-26, wherein the first signal fraction is rejected if one or more features Ci,i-_n of the first signal fraction deviates from one or more corresponding features of the reference first signal fraction by a difference Di,i-_n.

28. Method according to embodiment 27, wherein the difference Di,i-_n is at least one standard deviation, preferably at least two standard deviations, from the average of the one or more features of the reference first signal fraction.

29. Method according to any one of the embodiments 24-28, wherein the second signal fraction is rejected if one or more features C₂,i-_n of the second signal fraction deviates from one or more corresponding features of the reference second signal fraction by a difference D₂,i-_n.

30. Method according to embodiment 29, wherein the difference D₂,i-_n. is at least one standard deviation, and preferably at most two standard deviations, from the average of the one or more features of the reference second signal fraction.

31. Method according to any one of the embodiments 22-30, further comprising timecorrelating the first signal fraction and the second signal fraction using at least one trigger from the first signal fraction and at least one trigger from the second signal fraction.

32. Method according to embodiment 31, wherein time-correlating the first and the second signal fraction comprises subtracting t_x,i from t_v,i ,or vice versa, to determine a time difference between the first signal fraction and the second signal fraction at trigger i (At); and subtracting t_x,i+i from t_v,i+i ,or vice versa, to determine a time difference between the first signal fraction and the second signal fraction at trigger i+1 (At_i+i) and, where in the evaluation of the first signal fraction and the second signal fraction defined by triggers i and i+1 both are accepted, adding time differences (At) and (At+i) to a plurality of time differences At between previously accepted first and second signal fractions and calculating an ensemble average time difference, and wherein determining the physiological parameter comprises using the ensemble average time difference.

33. Method according to embodiment 32, as far as depending on embodiment 25, wherein the ensemble average time difference is used to correct timing of representative features of the ensemble average of previously accepted first signal fractions and/or timing of representative features of the ensemble average of previously accepted second signal fractions, preferably by adding or subtracting the ensemble average time difference.

34. Method according to any one of the embodiments 22-33, wherein in case of a rejected first signal fraction and an accepted second signal fraction, the physiological parameter is determined from the second signal fraction and the reference first signal fraction for a predetermined period; or, in case of an accepted first signal fraction and a rejected second signal fraction, the physiological parameter is determined from the first signal fraction and the reference second signal fraction for a predetermined period.

35. Method according to any one of the embodiments 22-34, wherein in case of a rejected first signal fraction and a rejected second signal fraction, the physiological parameter is determined from the reference first signal fraction and the reference second signal fraction for a predetermined period.

36. Method according to embodiment 34 or 35, as far as dependent on embodiment 25, wherein during the predetermined period a plurality of rejected first signal fractions and/or rejected second signal fractions are compared to determine whether to calculate a new ensemble average first signal and/or a new ensemble average second signal.

37. Method according to any one of the embodiments 22-36, in which the cross-check algorithm comprises the steps of:

(iv) comparing the difference against a reference and calculating a deviation from the reference; (v) determining whether to accept or reject both signal fractions, or to reject one signal fraction while accepting the other signal fraction, based on (1) the output of step (iv) and (2) the result of the intrinsic check, wherein: if the output of step (iv) is that the deviation from the reference exceeds a predetermined threshold; and the result of the intrinsic check is that the first and second signal fractions were either both accepted or both rejected, then both signal fractions are rejected; if the output of step (iv) is that the deviation from the reference exceeds the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then only the signal fraction rejected in the intrinsic check is rejected and the other signal fraction accepted; if the output of step (iv) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that both the first and second signal fractions were either both accepted or both rejected, then both signal fractions are accepted; and if the output of step (iv) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then both signal fractions are rejected.

38. Method according to any one of the embodiments 22-36, in which the cross-check algorithm comprises the steps of:

39. Method according to embodiment 38, wherein the reference comprises: (i) statistics of Windkessel model parameters of previously accepted relations between the first and second signal fractions; and/or (ii) a template and Windkessel model parameter limits.

40. Method according to any one of the embodiments 22-36, in which the cross-check algorithm comprises the steps of:

(ii) calculating a synthetic second signal fraction from the first signal fraction or calculating a synthetic first signal fraction from the second signal fraction with a Windkessel model and template Windkessel model parameters;

(iii) subtracting the synthetic first or second signal fraction from the first of second signal fraction and determining a residue rft); comparing the residue r(t) against a reference and calculating a deviation from the reference;

(iv) determining whether to accept or reject both signal fractions, or to reject one signal fraction, based on (1) the output of step (iii) and (2) the result of the intrinsic check, wherein: if the output of step (iii) is that the deviation from the reference exceeds a predetermined threshold; and the result of the intrinsic check is that the first and second signal fractions were either both accepted or both rejected, then both signal fractions are rejected; if the output of step (iii) is that the deviation from the reference exceeds the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then only the signal fraction rejected in the intrinsic check is rejected and the other signal fraction accepted; if the output of step (iii) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that both the first and second signal fractions were either both accepted or both rejected, then both signal fractions are accepted; and

If the output of step (iii) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then both signal fractions are rejected.

41. Method according to embodiment 40, wherein the template Windkessel model parameters is an ensemble average of previously accepted Windkessel model parameters.

42. Method according to embodiments 40 or 41, wherein step (iii), comparing the residue r(t) against a reference and calculating a deviation from the reference comprises: (a) comparing residue r(t') against an ensemble average of previously accepted residues and calculating a deviation; and/or (b) calculating a root mean square of the residue (RMS(r)) and subtracting RMS(r) from the root mean square of the second signal fraction (RMS(v)).

43. Device for use in a system according to any one of embodiments 1-21.

44. Device according to embodiment 43, wherein the device is a wearable device and comprises the sensor, or the plurality of sensors, and a communication module to transmit the first and second signals to an external device comprising the evaluation unit and controller.

45. Device according to embodiment 43, comprising the evaluation unit and/or the controller, wherein the evaluation unit is configured to receive the first and second signals from a further device comprising the sensor, or the plurality of sensors.

46. Computer program comprising instructions which, when executed by a computer or computing system, cause the computer or computing system to carry out the method of any one of embodiments 22-42.

47. Computer readable medium storing the computer program of embodiment 46.

Claims

2. System according to claim 1, wherein the evaluation unit is configured to bypass the cross-check algorithm if both signal fractions are accepted in the intrinsic check, and the controller is further configured to determine the physiological parameter using the first signal fraction and/or the second signal fraction if both were accepted in the intrinsic check.

3. System according to claim 1 or claim 2, wherein the intrinsic check of step (c) comprises comparing :

4. System according to claim 3, wherein the reference first signal fraction is an ensemble average of previously accepted first signal fractions; and the reference second signal fraction is an ensemble average of previously accepted second signal fractions.

5. System according to claims 3 or 4, wherein: the first signal fraction is rejected if one or more features Ci,i-_n of the first signal fraction deviates from one or more corresponding features of the reference first signal fraction by a difference Di,i-_n, wherein the difference Di,i-_n is at least one standard deviation, preferably at least two standard deviations, from the average of the one or more features of the reference first signal fraction; and/or the second signal fraction is rejected if one or more features C₂,i-_n of the second signal fraction deviates from one or more corresponding features of the reference second signal fraction by a difference D₂,i-_n, wherein the difference D₂,i-_n. is at least one standard deviation, and preferably at most two standard deviations, from the average of the one or more features of the reference second signal fraction.

6. System according to any one of the preceding claims, wherein the evaluation unit is further configured to time-correlate the first signal fraction and the second signal fraction using at least one trigger from the first signal fraction and at least one trigger from the second signal fraction, wherein time-correlating the first and the second signal fraction comprises subtracting t_x,i from t_v,i, or vice versa, to determine a time difference between the first signal fraction and the second signal fraction at trigger i (At); and subtracting t_x,i+i from t_v,i+i, or vice versa, to determine a time difference between the first signal fraction and the second signal fraction at trigger i+1 (At_i+i) and, where in the evaluation of the first signal fraction and the second signal fraction defined by triggers i and i+1 both are accepted, adding time differences (At) and (At+i) to a plurality of time differences At between previously accepted first and second signal fractions and calculating an ensemble average time difference, and wherein determining the physiological parameter comprises using the ensemble average time difference.

7. System according to any one of the preceding claims, wherein in case of a rejected first signal fraction and an accepted second signal fraction, the controller is configured to determine the physiological parameter from the second signal fraction and the reference first signal fraction for a predetermined period; or, in case of an accepted first signal fraction and a rejected second signal fraction, the controller is configured to determine the physiological parameter from the first signal fraction and the reference second signal fraction for a predetermined period.

8. System according to any one of the claims 1-7, in which the cross-check algorithm comprises the steps of:

(v) determining whether to accept or reject both signal fractions, or to reject one signal fraction while accepting the other signal fraction, based on (1) the output of step (iv) and (2) the result of the intrinsic check, wherein: if the output of step (iv) is that the deviation from the reference exceeds a predetermined threshold; and the result of the intrinsic check is that the first and second signal fractions were either both accepted or both rejected, then both signal fractions are rejected; if the output of step (iv) is that the deviation from the reference exceeds the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then only the signal fraction rejected in the intrinsic check is rejected and the other signal fraction accepted; if the output of step (iv) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that both the first and second signal fractions were either both accepted or both rejected, then both signal fractions are accepted; and if the output of step (iv) is that the deviation from the reference does not exceed the predetermined threshold; and the result of the intrinsic check is that one of the first and second signal fractions were rejected, then both signal fractions are rejected.

9. System according to any one of the claims 1-7, in which the cross-check algorithm comprises the steps of:

10. System according to claim 9, wherein the reference comprises: (i) statistics of Windkessel model parameters of previously accepted relations between the first and second signal fractions; and/or (ii) a template and Windkessel model parameter limits.

11. System according to any one of the preceding claims 1-7, in which the cross-check algorithm comprises the steps of:

12. System according to claim 11, wherein the template Windkessel model parameters are averages of previously accepted Windkessel model parameters.

13. System according to claims 11 or 12, wherein step (iii), comparing the residue r(t) against a reference and calculating a deviation from the reference comprises: (a) comparing residue r(t') against an ensemble average of previously accepted residues and calculating a deviation; and/or (b) calculating a root mean square of the residue (RMS(r)) and subtracting RMS(r) from the root mean square of the second signal fraction (RMS(v)).

14. Computer-implemented method for determining a physiological parameter of a body comprising blood perfused tissue, comprising : receiving a first signal from a sensor, or a plurality of sensors, for measuring radiation from a blood perfused tissue, the first signal comprising first information relating to an amount of blood in a blood perfused tissue as a function of time (x(t)); and receiving a second signal from the sensor, or the plurality of sensors, the second signal comprising second information relating to a speed of blood in the tissue as a function of time (v(t)); evaluating the first and second signals by:

(a) determining at least two triggers (t_x,i and t_x,i+i) in the first signal and defining the first signal between the two triggers as a first signal fraction;

(c) determining in an intrinsic check whether to accept or reject the first signal fraction by comparing the first signal fraction with a first reference; and whether to accept or reject the second signal fraction by comparing the second signal fraction with a second reference; (d) if in the intrinsic check one or both of the first signal fraction or the second signal fraction have been rejected, and, optionally, if both the first and second signal fractions have been accepted, determining whether to accept or reject both signal fractions, or to reject one signal fraction while accepting the other signal fraction, using a cross-check algorithm that has as its input both signal fractions and the result of the intrinsic check; and determining the physiological parameter using, if the cross-check algorithm was performed, the first signal fraction if accepted by the cross-check algorithm and/or using the second signal fraction if accepted by the cross-check algorithm.

15. Computer program comprising instructions which, when executed by a computer or computing system, cause the computer or computing system to carry out the method of claim 14.