GB2512304A

GB2512304A - Apparatus and method for estimating energy expenditure

Info

Publication number: GB2512304A
Application number: GB1305387.1A
Authority: GB
Inventors: Ksawery Wieczorkowski-Rettinger; Su-Shin Ang; Miguel Hernandez-Silveira
Original assignee: Toumaz Healthcare Ltd
Current assignee: Toumaz Healthcare Ltd
Priority date: 2013-03-25
Filing date: 2013-03-25
Publication date: 2014-10-01
Also published as: GB201305387D0

Abstract

An energy expenditure estimating apparatus is disclosed with an accelerometer input configured to receive accelerometer data from an accelerometer and an ECG input configured to receive ECG data from an ECG sensor. A parameter estimation module is configured to calculate an ECG-derived respiration rate and/or other parameters including, heart rate above sleep, frequency-domain heart rate variability, accelerometer activity count, using the ECG and accelerometer data. There is further an energy expenditure estimating apparatus which estimates energy expenditure using the parameters received from the parameter estimation module.

Description

Apparatus and Method for Estimating Energy Expenditure

TECHNICAL FIELD

The invention relates to apparatus and a method for estimating a user's energy expenditure.

BACKGROUND

Estimation of energy expenditure in humans can be used for a variety of reasons, including to prevent or treat obesity or to help a person manage their health. A number of different techniques have been developed in order to estimate the energy expenditure of a person whether at rest, undertaking low intensity exercise or undertaking high intensity exercise.

Known methods of energy expenditure assessment include activity questionnaires, direct calorimetry, indirect calorimetry and the noncalorimetric doubly labelled water technique. Unfortunately, these methods suffer from several limitations. In particular, the accuracy of activity questionnaires is affected by such factors as subject compliance, misreporting and miscoding of activities, as well as inaccurate estimation of activity intensity and duration. Similarly, the assessment of energy expenditure using the calorimetric and doubly-labelled water techniques is limited due to the high cost, confinement and intrusiveness of such systems, as well as the logistical complexity and expertise required to maintain them. Moreover, certain forms of indirect calorimetry (e.g. closed-loop spirometry) are unsuitable for the monitoring of energy expenditure even at moderate exercise intensities, while the doubly-labelled water technique is limited to the assessment of average metabolic rate only.

In order to overcome limitations of the aforementioned techniques, methods of physical activity expenditure estimation (PAFE) based on physiological and biomechanical parameters derived from sensors have been developed. Initially, the sole use of two physiological and biomechanical parameters, heart rate (HR) and accelerometer activity counts (AAC) that are known to correlate well with physical activity intensity (PAl) were used to estimate energy expenditure. However, it soon became apparent that PAI/PAEE models based on either one of these two parameters are affected by a number of issues. In particular, HR is influenced by such factors as mental stress and hydration of the subject, and the HR-PAl relationship becomes progressively nonlinear allow exercise intensities. Similarly, although the AAC-PAI relationship is accurate when considering low exercise intensities and may be used to effectively distinguish between activity and non-activity, it becomes progressively more nonlinear and imprecise at higher exercise intensities. Consequently, there have been attempts to use both of the physiological and biomechanical variables when estimating PAl and PAEE model.

One method of integrating Iwo variables to estimate PAEE is to use a multiple linear regression PAEE model, however such algorithms were found to require individual subject calibration. The requirement for individual subject calibration means that the models are less readily applicable to a user outside a controlled environment such as the laboratory. To ameliorate this problem a branch equation model was proposed by Brage et al.. The branch equation model selectively utilises HR and AAC to estimate energy expenditure depending on the intensity of the activity performed by a user. This method was found to provide reasonable estimates of PAl and PAEE under laboratory conditions without requiring individual calibration.

In addition to the use of the physiological and biomechanical variables activity classification is also used when estimating PAl and PAEE. Activity classifiers are commonly based on classification trees, which utilise variables derived from portable EGG and inertial sensors, to distinguish between different activities. The classification of an activity may then be used, in conjunction with such variables as AAC and an activity-based regression model, to estimate energy expenditure. An example of the use of activity classification in energy estimation is described in EP10701735.

One example of a known device for estimating energy is described in US6997882.

US6997882 describes a portable device and method for the estimation of PAEE using multiple axis accelerometry, average head rate and a neural network model.

Another example of a known device for estimating energy is described in US7502643.

In US7502643 the device monitors of human heart-related parameters and uses one of two energy expenditure estimation models. The first model makes use of multiple-axis accelerometry, the heat flux and galvanic skin response of the subject, and an activity classifier (which is able to distinguish between resting and active states) to determine PAEE. The second model makes use of multiple-axis accelerometry, average heart rate and an activity classifier (which is able to distinguish between resting, active, stressful and vehicular motion events) to determine PAFE.

A footwear system is described in ER1 0744384. The footwear system is for weight and posture allocation, physical activity classification and energy expenditure calculation.

The energy expediture calculation uses foot sole pressure and accelerometry.

However, current algorithms based on these parameters tend to provide erroneous estimates of energy expenditure, particularly at low exercise intensities and during activities involving isolated limb motion (e.g. stationary cycling), limiting their applicability outside a controlled environment. These discrepancies result from the sensitivity of HR to factors such as a person's hydration, stress, any chemicals, such as caffeine, that the person has ingested, the inability of inertial sensors to detect low impact activities, as well as the consideration of exercise intensity levels only (as opposed to the type of activity performed).

SUMMARY

According to an aspect of the present invention there is provided an energy expenditure estimating apparatus. The energy expenditure estimating apparatus includes an accelerometer input configured to receive accelerometer data from an accelerometer and an electrocardiogram (EGG) input configured to receive EGG data from an FCC sensor. The energy expenditure estimating apparatus further includes a parameter estimation module configured to calculate one or more estimated parameters using at least one of the accelerometer data and the EGG data, the one or more estimated parameters including an EGG-derived respiration rate. An energy expenditure module of the energy expenditure estimating apparatus receives one or more of the one or more estimated parameters from the parameter estimation module and calculates an estimated energy expenditure using at least the received EGG-derived respiration rate. By using the ECG-derived respiration rate instead of the traditional method of estimating respiration rate by calculating V02 the accuracy of estimation of energy expenditure values such as PAl is increased at high activity intensities.

The energy expenditure estimating apparatus may also include an activity classification module configured to determine the active state of the user from the one or more estimated parameters and a memory including a plurality of algorithms each algorithm associated with an active state. The energy expenditure module selects an algorithm from the plurality of the algorithms to calculate the estimated energy expenditure from a plurality of algorithms dependent upon the determined active state of the user OR to select an estimated energy expenditure calculated by the algorithm associated with the determined active state from a plurality of estimated energy expenditures, each estimated energy expenditure being calculated by one of the plurality of algorithms.

This selection of algorithms enables an algorithm which weights each of the different parameters according to the activity intensity being undertaken thereby increasing the accuracy of energy estimation further.

The energy expenditure estimating apparatus may include a memory including a first algorithm to be used when the FOG-derived respiration rate is determined to be valid and a second algorithm to calculate energy expenditure when the FOG-derived respiration rate is determined to be invalid wherein the energy expenditure module is configured to select one of the first and second algorithm dependent upon whether the FOG-derived respiration rate received from the parameter estimation module is valid.

This enables energy estimation to still occur even when FOG-derived respiration waveforms have become corrupted as result of FOG signal artef acts, and a respiration rate value cannot be output.

The energy expenditure estimating apparatus may include an activity classification module configured to determine the active state of the user from the one or more estimated parameters and a memory including a plurality of first and second algorithms, each pair of first and second algorithms being associated with an active state wherein the energy expenditure module is configured to select an algorithm to calculate the estimated energy expenditure from a plurality of algorithms dependent upon the determined active state of the user and whether the EOG-derived respiration rate is valid or invalid OR to select an estimated energy expenditure calculated by the algorithm associated with the determined active state and validity of the EOG-derived respiration rates from a plurality of estimated energy expenditures, each estimated energy expenditure being calculated by one of the plurality of algorithms.

The energy expenditure estimating apparatus may calculate the EOG derived respiration rate using ORS peak locations in order to acquire values of instantaneous heart rate, deriving the respiratory waveform using the values of instantaneous heart rate and calculating the EGG derived respiration rate by detecting peaks in the derived respiratory waveform.

The energy expenditure estimating apparatus may, alternatively, calculate the EGG derived respiration rate by detecting peaks in the derived respiratory waveform comprises calculating the EGG derived respiration rate from the change in amplitude of QRS peaks.

The estimated parameters may include one or more of heart rate above sleep, heart rate variability (HRV) and accelerometer activity counts (AAC).

The estimated energy may be physical activity intensity and calculated using the equation: PAl =(W11f(AAC)) +(Wf2(HR))+(WJ(EDR))+(W14g1(HR V1)) +... + where W is a coefficient and f() and gQ are a regression equations for the parameter wherein the coefficient and regression equation are dependent upon at least the activity type.

The HRV parameters may, advantageously be, time-domain HRV parameters. The use of time-domain HRV parameters increases the accuracy of energy estimations at low activity intensity when compared to energy estimations using frequency-domain HRV parameters. The time-domain HRV parameters may be one or more of the standard deviation of NN intervals (SDNN), standard deviation of differences between successive NNs (SDSD) and the square root of the mean squared difference of successive NNs (RMSSD).

According to another aspect of the present invention there is provided a system configured to output an estimated energy expenditure comprising an accelerometer, an EGG sensor and an energy expenditure estimating apparatus. The energy expenditure estimating apparatus includes an accelerometer input configured to receive accelerometer data from an accelerometer and an electrocardiogram (EGG) input configured to receive ECG data from an ECG sensor. The energy expenditure estimating apparatus further includes a parameter estimation module configured to calculate one or more estimated parameters using at least one of the accelerometer data and the FCC data, the one or more estimated parameters including an FOG-derived respiration rate. An energy expenditure module of the energy expenditure estimating apparatus receives one or more of the one or more estimated parameters from the parameter estimation module and calculates an estimated energy expenditure using at least the received ECO-derived respiration rate. By using the FOG-derived respiration rate instead of the traditional method of estimating respiration rate by calculating V02 the accuracy of estimation of energy expenditure values such as PAl is increased at high activity intensities.

The accelerometer and EGG sensor may be configured to transmit data to the energy expenditure estimating apparatus over a wireless connection.

According to a further aspect of the present invention there is provided a method to estimate energy expenditure. The method includes: receiving accelerometer data from an accelerometer; receiving electrocardiogram (FCC) data from an FOG sensor; calculating one or more estimated parameters using at least one of the accelerometer data and the ECO data, the one or more estimated parameters including an ECO-derived respiration rate; and calculating an estimated energy expenditure using at least the received EGG-derived respiration rate.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 is a flow diagram showing steps according to an embodiment of the invention; Figure 2 illustrates steps according to an embodiment of the invention; Figure 3 illustrates a parameter estimation module according to an embodiment of the invention; and Figures 4 and 5 illustrate alternative embodiments of an energy expenditure calculation module according to an embodiment of the invention.

DETAILED DESCRIPTION

As discussed above, where the physical activity undertaken by the user is of low physical intensity inaccuracies may occur when estimating energy expenditure. A method of improving energy estimation at low exercise intensities is described below with reference to Figures 1 to 5.

The exemplary embodiment of the invention is shown in the flow diagram of Figure 1.

Si. A device including a heart monitor to measure FOG and an accelerometer is attached to a user. At predetermined time intervals a processing unit in the device receives an FOG measurement and tn-axial acceleration (Xaac, Yaac and Zaac) from the heart monitor and accelerometer respectively. The device also includes a memory including a measure of resting heart rate.

S2. The received FOG measurements and tn-axial acceleration are input into a parameter estimation module, such as that illustrated in Figure 3. The parameter estimation module estimates physiological and biomechanical parameters of the user as discussed below with reference to Figure 3. The estimated parameters include heart rate above sleep (HRas), AAO, and EOG-derived respiration rate (EDR).

S3. The parameter estimation module also classifies the active state of the user.

Olassification of the active state of the user may be achieved using any suitable activity classification algorithm such as that described in FP1 0701 735 and incorporated by reference herein.

54. The parameter estimation calculation module receives the estimated parameters and activity type for the time interval from the parameter estimation module.

35. The apparatus includes, in a memory a plurality of algorithms for calculating estimated PAl and PAEE, each of the plurality of algorithms being associated with a different activity classification. The energy expenditure calculation module upon receiving the estimated parameters uses each of the algorithms stored within the memory to calculate an estimated PAl and/or PAFF. The apparatus then selects the estimated PAl and/or PAEE calculated by the algorithm associated with the activity type determined by the apparatus for output. This is described in more detail with reference to Figures 4 and 5.

S5. The estimated PAl and/or PAEE of the user is output by the energy expenditure calculation module. It will be understood that the estimated PAl and/or PAEE may be output to any suitable destination. For example, the estimated PAl and/or PAFE may be output to a screen for visually displaying data, an audio speaker, a memory for storage and retrieval at a later date. Alternatively, the estimated PAl and/or PAEE may be output to a device output for transmission over a wired or wireless co nnection to a separate device such as a computer or cellular phone for immediate display or storage and later display.

The above steps are shown diagrammatically in Figure 2.

Figure 3 illustrates a detailed outline of the exemplary embodiment of a parameter estimation module of an energy estimation apparatus.

As discussed above, the parameter estimation module 1 receives ECG and tn-axial acceleration (Xaac, Yaac and Zaac) data for time interval of predetermined length from a heart rate monitor and accelerometer. The parameter estimation module 1 also retrieves the resting heart rate (RHR) of the user from a memory of the apparatus (not shown).

The parameter estimation module 1 includes the complete set of algorithms, and intermediate parameters used, in order to derive the final set of parameters utilized in the estimation of PAl and FAEE. The algorithms utilized in the parameter estimation module may include, but are not limited to, a HR algorithm 3, HR normalization algorithm 4, AAG algorithm 5, EDR algorithm 6, HRV algorithm 7, and the activity classification algorithm 8. The computations performed by each component of the module are outlined below.

The HR algorithm 3 receives raw EGG epochs for the time interval. The average heart rate may be calculated using any suitable method but is preferably calculated in the units of beats per minute (bpm).

The HR algorithm 3 also calculates the ORS complex peak locations and peak amplitudes of the EGG epochs for the time interval. Several methods may be used to calculate the QRS complex peak locations and peak amplitudes. One example is the Open Source ECG Algorithm (OSEA), which is computationally efficient and has a high level of accuracy.

Advantageously, in order to reduce the impact of inter-subject variability on the relationship between HR and PAl the raw heart rate value is normalized with reference to the sleeping heart rate value (SHR). The HR normalization algorithm 4 performs this normalization of the heart rate value. The SHR value may be approximated as 10 beats below the RHR. The normalized value of heart rate, referred to as HRas, may then be obtained by subtracting the SHR from the present value of HR.

The AAC algorithm 5 receives raw tn-axial accelerometer epochs for the time interval, and calculates a single value of accelerometer activity counts. This algorithm is comprised of three main signal processing stages. In the first stage, each accelerometer channel is de-trended in order to remove the static component of gravity from each signal. The signals are then band-passed filtered (between 0.25 and 8 Hz) in the second processing stage, in order to reduce the effects of noise. Finally, the three preprocessed signals are rectified, integrated and summed together, to generate a single AAG value.

The EDR algorithm 6 employs two techniques in order to obtain EGG-derived respiration waveforms from the ORS complex peak values and peak locations acquired using the HR algorithm. The first technique involves the use of the QRS peak locations in order to acquire values of instantaneous HR. These values are then used to derive the respiratory waveform using the principle of respiratory sinus arrhythmia. The second technique exploits the fact the magnitude of the EGG waveform is modulated as the heart rotates during the breathing cycle, and that the respiratory waveform may therefore be derived from the change in amplitude of QRS peaks. Following the derivation of both waveforms, the most appropriate signal is selected using a confidence indicator, and a value of respiration rate is derived using a peak detector, which identifies inhalation and exhalation events within the selected trace.

The confidence indicator may be, for example, calculated using the mean absolute deviation (MAD) of the peak-to-peak intervals within the signal. The MAD may be calculated using any suitable method, such as that described in 0B2488316A which is incorporated herein by reference.

The HRV algorithm 7 utilises the QRS complex peak locations acquired using the HR algorithm 3, in order to derive several indices of HRV. For this purpose, the ORS peak locations are processed in a number of steps. Firstly, the differences between individual peak locations are used in order to construct an R-R interval time series. The time-series is then detrended and filtered using an impulse-rejection filter in order to remove ectopic beats and artefacts, which may impede the estimation of HRV. The resultant trace, referred to as the N-N interval tachogram, is then passed through an additional signal interpolation and re-sampling stage, in order to derive frequency-domain indices of HRV.

The activity classification algorithm 8 utilised in the parameter estimation module 1 uses the HRas, AAC and Yacc variables in order to determine the active state of the user, however other variables may be used as well as, or instead of, those listed.

Several techniques can be employed in the implementation of the activity classifier, including automated decision trees and neural networks. For example, a decision tree may be constructed using a training dataset consisting of the relevant set of metrics and class labels indicating the correct activity type. The resulting tree consists of several nodes, including the root node, intermediate node and leaf nodes. The root node and internal nodes of the tree represent tests on the value of the variables used, while the branches of the tree represent the outcomes of these tests. The leaf nodes of the tree represent the final outcome of the branching process, and are the class labels corresponding to different activity types. As an example, possible class labels could include resting, stepping, cycling and walking/running activities.

Figure 4 illustrates a detailed outline of the exemplary embodiment of an energy expenditure calculation module 2 of an energy estimation apparatus.

The energy expenditure calculation module 2 receives the estimated parameters and the activity classification for the time period from the parameter estimation module 1.

The energy expenditure calculation module 2 runs an energy expenditure algorithm.

The energy expenditure algorithm is based on a model comprising multiple branches (Activity: 1, Activity: 2... Activity: N), where each branch corresponds to the active state of the user (determined from the output of the activity classification algorithm).

Each branch contains an appropriate set of empirically-derived regression equations, relating each of the parameters to PAl. Each regression equation is a single linear equation comprising a sum of weighted values of one or more of the parameters. The energy expenditure calculation module 3 inputs the received parameters into each of the algorithms and selects the result determined by the algorithm corresponding to the activity type received from the parameter estimation module 1.

The determined result may be output in any suitable way as discussed with reference to Figure 1.

Optionally, each activity may be provided with two algorithms as illustrated in Figure 5.

In such an embodiment an additional branching stage based on the output of the EDFI algorithm occurs. This step is required as at certain points in time, ECG-derived respiration waveforms may become corrupted as result of ECG signal artefacts, and a respiration rate value is not output by the algorithm. At these points in time, the EDR variable cannot be used in the estimation of energy expenditure, and must therefore be excluded from further analysis.

The coefficients for each parameter in each regression equation (indicated by f in Figure 4) and coefficients (indicated by W' in Figure 4) are specific to the type of activity performed and are used to define the way in which the regression equations are combined, in order to obtain a final estimate of energy expenditure. The empirical techniques that are used to determine the relationship between each variable and PAl, as well the set of branch-specific coefficients, are outlined below.

The derivation of activity-related regression equations and branch-specific coefficients involves conducting controlled experiments, during which subjects undergo several pre-defined activities over a wide range of exercise intensities. In these experiments, EGG and accelerometer recordings should be collected concurrently from each subject, together with corresponding values of energy expenditure acquired using a reference device. An example of a reliable reference device is the indirect calorimeter, where energy expenditure is estimated as a function of oxygen consumption (Va2) and carbon dioxide production (VGa2). Following these experiments, the resultant datasets should be partitioned according to activity type. From these sub-datasets, the relationships between the various indicators and the PAl can be obtained by means of least square regression techniques. The set of branch-specific coefficients can be determined by further partitioning the datasets according to the validity of the EDFI output, and using such automatic optimization techniques as simulated annealing.

As an example, a set of possible regression equations and coefficients is outlined in Tables 1, 2 and 3. The relations were derived by utilising an experimental dataset acquired from subjects performing four different activities, including resting, stepping, cycling and walking/running. The variables that were considered included HRas, AAC, the time-domain HRV measurements of the standard deviation of NN intervals (SDNN), standard deviation of differences between successive NNs (SDSD) and the square root of the mean squared difference of successive NNs (RMSSD), and EDR. Each set of branch-specific coefficients was determined using the simulated annealing optimisation process.

Table 1: Example set of regression equations, derived for resting and stepping activities.

Parameter Resting (f1) Stepping (f2) AAC 0.1067 $44C 0.0723 sA4C HR 0.3609 sj'Rg -3,1*90* 1 L357L.hs -1.6206 01111fl 39.6447 t 4 37.7494 + .67.6117.

SDSD 30.9653 * 42,1857 4 l0702 * flIVIOOLJ 33473 * 4 4.1.1504 + 93.$9E1 * r 2 EDR Z426s30R -14.8190 23822 *EDR Table 2: Example set of regression equations, derived for cycling and walking/running activities.

Parameter Cycling (f3) Walking/Running (f4) O,D437 * AAC AAC 0:662 ft:mos1r)sWc _11s9÷49.7:8e4 ____________ __________________________ TPr =1139 HR.29g3:p-12312 i1364*?Rs±43.22g TP =57 nn*I*I ______ ______ OUI"II'l 4fl37 ---2293S3 s 2c.7997 -4-1553534 s 2 SDSD 50.2550± :1S4.02:ii 95.7325: --489.3375 RMSSD 43,6546 ± 142:1495 535423 ± 5475575 2.5225 s33fl EDR 29551.ED:R 6.1064. EOR -65.2917 _________ ____________________ TPri7.i023 Where IPAAC indicates the transition point between the first (top) and the second (bottom) AAC regression equations; PHd indicates the transition point between the first (top) and the second (bottom) HR regression equations; and TPEDR indicates the transition point between the first (top) and the second (bottom) EDR regression equations.

Table 3: Example set of branch-specific coefficients, derived for resting, stepping, cycling and walking/running activities. Note that the numerical subscript of each branch-specific coefficient has been replaced with the name of its corresponding regression equation.

EDR

Activity Type WAAC WHRaS WSDNN WSDSD WRMSSD WEDR Output Valid 0.3150 0.6434 0.0139 0.0024 0.0191 0.0061 Resting Invalid 0.2759 0.7081 0.0081 0.0059 0.0019 -Valid 0.3838 0.5809 0.0238 0.0068 0.0032 0.0014 Stepping Invalid 0.4001 0.5863 0.0037 0.0047 0.0052 -Valid 0.1860 0.6077 0.0029 0.0984 0.1034 0.0015 Cycling Invalid 0.1247 0.6643 0.0011 0.1969 0.0130 -Valid 0.2452 0.7205 0.0112 0.0085 0.0044 0.0102 Walking/Running Invalid 0.4933 0.3514 0.1511 0.0021 0.0021 -Although the present invention has been described with reference to calculating PAl the skilled person will understand that suitable algorithms may be substituted to calculate PAEE. For example, as PAEE is equal to the area under the plot of PAl v.

time, PAEE may be calculated using the trapezoidal rule where PAl is given as a function of time.

As will be clear to the skilled person each time interval may be any suitable duration.

For example, they may be 60 seconds in duration.

The RHR of the user may be calculated using any suitable mechanism, for example by having the user cause a measurement from the EGG sensor to be stored as the RHFI in the memory.

The apparatus may be further configured to calculate a cumulated energy expenditure.

For example, the apparatus may be configured to calculate a sum of the PAl and PAEE respectively over a time period selected by the user. The calculation may occur in real time or may be calculated at a later point in time.

The heart rate monitor and accelerometer may be integral with the apparatus.

Alternatively, the heart rate monitor and accelerometer may comprise one or more separate devices which transmit EGG and accelerometer data to the energy estimation apparatus over any suitable connection such as a wired connection or a wireless connection.

The skilled person will also understand that the functions of the parameter estimation module and the energy expenditure calculation module may be provided on a single processor or in any other suitable configuration; for example, the activity classification may be performed by and activity classification module separate from the parameter estimation module.

Although the energy expenditure calculation module has been described as calculating multiple values of PAl and/or FAEE and selecting from those multiple values using the activity type determined by the activity classification it is clear that the energy expenditure calculation module may select one of the plurality of algorithms used to calculate energy expenditure using the activity classification and then calculate a single estimated PAl and/or PAEE value. Where such a method is implemented in conjunction with selecting an appropriate algorithm based on whether the EDR value is valid the selection based upon activity type may occur before or after algorithm selection based upon whether the EDR is valid or invalid.

Optionally, the HRV algorithm 7 utilizes the ORS complex peak locations derived by the HR algorithm 3 to derive time domain indices of HRV.

For this purpose, the QRS peak locations are processed in a number of steps. Firstly, the differences between individual peak locations are used in order to construct an fl-P interval time series. The time-series is then detrended and filtered using an impulse-rejection filter in order to remove ectopic beats and artefacts, which may impede the estimation of HRV. The resultant trace, referred to as the N-N interval tachogram, may then be used directly to derive (both short-term and long-term) time-domain and geometrical indices of HRV in addition to performing signal interpolation and re-sampling, in order to derive frequency-domain indices of HRV.

Three short-term time-domain indices of HRV, including SDNN, SDSD and RMSSD have been found to correlate well with incremental levels of physical activity. Moreover, the efficacy of these indicators does not fall when the ECG epoch duration is reduced from the conventional length of five minutes to one minute. These parameters are thus likely to be sensitive rapid changes in energy expenditure, and are thus considered suitable candidate measures for the estimation of PAI/PAEE. The invention is not limited to these parameters only, and other indices of HRV that are related to PAI/PAEE may be used, either instead of, or in conjunction with those listed. The use of time domain HRV parameters is described in more detail in UK Patent Application entitled Apparatus and Method for Estimating Energy Expenditure" filed by Toumaz Healthcare Limited on 25 March 2013.

Claims

CLAIMS: 1. An energy expenditure estimating apparatus comprising: an accelerometer input configured to receive accelerometer data from an accelerometer; an electrocardiogram (FOG) input configured to receive FOG data from an FOG sensor; a parameter estimation module configured to calculate one or more estimated parameters using at least one of the accelerometer data and the EOG data, the one or more estimated parameters including an ECG-derived respiration rate; and an energy expenditure module configured to receive one or more of the one or more estimated parameters from the parameter estimation module and calculate an estimated energy expenditure using at least the received FOG-derived respiration rate.
2. The energy expenditure estimating apparatus of claim 1 further comprising an activity classification module configured to determine the active state of the user from the one or more estimated parameters; a memory including a plurality of algorithms each algorithm associated with an active state wherein the energy expenditure module is configured to select an algorithm to calculate the estimated energy expenditure from a plurality of algorithms dependent upon the determined active state of the user OR to select an estimated energy expenditure calculated by the algorithm associated with the determined active state from a plurality of estimated energy expenditures, each estimated energy expenditure being calculated by one of the plurality of algorithms.
3. The energy expenditure estimating apparatus of claim 1 further comprising a memory including a first algorithm to be used when the ECG-derived respiration rate is determined to be valid and a second algorithm to calculate energy expenditure when the EOG-derived respiration rate is determined to be invalid wherein the energy expenditure module is configured to select one of the first and second algorithm dependent upon whether the ECG-derived respiration rate received from the parameter estimation module is valid.
4. The energy expenditure estimating apparatus of claim 3, further comprising an activity classification module configured to determine the active state of the user from the one or more estimated parameters; wherein the memory includes a plurality of first and second algorithms, each pair of first and second algorithms being associated with an active state wherein the energy expenditure module is configured to select an algorithm to calculate the estimated energy expenditure from a plurality of algorithms dependent upon the determined active state of the user and whether the FOG-derived respiration rate is valid or invalid OR to select an estimated energy expenditure calculated by the algorithm associated with the determined active state and validity of the EGG-derived respiration rates from a plurality of estimated energy expenditures, each estimated energy expenditure being calculated by one of the plurality of algorithms.
5. The energy expenditure estimating apparatus of any one of the preceding claims, wherein the EGG derived respiration rate is calculated using ORS peak locations in order to acquire values of instantaneous heart rate, deriving the respiratory waveform using the values of instantaneous heart rate and calculating the FOG derived respiration rate by detecting peaks in the derived respiratory waveform.
6. The energy expenditure estimating apparatus of any one of claims 1 to 4, wherein calculating the EGG derived respiration rate by detecting peaks in the derived respiratory waveform comprises calculating the FCC derived respiration rate from the change in amplitude of QRS peaks.
7. The energy expenditure estimating apparatus of claim any one of the preceding claims wherein the estimated parameters further comprise heart rate above sleep, heart rate variability (HRV) and accelerometer activity counts (AAG).
8. The energy expenditure estimating apparatus of claim any one of the preceding claims wherein the estimated energy is physical activity intensity and is calculated using the equation: PAl = (WI1 f1 (AAC)) + (WI2r (BR)) + (WI3R (EDR)) + (W14g1 (HRt'I)) + + (Tic g,(HRV7)) where W is a coefficient and f() and go are a regression equations for the parameter wherein the coefficient and regression equation are dependent upon at least the activity type.
9. The energy expenditure estimating apparatus as claimed in claim 7, or claim 8 when dependent upon claim 7, wherein the HRV parameters are time-domain HRV parameters.
10. The energy expenditure estimating apparatus as claimed in claim 9 wherein the time-domain HRV parameters comprise one or more of the standard deviation of NN intervals (SDNN), standard deviation of differences between successive NNs (SDSD) and the square root of the mean squared difference of successive NNs (RMSSD).
11. A system configured to output an estimated energy expenditure comprising: an accelerometer; an EGG sensor; an energy expenditure estimating apparatus as recited in any one of claims ito 10.
12. A system as claimed in claim ii wherein the accelerometer and ECG sensor are configured to transmit data to the energy expenditure estimating apparatus over a wireless connection.
13. A method to estimate energy expenditure comprising: receiving accelerometer data from an accelerometer; receiving electrocardiogram (ECG) data from an EGG sensor; calculating one or more estimated parameters using at least one of the accelerometer data and the EGG data, the one or more estimated parameters including an EGG-derived respiration rate; and calculating an estimated energy expenditure using at least the received EGG-derived respiration rate.