HK40000217A

HK40000217A - Patient diabetes monitoring system with clustering of unsupervised daily cgm profiles (or insulin profiles) and method thereof

Info

Publication number: HK40000217A
Application number: HK19123419.4A
Authority: HK
Inventors: David L. Duke; Bernd Steiger; Chinmay Uday MANOHAR
Original assignee: F. Hoffmann-La Roche Ag
Priority date: 2016-03-02
Filing date: 2017-02-23
Publication date: 2020-02-07

Description

Patient diabetes monitoring system with unsupervised clustering of daily CGM profiles (or insulin profiles) and methods thereof

Cross Reference to Related Applications

This application claims the benefit of U.S. patent application 15/058,271 filed 2016, 3, 2, which is incorporated herein by reference in its entirety.

Technical Field

The following disclosure relates generally to diabetes management and, in particular, to a diabetes monitoring system and method for identifying patients with inadequate diabetes control therapy(s) using a cluster of similar unsupervised daily CGM profiles (or insulin profiles).

Background

Diabetes can be characterized by hyperglycemia and relative insulin deficiency. There are two major types of diabetes, type I diabetes (insulin-dependent diabetes) and type II diabetes (non-insulin-dependent diabetes). In some cases, diabetes is also characterized by insulin resistance.

The insulin secretion function serves to control the level of blood glucose so as to maintain the glucose level at an optimum level. Health care for patients diagnosed with diabetes may include both establishing a therapy program and monitoring the progress of the person with the disease. Monitoring blood glucose levels is an important process used to help diabetics keep blood glucose levels as close to normal as possible throughout the day. Monitoring may also allow for successful treatment of diabetes by changing the therapy as necessary. Monitoring may allow a diabetic patient to more closely monitor his or her condition and, in addition, may provide valuable information to the health care provider in both determining the patient's progress and detecting any need to change the patient's therapy program.

Over the past few years, advances in the electronics field have brought significant changes to medical diagnostic and monitoring equipment, including self-care monitoring. In the control and monitoring of diabetes, relatively inexpensive and easy to use blood glucose monitoring systems have become available that provide reliable information that allows a diabetic patient, as well as his or her health care professional ("HCP"), to establish, monitor, and adjust a treatment plan.

There are two main types of blood glucose monitoring systems used by patients: single point (or non-continuous) systems and continuous systems. A non-continuous system consists of a meter and test strip and requires that a blood sample be drawn from the fingertip or alternative site, such as the forearm and leg. One example of a non-continuous system may require a diabetic patient to apply a blood sample to the reagent-impregnated region of the test strip, wipe the blood sample from the test strip after a predetermined period of time, and determine the blood glucose level after a second predetermined period of time by comparing the color of the reagent-impregnated region of the test strip to a color chart supplied by the test strip manufacturer. These systems may also rely on the incision and manipulation of fingers or alternative blood drawing sites, which can be very painful and inconvenient, especially for children.

One example of a continuous system is a continuous glucose monitor ("CGM"), which may be subcutaneously implanted and measures glucose levels in interstitial fluid at various times throughout the day, providing data showing trends in glucose measurements over a period of time. The CGM can provide a large amount of data that needs to be processed to find out the pattern of similar data. This data can be used to identify adverse patient behavior or to help optimize therapy based on similar past experiences. It can also be used to monitor glucose over time to determine blood glucose patterns. Due to the large amount of data involved, an efficient algorithm may be needed to implement pattern analysis for devices with limited processing power. Furthermore, while dynamic glucose maps (AGPs) provide both graphical and quantitative characterization of daily glucose patterns, such characterization does not provide HCPs with sufficient information to identify weak points associated with adherence or effectiveness of therapy.

While various devices and techniques for continuously monitoring a patient over time may exist, it is believed that no one prior to the inventors has made or used the inventive embodiments as described herein.

Disclosure of Invention

Against the above background in accordance with various embodiments disclosed herein, clustering of datasets based on unsupervised CGM data of similar days therefore greatly enhances the HCP ability to identify problem areas (times) day-to-night and can help optimize therapy focusing on these weak points. Various embodiments of the invention disclosed herein propose a way to automatically analyze unsupervised CGM spectra and cluster them based on similarity index. Various embodiments of the invention disclosed herein also illustrate a method for determining a minimum number of similar clusters to find in a data set.

In one example, a patient diabetes monitoring system is disclosed. The system may include: a physiological data input device that acquires a plurality of physiological measurements of a patient over a time window to generate at least one time window dataset of a collected unsupervised daily monitoring atlas; a memory storing an unsupervised daily monitoring graph clustering algorithm; and a processor in communication with the input device to receive the generated at least one time window dataset, and in communication with the memory to execute the unsupervised daily monitoring graph clustering algorithm, wherein the unsupervised daily monitoring graph clustering algorithm, when executed by the processor, causes the processor to automatically: the method includes preprocessing a data set to control an amount of bias/aggression according to a collected unsupervised daily monitoring atlas to generate a preprocessed data set, building a similarity matrix from the preprocessed data set, and outputting an optimal number of similarity clusters found by a processor from the similarity matrix.

In another embodiment, disclosed is a non-transitory computer readable medium storing a program that when executed by a processor causes the processor to execute via a patient glucose monitoring system having a physiological data input device that acquires a plurality of physiological measurements of a patient over a time window to generate at least one time window dataset of a collected unsupervised daily monitoring atlas and that is in communication with the processor to cause the processor to receive the generated at least one time window dataset and that is in communication with the memory, an unsupervised daily monitoring atlas clustering algorithm to cause the processor to automatically: the method includes preprocessing a data set to control an amount of bias/aggression according to a collected unsupervised daily monitoring atlas to generate a preprocessed data set, building a similarity matrix from the preprocessed data set, and outputting an optimal number of similarity clusters. In another embodiment of the non-transitory computer readable medium, the CGM map or the insulin map is at least one time window dataset from the patient and comprises raw data, transformed data, raw data associated with a related data tag, transformed data associated with a related data tag, or a combination thereof.

In yet another embodiment, a method for identifying day(s) of inadequate diabetes control therapy for a patient using a monitoring system comprising a display device, a physiological data input device, and a processor is disclosed. The method includes automatically receiving a plurality of physiological measurements of a patient over a time window from a physiological data input device to generate at least one time window dataset of a collected unsupervised daily monitoring atlas; and executing the stored unsupervised daily monitoring graph clustering algorithm from the memory and causing the processor to automatically: the method includes preprocessing the data set to control an amount of bias/aggression in accordance with the collected unsupervised daily monitoring atlas, thereby generating a preprocessed data set, building a similarity matrix from the preprocessed data set, and outputting on a display an optimal number of similarity clusters found by the processor from the similarity matrix.

While the specification concludes with claims particularly pointing out and distinctly claiming the invention, it is believed that the present invention will be better understood from the following description of certain examples taken in conjunction with the accompanying drawings. In the drawings, like numerals refer to like elements throughout the several views.

Drawings

FIG. 1 depicts a diagram of an exemplary version of a patient diabetes monitoring system associated with a diabetic patient.

Fig. 2 depicts a block diagram of an exemplary version of the patient diabetes monitoring system of fig. 1.

Fig. 3 depicts a block diagram of an exemplary version of a patient diabetes monitoring system.

Fig. 4 depicts a block diagram of an exemplary version of a patient diabetes monitoring system.

Fig. 5 depicts a block diagram of an exemplary version of a patient diabetes monitoring system.

Fig. 6 depicts a flow diagram of an exemplary unsupervised daily monitoring profile clustering process using a patient diabetes monitoring system.

Figure 7A depicts CGM profile traces from two days that are more or less similar but have different peak amplitudes.

Fig. 7B depicts the CGM map trace from fig. 7A in the transform space.

Fig. 8 depicts the proposed glucose transformation for retrospective analysis.

Fig. 9 depicts the distance between two points in respective corresponding time series.

Fig. 10A and 10B depict the time response of changes in an otherwise similar CGM map trace.

Fig. 11 depicts the general idea behind dynamic time warping.

Figure 12 depicts the respective first and second time series of test and target CGM map traces together with the resulting shortest alignment path therebetween.

Fig. 13A and 13B depict glucose measurement data arranged in a distance matrix and a corresponding graphical representation of the distance matrix, respectively.

14A and 14B depict a tree diagram showing the 'relationships' between members of a data set and the 'distances' between clusters and the remaining data, respectively, as the onset of merging.

FIG. 15 depicts a graphical representation of finding a minimal cluster.

FIG. 16A depicts a displayed output of a graphical plot of an original monitoring data set.

16B-16F depict the displayed output of the smallest cluster found from the original monitoring data set depicted in FIG. 16A.

The drawings are not intended to be limiting in any way and it is contemplated that the various embodiments of the invention may be practiced in a variety of other ways, including those not necessarily depicted in the drawings. The accompanying drawings incorporated in and forming a part of the specification illustrate several aspects of the present invention and, together with the description, serve to explain the principles of the invention; it is to be understood, however, that the invention is not limited to the precise arrangements shown.

Detailed Description

The following description of certain examples should not be used to limit the scope of the present invention. Other features, aspects, and advantages of the versions disclosed herein will become apparent to those skilled in the art from the following description, which is by way of illustration, one of the best modes contemplated for carrying out the invention. As will be realized, the versions described herein are capable of other different and obvious aspects, all without departing from the invention. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive.

Exemplary apparatus and methods。

Fig. 1 depicts an exemplary configuration of a patient diabetes monitoring system 100 associated with a patient 102. The patient 102 may be a diabetic patient or a patient with a physiological condition that requires routine or continuous monitoring. The monitoring system 100 includes hardware and software components that may be utilized to implement unsupervised daily monitoring profile clustering features as further described herein. As illustrated, the monitoring system 100 includes a device 105. The device 105 may be a handheld system with limited processing capabilities, such as a PDA, a mobile phone, a glucose meter, and so forth. The device 105 may also be a personal computer. As further shown in fig. 2, device 105 may include physiological data input device(s) 110, data interface 115, processor 120, database 130, memory 135 along with analysis logic 132, and display 140. These components are "operably connected" to each other and may include one or more components that are coupled to one or more other components, either directly or through one or more intervening components, such that they may communicate and transfer information as desired to perform at least the processes and functions described below. The connection may be mechanical, electrical or a connection that allows signals to be transmitted between the components, e.g., wired or wireless.

The device 105 may further include an input mechanism or user interface 145 for inputting information and/or making data/output requests. Exemplary input mechanisms or user interfaces 145 may include a touch screen, input buttons, a keyboard, a mouse, a microphone, and combinations thereof. In one embodiment, the patient diabetes monitoring system 100 implements continuous glucose monitoring in which the device 105 is operable to make multiple measurements of the concentration of glucose or a substance indicative of the concentration or presence of glucose via the physiological data input device 110 and process the data set (e.g., data set 131 containing multiple unsupervised daily CGM glucose measurements (CGM profiles)) using the processor 120 to find similar patterns represented in the data set. As used herein, continuous (or uninterrupted) glucose monitoring (or monitoring) may include periods in which monitoring of glucose concentration is performed continuously, uninterruptedly, and/or intermittently (e.g., regularly or irregularly).

Referring to fig. 2, for example, in one embodiment the physiological data input device 110 can be one or more sensors that automatically collect patient-specific physiological data (such as, for example, blood glucose, blood viscosity, or other information about the blood chemistry of the patient 102), physical activity, temperature, heart rate, blood pressure, breathing patterns, other patient-specific physiological parameters, and combinations thereof. In one embodiment, the physiological data input device 110 may be a component or region of the patient glucose monitoring system 100 by which glucose may be quantified and configured to generate a signal indicative of the glucose concentration of the patient 102. In operation, the physiological data input device 110 may pass through a glucose sensor that measures and directly or indirectly acquires a detectable signal (e.g., a chemical signal, an electrochemical signal, etc.) from glucose or a derivative thereof indicative of the concentration or presence of glucose, and may then transmit the signal to the processor 120 for further processing and/or storage in a database 130 (illustrated only in fig. 2 for ease of illustration) as a data set 131. The physiological data input device 110 can be in communication with a processor 120.

As used herein, the physiological data input device 110 can be a continuous device, such as a subcutaneous, transdermal (e.g., percutaneous), or intravascular device. However, it should be understood that the devices and methods described herein may be applied to any device (including external devices) capable of detecting the concentration of glucose and providing an output signal indicative of the concentration of glucose. In another embodiment the physiological data input device 110 can be hardware and/or software that can analyze a plurality of intermittent biological samples (e.g., blood, interstitial fluid, other desired biological fluids, etc.). The physiological data input device 110 may use any method of glucose sensing, including enzymatic, chemical, physical, electrochemical, spectrophotometric, polarimetric, calorimetric, radiometric, and the like. The physiological data input device 110 can provide an output signal indicative of, for example, glucose concentration or other physiological data using any method, including invasive, minimally invasive, and non-invasive sensing techniques. The output signal may be a raw data measurement that is used to provide a useful glucose value to a user (such as a patient or physician) who may be using the device. Smoothing, evaluation methods, and the like may be applied to the raw data measurements to provide transformed data measurements to the user, such as discussed later in later sections with reference to fig. 6.

The data measurements provided in the data set 131 may be derived from intermittent collection of data including measurements made by a device (such as, for example, the physiological data input device 110) (e.g., current measurements that ultimately correspond to glucose levels or concentrations). The data measurements may further be associated with related data tags. By way of example only, the data tags may include when to eat a meal, when to administer insulin, when to exercise, and the like. In addition, the data tags may include the amount of nutritional components in the meal, insulin, oral medications, exercise, and the like. The data measurements may further include determining transformed data measurements from one or more raw data measurements and associating those transformed data measurements with the relevant data tags.

The data measurements in the data set 131 are obtained from a particular biological system (e.g., blood, interstitial fluid, etc.) using a device that remains in operative contact with the biological system over a time window, such as, for example, the physiological data input device 110. The time window may be a defined period of time (e.g., hour(s), day(s), etc.) to obtain a series of data measurements (e.g., seconds(s), minutes(s), hours(s), etc.) that result in at least one time window data set (e.g., data set 131). The time window may also be started and stopped by the diabetic 102. By way of example only, the diabetic 102 may start the time window at the beginning of a meal and stop the time window at some later date after the meal. The at least one time window data set (or data measurements) 131 may be collected from a single individual. Alternatively, the at least one time window data set (or data measurements) 131 may be collected from multiple individuals and compiled into the database at the time or later at which the at least one time window data set (or data measurements) 131 is collected. The at least one time window data set 131 may include raw data measurements, transformed data measurements, raw or transformed data measurements associated with data tags, or a combination thereof from sensors.

The physiological data input device 110 may be capable of measuring only glucose in one embodiment. Alternatively, in other embodiments, the physiological data input device 110 may be capable of measuring any other physiological analyte of interest as a specific substance or component detected and/or measured by chemical, physical, enzymatic, or optical analysis. The data set 131 for each physiological analyte is collected and compiled into a multi-analyte database (such as, for example, database 130). In another example, the database 130 may also be formulated by compiling data measurements collected using a plurality of monitors, each of which measures a single substance, resulting in a multi-analyte database.

Examples of physiological analytes may include any specific substance, component, or combination thereof that a person desires to detect and/or measure in a chemical, physical, enzymatic, or optical assay. Such physiological analytes include, but are not limited to, urate/uric acid, glucose, urea (blood urea nitrogen), lactate and/or lactic acid, hydroxybutyric acid, cholesterol, triglycerides, creatine, creatinine, insulin, hematocrit, and hemoglobin, carbonate, calcium, potassium, sodium, chloride, bicarbonate, blood gases (e.g., carbon dioxide, oxygen, etc.), heavy metals (e.g., lead, copper, etc.), lipids, amino acids, enzyme substrates or products indicative of a disease state or condition, other markers of a disease state or condition, and the like. In the case of a multi-analyte data database, all physiological analytes may be associated with a single physiological state or condition; alternatively, in other embodiments, each physiological analyte may be associated with a different physiological state or condition.

In still other embodiments, one or more of the physiological data/information described above can be manually input by the patient 102 included in the data set 131 via the user interface 145, and requested for output (e.g., displayed on the display 140, sent to another external device via the data interface 115, etc.). In still other embodiments, the input device 110 may also include, for example, a controller, microcontroller, processor, microprocessor, or the like configured to receive and/or process signals, communicate with the processor 120, and generate a CGM map (or insulin map). The CGM profile (or insulin profile) may be a recent data set 131 (e.g., a recent at least one time window data set collected by the input device 110, a data set from a day, hour, minute, etc. provided in the memory 135 and/or database 130), and/or any other data set of interest (e.g., historical data of the patient 102 (days, weeks, months, years, etc.) before, weeks, months, years, etc.) Or any method performed by external device(s) operating on the data (and provided to the device via the data interface 115) generates a CGM profile (or insulin profile) in which the mode(s) of interest (such as, for example, the one or more glucose curves 133 depicted by fig. 16A) are provided on the display (140).

An exemplary method for generating the glucose curve 133 may include: causing the processor 120 to plot a glucose curve using the glucose data measurements provided by the physiological data input device 110, causing the processor 120 to plot a glucose curve using the glucose data measurements read from the database 130 and/or memory 135 over at least one time window or other period, causing the processor 120 to plot a glucose curve using inputs received via the user interface 145, causing the processor 120 to select a glucose curve representing a common behavior or condition detectable in the data of the patient 102 (e.g., lowering blood glucose during exercise, raising blood glucose after a meal, etc.), and combinations thereof. In other embodiments, as discussed above with respect to historical and/or external data, the glucose curve need not be selected from actual glucose data measurements. The CGM map (or insulin map) may also be generated from data entered via the user interface 145 and generated by the processor 120 from queries run on the most recent data collected by the input device 110 or stored data provided in the database 130, memory 135 and/or other external source queried by the processor 120 via the data interface 115. The CGM profile (or insulin profile) may also include any relevant data tags or multi-analyte data, and the generated and/or received CGM profile (or insulin profile) may be stored in the database 130 and/or memory 135 until required by the processor 120 for the unsupervised daily monitoring profile clustering process discussed later in later sections.

The data interface 115 may be hardware and/or software that provides the device 105 with the ability to communicate with other devices and components as discussed hereinafter, in some embodiments, and with the ability to read from and write to a non-transitory computer readable product or storage medium (such as the non-transitory computer readable medium 148) in other embodiments. For the purposes of this description, a non-transitory computer readable product or storage medium may be any means that can contain or store the program and/or code for use by or in connection with a processor, apparatus, or device. Examples of a non-transitory computer readable product or storage medium include a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a Random Access Memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk. Current examples of optical disks include compact disk read only memory (CD-ROM), compact disk read/write (CD-R/W), and DVD.

Still referring to fig. 2, the processor 120 may include any general purpose processor or any processing element configured to provide, receive, and execute sequences of instructions (such as from memory 135). For example, the processor 120 may perform the calculation using at least one time window data set 131 (or data measurements) from the physiological data input device 110 and/or a CGM map (or insulin map) from the input device 110 (when provided with the processing means), which CGM map (or insulin map) may also be viewed as the time window data set 131 generated by the input device 110. In another example, the processor 120 may also compress the at least one time window data set 131 (or data measurements) into a reduced-rank basis as will be further described herein. In another example, the processor 120 may perform unsupervised daily monitoring profile clustering with at least one time window data set 131 (or data measurements) as will be further described herein. Processor 120 may be implemented as a single computing device or a combination of computing devices, e.g., a combination of a digital signal processor and a microprocessor, a plurality of microprocessors, one or more microcontrollers, one or more microprocessors in conjunction with a digital signal processor core, or any other such configuration.

Still referring to FIG. 2, the display 140 may include a liquid crystal display ("LCD"), a touch sensitive screen, a network interface, and so forth. A touch screen or network interface may provide a convenient way to enter various commands and/or select various programmable options. In operation, the display 140 can display information for, for example, at least one time window data set 131 (or data measurements) unsupervised clustering results, marker regions to identify regions of interest, data tagging information, CGM profiles (or insulin profiles), and the like. By way of example only, the displayed information may include at least one time window data set 131 (or data measurements) that may or may not require processing by the display device prior to display. The displayed at least one time window data set 131 (or data measurements) may be raw data, real-time data, transformed data, or the like. The display 140 may include hardware and/or software including display instructions (e.g., software programming including instructions) configured to enable display of information on the display 140 and/or obtain information from the database 130. The data in the database 130 may be queried by the processor 120 and/or displayed on the display 140.

Still referring to FIG. 2, the memory 135 may be any type of memory known in the art including, but not limited to, a hard disk, magnetic tape, optical disk, semiconductor memory, floppy disk, CD-ROM, DVD-ROM, ROM memory, a remote site accessible by any known protocol, or any other memory device for storing algorithms and/or data. In operation, the memory 135 may include hardware and software for unsupervised daily monitoring of clusters of diabetes or insulin profiles, such as, for example, via the included analysis logic 132. The analysis logic 132 may be suitably configured to store, interpret and process incoming information, such as at least one time window data set 131 (raw or transformed), etc., received from the input device 110, the user interface 145 and/or resulting from a query for data available from the input device 110, the database 130, the memory 135 and/or an external source via the data interface 115, and/or the processor 120 to perform such storage, interpretation and processing of incoming information. As will be discussed in more detail below, the analysis logic 132 may include a profile clustering algorithm, one or more storage algorithms, one or more data preprocessing algorithms, and/or an initialization algorithm for performing clustering of unsupervised daily monitoring of diabetes or insulin profiles.

Referring again to fig. 2, the database 130 may include a database capable of receiving and storing measured and/or detected and/or identified characteristic information (e.g., at least one time window data set 131, raw data measurements (e.g., numerical values corresponding to physical measurements), compressed data measurements, transformed data measurements), and may include additional pertinent information as described above, such as data tags, pointers, etc. and/or one or more storage algorithms. When the processor 120 executes the one or more storage algorithms, it causes the processor 120 to store at least one time window data set 131, raw data measurements, compressed data measurements, transformed data measurements, single digital results calculated or derived from one or more raw data points, and the like in the database 130. The processor 120 may also be caused to read at least one time window data set 131, raw data measurements, compressed data measurements, transformed data measurements, and the like from the database 130. The processor 120 may also be caused to index the at least one time window data set 131, raw data measurements, compressed data measurements, transformed data measurements, etc. from the input device 110 by time and/or date. The database 130 may automatically collect and receive data measurements via the input device 110 over a time window, thereby generating and storing a time window data set 131. The data of the data set 131 may be stored in a proprietary data structure format for organizing and storing the data. Exemplary data structure types may include arrays, files, records, tables, trees, and so forth. The data structure may be designed to organize the data to suit a particular purpose so that it may be accessed by and work accordingly with the functions of the system 100.

Fig. 3 depicts another exemplary configuration of the patient diabetes monitoring system 100, and only the differences from the configuration depicted by fig. 2 are discussed hereafter for purposes of brevity. In this embodiment, the patient diabetes monitoring system 100 includes a device 105, an input device 110 as a separate component from the device 105, and a network interface 150. Device 105 includes data interface 115, processor 120, database 130, memory 135 along with analysis logic 132, display 140, and user interface 145. The input device 110 is coupled to the device 105 via a network interface 150. The network interface 150 may include wired or wireless connections, as well as any wired or wireless networking hardware, such as modems, LAN ports, wireless fidelity (Wi-Fi) cards, WiMax cards, mobile communication hardware, and/or other hardware for communicating with other networks and/or devices. Device 105 may implement data storage, unsupervised daily monitoring atlas clustering, and display of clustering results via use of analysis logic 132.

Fig. 4 depicts another exemplary configuration of the patient diabetes monitoring system 100, and only the differences from the configuration depicted by fig. 3 are discussed hereafter for purposes of brevity. In this embodiment, the patient diabetes monitoring system 100 includes a device 105, an input device 110 that is a separate component from the device 105, a first network interface 155, a second network interface 170, and a server 180. The input device 110 may provide input to the device 105 via the first network interface 155. Device 105 may be coupled to server 180 via second network interface 170. As noted above with the network interface of fig. 3, the first and second network interfaces may also include wired or wireless connections, as well as any wired or wireless networking hardware for communicating with a network and/or device. Device 105 includes a data interface 115, a processor 120, a display 140, and a user interface 145. Device 105 may handle data preprocessing, input of data requests, input of data queries, and display of data results. Server 180 includes a database 130 and a memory 135 along with analysis logic 132. In one example, the server 180 may also include a processor 185 that may be configured to store data measurements in the database 130 and perform unsupervised daily monitoring atlas clustering via use of the analysis logic 132.

Fig. 5 depicts another exemplary configuration of the patient diabetes monitoring system 100, and only the differences from the configuration depicted by fig. 5 are discussed hereafter for purposes of brevity. In this embodiment, the patient diabetes monitoring system 100 includes a device 105, an input device 110 that is a separate component from the device 105, a first network interface 155, a second network interface 170, and a server 180. Device 105 includes display 140 and user interface 145 and is configured to send raw data to server 180. Server 180 includes data interface 115, processor 120, database 130, and memory 135 along with analysis logic 132. Server 180 is configured to compress raw data measurements, store the data in database 130, and perform unsupervised daily monitoring atlas clustering, discussed below with reference to fig. 6, via use of analysis logic 132.

FIG. 6 depicts a flow diagram illustrating the general logic of the unsupervised daily monitoring atlas clustering algorithm 200, which includes the following processes: preprocessing 202, building of a similarity matrix 204, and aggregating (aggregative) clustering 206. Clustering of similar unsupervised daily monitoring CGM profiles (or insulin profiles) by the analytical logic 132 helps identify the day(s) on which diabetes control therapy is inadequate. As used herein, the term "unsupervised" means that data is collected by device 105 daily during an unsupervised condition of free life, such as during the course of normal daily life when a person is at home, school, and/or at work, as opposed to data collected according to a doctor-guided planning/testing group with controlled life, supervised conditions. As discussed in more detail hereinafter, the clustering algorithm 200 is based on both the collected unsupervised glucose data and the rate of change of the collected unsupervised glucose data. It will be appreciated that the high frequency and long duration of continuous glucose monitor capture data makes analysis of such collected data cumbersome and time consuming. Furthermore, health care professionals, including general practitioners, family physicians, and endocrinologists (researchers) have only limited time to interact with patients with diabetes (PwD). Therefore, there is a need for an efficient and automated way of analyzing CGM datasets for better analysis and optimization of therapy. The clustering algorithm 200 works in a hierarchical manner and helps physicians identify potential patterns in the data set and their similarities to other members (days) in the data set.

Pretreatment of。

The purpose of the pre-processing 202 of the data set 131 is to control the amount of any penalized bias/aggression to the high or low side that may exist in the data set due to unsupervised conditions of data collection, as well as to provide a data transformation that makes the transformed data symmetric for better statistical analysis. It is to be appreciated that the current consensus of acceptable normal blood glucose levels for healthy people is between 4.0 and 6.0 mmol/L during the pre-prandial state and up to 7.0 mmol/L during the post-prandial state. For persons with diabetes (T1 or T2), the recommended normal glucose level is 4.0-9.0 mmol/L in between. Outside these ranges, a person is at "risk" for hyperglycemia if above 9 mmol/L or at risk for hypoglycemia if below 4.0 mmol/L. Others have proposed using a hazard function for SMBG measurements to assess the risk associated with each BG value. In particular, others have proposed using equation (1), often referred to as a "Kovatchev function" (for its part a hazard function). Equation (1) is as follows:

here "h (g)" is the converted blood glucose, and "g" is the blood glucose concentration measured in millimoles per liter. See, for example, Kovatchev et al, "symmetry of the blood glucose measurement scales and updates"Diabetes Care,1997, 20, 1655-1658. The risk function described above in equation (1) has a center at 112.5 mg/dl (6.3 mmol/L), which is referred to as the optimal blood glucose concentration. Furthermore, the risks associated with hypoglycemia rise significantly faster than high blood glucose. However, this risk function is not useful for retrospective analysis of CGM data, as health care professionals may have the same concern for people with post-prandial peak glucose levels of 12 mmol/L or 15 mmol/L,although the risk calculated by equation (1) will vary significantly, for example as shown in fig. 7A and 7B for illustrative purposes.

Figure 7A depicts CGM profile traces for two days that are more or less similar but have different peak amplitudes. Fig. 7B depicts the same CGM map trace in the transform space using equation (1). From the physician's perspective, both traces will be treated similarly in a retrospective analysis, showing the need to ' blunt ' the response to the high side of the glucose response spectrum of the CGM trace. A similar need is also shown on the low side of the glucose response spectrum. In view of the above, equation (2) below describes an improved transformation that is more useful than equation (1) for retrospective analysis. Equation (2) is as follows:

where the parametersAnd parameter ofWherein the parametersT _cIs the center of the transformation space, the parameterD _rIs a minimum defined glucose level, parameterG _tIs a transformed glucose level value, and the parameterGIs the initial glucose level value in the data set 131. Fig. 8 graphically depicts the transformation for retrospective analysis performed on the same CGM trace depicted by fig. 7A according to equation (2). Equation (2) gives better control over where the zero risk occurs than equation (1), which is indicated by reference sign 801 in fig. 8. In addition, by changing/transforming the glucose value at the upper limit indicated by reference 802 and the glucose value at the lower limit indicated by reference 803, the deviation/aggressiveness of any penalty to the high side or low side can be controlled. In addition, the parameter has been passedD _rIn low bloodThe glucose value at the lower limit 803 is chosen in the sugar range and therefore from a retrospective analysis perspective, any glucose value at this level of the lower limit 803 should not have any additional risk and therefore cover a negative or low risk (cap) below the lower limit 803 at a value equal to the risk at the lower limit 803.

Similarity matrix procedure。

When using glucose along the y-axis and time mapping along the x-axis, the similarity between two CGM map traces can be calculated by calculating the straight-line distance between the data points along each of the two time series. This is called L²And (4) norm. For a time series vector X with data points i = 1,2,3 … n_iAnd another time series vector Y with member I = 1,2,3 … n_iThe distance between two time series vectors can be calculated according to equation (3) defined as follows:

。

it will be appreciated that the euclidean distance in equation (3) is one of the most commonly used, but such equations fail to take into account the dynamic nature of the time series. Thus, according to an embodiment of the invention, the distance metric is disclosed in equation (4), which equation (4) considers the glucose value (in the actual or transformed space) and the dynamic component (i.e., the glucose change slope or rate) to calculate the distance between two time series. Equation (4) is defined as follows:

where the parameter X_iIs the glucose value at time i in the first time series X; parameter Y_iIs the glucose value at time i in the second time series Y; the parameter k is a weighting factor; parameter m_xIs directed to the time series XiThe slope at time i of (a); and parameter m_yIs the slope at time i for the time series Yi. A distance metric 900 of equation (4) is illustrated in fig. 9 that depicts a distance 902 between two associated points 904, 906, 908, 910, 912, and 914 in respective time series 916 and 918. The sum of each distance 902 between the two time series 916 and 918 is calculated by the algorithm 200 and then used in an elastic alignment procedure called dynamic time warping, described briefly below.

It is to be appreciated that patient behavior may be inconsistent within a day or between two days and may show out-of-phase or have a delayed or compressed response to the prescribed therapy as a result of the unsupervised CGM profile trace (such as, for example, a delayed or compressed response to a correction bolus, an insulin bolus, a meal or physical activity, etc.). For example, fig. 10A shows an example of a temporal response 1000 that may exist in a data set 131 in an otherwise similar CGM map trace depicted with signals a and B. If the time response 1000 of the change is not accounted for (e.g., along sections 1002 and 1004), the CGM trace that appears similar will be penalized by the distance metric described above, which simply uses the difference in glucose values and the instantaneous rate of change information, as depicted by fig. 10B. Therefore, it is beneficial to further modify the above-described distance metric of equation (4) in algorithm 200 during process 204 to accommodate the time offset in the CGM map trace via dynamic time warping.

Dynamic time warping allows elastic matching of two time sequences, which may be a local compression or extension along the time axis. See Lucero, J.C. et al, Munhall, K.G., Gracco, V.G., Ramsay, J.O. (1997) "the Registration of Time and the Patterning of Speech modifications", journal of Speech, Language, and Heart Research 40: 1111-; see also Sakoe, Hiroak, Chiba, Seibi, "Dynamic programming optimization for spoken word recognition", IEEE Transactions on Acoustics, Speech and SignalProcessing 26 (1): 43-49. The principle of dynamic time warping 1100 is briefly described with reference to fig. 11. Briefly, the purpose of the dynamic time warping 1100 is to find the best alignment between time series a and time series B shown in fig. 11, shown with path P (depicted with lighter dots) 1102. As depicted, the path 1102 spans the entire length of the two time series a and B, and the function that minimizes the overall length of the path 1102 and, conversely, the distance between the curves of the two time series a and B (depicted with the deeper points 1103) is referred to as a dynamic warping function. It is to be appreciated that the shortest alignment path 1102 is thus derived to account for valid alignment path requirements to satisfy the following condition:

(a) monotonicity, i.e., path 1102 only moves forward in time;

(b) continuity, i.e., path 1102 cannot have an interruption, i.e., cannot skip data while moving forward as depicted by arrow 1104;

(c) the boundary condition is met, i.e., the path 1102 must travel the entire length and some samples are not allowed to match (e.g., block 1106);

(d) the search window is met, i.e., the local time shift of alignment path 1102 must be within a predetermined search width (e.g., line 1108); and

(e) the slope is satisfied, i.e., the temporal compression or elongation should not exceed a predetermined width (e.g., line 1110).

It is to be appreciated that during the alignment process, the dynamic warping function locally compresses or lengthens the curve in time, and for better results, the inventors have the algorithm 200 add a penalty to the total distance between two curves of the time series. To illustrate, the first and second time series X, Y of transformed data sets are processed with penalties as follows:

(a) starting at the origin, the distance between the curves of the first time series X and the second time series Y is: x (1,1) = Y (1, 1);

(b) maintaining the first row at a constant distance by X (i,1) = X (i-1,1) + Y (i, 1);

(c) the first column is kept constant by X (1, j) = X (1, j-1) + Y (1, j); and

(d) proceed to the end of the search space of the transformed dataset as defined by:

X(i, j) = min(X(i, j–1), X(i–1, j–1), X(i – 1, j)) + Y(i, j) (5)。

in equation (5) above, instead of using the simple L2 norm (euclidean distance), the distance metric described earlier is used in equation (4). Thus, when the two curves X and Y are aligned, the dynamic warping function returns to account for the difference along the time axis and the total distance between the two curves for the glucose value. Fig. 12 depicts such results 1200 of the above process, in which the original (test) and warped (target) CGM map traces 1202 and 1204 are shown, respectively, along with the shortest alignment path 1206 employed. In view of the above, the output of the similarity matrix process 204 may be summarized as follows. For a given data set 131 with 1 … n days, the distance matrix modified according to equation (4) and with equation (5) is calculated by measuring the similarity between pairs of days in the CGM map. Thus, the data set 131 for N days may be described using a matrix N × N as shown in fig. 13A and 13B. Fig. 13A and 13B depict such output glucose measurement data from the similarity matrix process 204, with distances shown in the distance matrix of fig. 13A in the similarity matrix process 204 and a corresponding graphical representation of the distance matrix shown by fig. 13B. The aggregation process 206 is now discussed hereafter.

Polymerisation process。

Because of its deterministic nature, hierarchical clustering produces consistent labels, i.e., cluster members do not migrate from one cluster to another on a repetitive run basis. See, e.g., Kaufman, L.; Rousseeuw, P.J. (1990). binding Groups in Data: An Introduction to Cluster Analysis (1 ed.). New York: John Wiley. ISBN 0-471-87876-6. This is particularly important, for example, in the case of electronic consultation where the health care provider and patient may remotely view the same data set on their respective computers, smartphones, etc. If the clustering algorithm 200 is not deterministic in nature, the HCP and patient may end up potentially looking at different cluster members within the same labeled cluster, which could lead to confusion and potentially cause medical errors (in the correct therapy).

Using the output of the similarity matrix process 204 (e.g., the distance matrix shown in fig. 13A), the aggregate clustering process 206 follows the following pseudo-code routine:

(a) calculating a distance matrix between the output data points;

(b) making each of the data points a cluster;

(c) the following are repeated:

i. merging the two closest clusters, an

Updating the distance matrix; and

(d) the iteration is performed until only a single cluster remains.

Without a predefined stop condition, the above process 206 starts with every data point in the distance matrix that is considered to be its own cluster, where the process automatically keeps moving forward until only one 'super' cluster remains.

It is to be appreciated that there are several ways to merge data points into clusters or two clusters into one cluster. Perhaps the most robust approach is to use Ward's link, which minimizes the overall increase in cluster variance. See Ward, J.H., Jr. (1963), "iterative Grouping to optimal an Objective Function," Journal of the American Statistical Association, 58, 236-. Hierarchical clustering, as the name implies, produces relationships between data points as clustering progresses from one stage to another. The relationship may be represented using a tree structure also referred to as a tree diagram 1400 as shown in fig. 14A and 14B. FIG. 14A depicts how members of the data set 131 relate to each other. Vertical line 1402 shows that two data points (individual days) 1404 are merged via a dissimilarity or distance 1406 between them. Although Ward's link ensures a minimum within cluster variance, dissimilarity (distance) 1406 within data set 131, as shown by curve 1408 in fig. 14B, increases greatly as the algorithm proceeds from level 1 to level 9. At each level, the number of clusters 1410 is reduced by 1. For example, in the case of FIG. 14A, at level 1, data points 1 and 3 merge and thus the number of possible clusters is 9, where the clustering continues during process 206, with all of the data in data set 131 grouped together to form a 1 cluster at level 9.

Finding the "correct" number of clusters can be one of the most challenging problems in data mining. This problem is solved by analyzing the 'relationships' as depicted in the tree 1400. Each stage of data consolidation gives an indication of the similarity of members within a data set, which is shown in FIG. 14B. Intuitively, the "correct" number of clusters may be defined as the level at which any addition to a cluster results in a sudden increase/enhancement in the dissimilarity or distance curve 1408. Mathematically, this may be calculated using a second derivative test to find the inflection or dip of the curve 1408 discussed hereafter with reference to FIG. 15.

Fig. 15 graphically depicts a plot of the clusters 1500 used to find the optimal (minimum) number. To find the optimal number of clusters, let usd(l)In order to be a distance curve 1502,is the first derivative of the distance curve 1502, andthe second derivative of the distance curve 1502. If it is notIf present, then along the curved(l)Set of optimal number ofA group can be considered to bePoint of (2)l. However, the above approach only works when the distance monotonically increases as the number of clusters 1503 decreases to 1. A better approximation is to calculate an inflection point 1504, labeled as a black point on the distance curve 1502 in fig. 15. The method used to calculate inflection point 1504 is described as follows:

(a) let is provided withpDistance curve of pointsd(l)At the beginning ofkThe points are 1,2, ….kAnd find the slope:

m ₁ = d(2) – d(1)/ (2-1),

m ₂ = d(3) – d(1)/ (3-1),...,

m _k = d(n) – d(1)/ (n-1);

(b) calculating the median of the slopes according to step (a):m _a = median(m ₁ , m ₂ … m _k );

(c) let is provided withpDistance curve of pointsd(l)Last of (3)nIs dotted asp-n,…,p-1,pAnd find the slope:

m _p = d(p) – d(p-1)/ (p-(p-1)),

m ₂ = d(p) – d(p-2)/ (p-(p-2)),…,

m _n = d(p) – d(p-n)/(p-(p-n))and

(d) calculating the median of the slopes according to step (c):m _b = median(m ₁ , m ₂ … m _n )，

where the first line 1506 is along a distance curve from the starting pointd(l)1502 median slope of the first pointm _aAnd a second line 1508 is defined by the starting point as being along the distance curved(l)Median slope of the endpoint of (1)m _bIs defined such that inflection point 1504 is a distance curved(l)Projection of intersection 1510 between first and second lines 1506 and 1508 andl _pto indicate. Next in process 206, if the inflection pointl _pNot an integer, the cluster is determined by the algorithm 200 in process 206 as followsL _minOptimal minimum number of:

。

for example, starting with a data set 131 from an unsupervised CGM atlas trace 133 of 10-day valuable data from a diabetic user wearing CGM (the data set 131 is graphically depicted by fig. 16A all drawn together on the display 140 of the device 105), and via processing the data set 131 with the analysis logic 132 of the clustering algorithm 200, the processor 120 automatically determines five distinct clusters 1600A, 1600B, 1600C, 1600D and 1600E each having a minimum within the cluster variance. The smallest cluster so found is then provided/output distinctly/discernibly on the display 140 by the processor 120, such as, for example, each of the distinct clusters 1600A, 1600B, 1600C, 1600D, and 1600E, respectively, as shown by fig. 16B-16F. Although in fig. 16B-16F each glucose trace 133 determined by processor 120 to be in the respective found cluster is shown in bold (darker) with a bolder solid or dashed/dotted line and the irrelevant (non-clustered) glucose traces are shown via a respectively weaker (lighter) and thinner solid line, the associated glucose trace(s) of each distinct minimal cluster may be made distinguishable from the irrelevant glucose trace(s) in many other ways, such as by different colors and/or different line representations (dashed, dot-dashed, real, darker, weaker, thinner, thicker, etc.) and by not showing/hiding the irrelevant glucose traces. It is to be appreciated that one of the advantages of determining, presenting, and displaying the glucose trace 133 via their associated found clusters (rather than via a typical data dump display showing all traces in the dataset, e.g., ten days worth of data with an average of 288 sample records per day as depicted by fig. 16A) is to make the rich informational content provided in the dataset 131 more easily and effectively discernable to the user even on mobile devices such as cell phones, bG instruments, PDAs, tablets, or the like having a space constrained display (i.e., a diagonal screen size of less than 10 inches). In other words, the device 105 provided with the clustering functionality disclosed above is used by a user to more effectively discern (than a similar prior art device without such clustering functionality) a day that diabetes control therapy is inadequate, and the similar prior art device only provides a data dump and cluster display with many sample records, such as depicted by fig. 16A.

Further, it is to be appreciated that the processor 120 that determines and generates clusters based on the data set 131 as disclosed herein, such as the displayed clusters 1600A, 1600B, 1600C, 1600D, and 1600E, creates symbolic groupings/representations of explicit qualitative data regarding the sufficiency of the user's diabetes control therapy, thereby providing a more specific and practical way to process and represent information (transformed via clustering) than previously found in the prior art. For example, as depicted by fig. 16B, a first distinct cluster 1600A determined by processor 120 based on data set 131 includes a day 6 trace because a constant band of data occurs between 11 am and 12:30 pm that has been identified and highlighted as a distinct cluster. The constant band period in day 6, depicted as having a constant, invariant signal input from CGM of 50 mg/dl of glucose sensed over one and a half hours, quickly and effectively indicates to the user (without a hypoglycemic episode) that the sensor of the CGM failed or was not properly immobilized during that period, but that the user's diabetes control therapy is generally good, as represented by the remaining plotted data for the day 6 trace. Although in fig. 16B, a sensor fault or incorrect sensor fixation is shown as a constant sensed glucose value, such a problem may also be shown as a dead time period of no signal input, a period with a depicted zero value (or some other false constant value) in the plotted data. However, if the user did have a hypoglycemic episode during the constant period between 11 am and 12:30 pm, the identified cluster indicates to the user and the observation Health Care Professional (HCP) that the diabetes control therapy was ineffective in the period after breakfast and before lunch because the user's glucose dropped below the alarm level of CGM (i.e., 50 mg/dl), resulting in a constant band period in day 6.

Also, as indicated in the next distinct cluster 1600B depicted by fig. 16C, in the absence of sensor failure/fixing problems, discernable from that cluster quickly, the user experiences a change in the speed of the sensor in the 1 st: 30 to 2 am: both the hypoglycemic period occurring during the fast sleep period between 30 and the hyperglycemic period occurring after 9 pm, because the user's glucose rises above the alarm level of CGM (i.e., 400 mg/dl), resulting in a constant band period at 400 mg/dl on day 2. While the sufficiency of the user's diabetes control therapy is generally good for the time remaining on the day 2 trace, attention/changes to the diabetes user's therapy control routine and/or lifestyle may be required for at least these two periods. This is also indicated in the next distinct cluster consisting of the first and third day traces which had hyperglycemic or near hyperglycemic periods before or just after midnight. Although such scenarios may be easily interpretable, such as if those days represent periods when the diabetic user neglects dietary restrictions, repetitions in such scenarios clearly indicate problems that should be addressed in the user's diabetes control therapy. The suggestion of how the above-mentioned problems should be solved in diabetes management therapy is also clearly reflected by clustering. For example, the user should simulate/mimic/follow the diabetes control therapy used on days 4, 9 and 10, as the next distinct cluster 1600D depicted by fig. 16E and the next (and last) distinct 1600E depicted by fig. 16F for days 5, 7 and 8 clearly show that the user's diabetes control therapy is adequate on these six days, without problems/periods of high or low blood glucose.

To facilitate and expedite viewing of the located distinct clusters 1600A, 1600B, 1600C, 1600D, and 1600E (e.g., as displayed by the initial data (dump) map 1605 from the data set 131 depicted by fig. 16A), one or more buttons (e.g., closing and viewing clusters) 1610 and 1620 of the user interface 1625 are displayed with the plotted data. Selection of the "close" button 1610 by the user using a finger, stylus, or cursor will cause the processor 120 to close the displayed diagram 1605 and return to the previous/default display image of the device 105. Selection of the "View Cluster" button 1620 by the user using a finger, stylus, or cursor will cause the processor 120 to display the first found distinct cluster, such as the cluster 1600A depicted by FIG. 16B. The user interface 1625 has a number of buttons, such as a "next cluster" button 1630 to cause the processor 120 to display the next found distinct cluster (such as cluster 1600B), "previous cluster" button 1640 (fig. 16C) to cause the processor 120 to display the previously displayed distinct cluster, and a "return to main menu" button 1650 (fig. 16E) to cause the processor 120 to return to display the original data map 1605 (fig. 16A). In other embodiments, the user interface 1625 may display buttons 1610, 1630, 1640, and 1650 along with each cluster for ease of navigation.

In view of the above disclosure, it is apparent that in one embodiment disclosed is a patient diabetes monitoring system for a patient. The system comprises: a physiological data input device that acquires a plurality of physiological measurements of a patient over a time window to generate at least one time window dataset of a collected unsupervised daily monitoring atlas; a memory storing an unsupervised daily monitoring graph clustering algorithm; and a processor connected with the input deviceCommunicate to receive the generated at least one time window dataset and communicate with the memory to execute the unsupervised daily monitoring graph clustering algorithm, wherein the unsupervised daily monitoring graph clustering algorithm, when executed by the processor, causes the processor to automatically: the method includes preprocessing a data set to control an amount of bias/aggression according to a collected unsupervised daily monitoring atlas to generate a preprocessed data set, building a similarity matrix from the preprocessed data set, and outputting an optimal number of similarity clusters found by a processor from the similarity matrix. In another embodiment of the system, the pre-processing of the data set controls the amount of bias/aggressiveness via a data transformation that symmetrizes the pre-processed data set for retrospective analysis of the data set. In another embodiment of the system, the data transformation for retrospective analysis is utilized by the data setObtained by processing a defined risk function, where the parametersAnd parameter ofWherein T is_cIs the center of the transformation space, D_rIs the minimum defined glucose level, G_tIs transformed data of a blood glucose concentration measurement provided in the data set, and "g" is an initial glucose level value of the blood glucose concentration measurement provided in the data set and measured in millimoles per liter. In another embodiment of the system, the physiological data input device is a CGM.

In another embodiment of the above-mentioned system, after the pre-processing of the data set, the pre-processed data set is then processed to establish a similarity matrix that takes into account time series dynamics in the pre-processed data set. In a further embodiment of the system, the time series in the preprocessed data set are taken into account by means of a distance matrixDynamics, the distance matrix takes into account the glucose value level in the actual space or in the transformed space and calculates the distance between each pair of similar time series of data present in the preprocessed data set via the rate of change of the glucose value level. In another embodiment of the system, byTo define a distance matrix, whereX _iIs a first time seriesXTime of (1)iThe value of glucose level at (c);Y _iis the second time seriesYTime of (1)iThe glucose value of (d);kis a weight factor;m _xis directed to a first time sequenceX _iTime ofiThe slope of (d); and ism _yIs directed to time seriesX _iTime ofiThe slope of (d). In another embodiment of the system, the first and second time series are used in an elastic alignment procedureXAndYthe sum of the distances between to account for the varying time response/displacement in the preprocessed data set. In another embodiment of the system, the elastic alignment procedure is a dynamic time warping process that allows passing of locally compressed or extended first and second time series along a time axisXAndYis matched with the elasticity of the rubber. In another embodiment of the system, the dynamic time warping process results in being added to the first and second time seriesXAndYany penalty of the sum of the distances between. In another embodiment of the system, the first and second time series are CGM curves. In another embodiment of the system, the first and second time series of preprocessed data sets are processed by the processor with penalties as followsXAndY：

(c) the first column is kept constant by X (1, j) = X (1, j-1) + Y (1, j); and

(d) proceed to the end of the search space of the preprocessed dataset as defined by X (i, j) = min (X (i, j-1), X (i-1, j)) + Y (i, j) for the next row and the next column.

In another embodiment of the system disclosed above, the output of the process of establishing the similarity matrix is examined with respect to one or more conditions to evaluate whether the determined alignment path is a valid path, the one or more conditions being: monotonicity, continuity, boundary conditions, search window and slope. In another embodiment of the system, the output of the similarity matrix process is then used in an aggregate clustering process to output similarity clusters, the aggregate clustering process having the following pseudo code:

(a) calculating a distance matrix between the output data points;

(b) making each of the data points a cluster;

(c) the following are repeated:

i. merging the two closest clusters, an

Updating the distance matrix; and

(d) the iteration is performed until only a single cluster remains.

In another embodiment of the system, inflection points in the distance matrix are calculated by the processor to find an optimal minimum number of clusters. In another embodiment of the system, ifd(l)Is a distance curve in the distance matrix,is the first derivative of the distance curve, andis the second derivative of the distance curve, andif it is notIf present, the processor will follow the curved(l)Is calculated as the optimal minimum number of clustersPoint of (2)l. In another embodiment of the system, the processor calculates the inflection point as follows:

m ₁ = d(2) – d(1)/ (2-1), m ₂ = d(3) – d(1)/ (3-1), ..., m _k = d(n) – d(1)/ (n-1);

m _p = d(p) – d(p-1)/ (p-(p-1)), m ₂ = d(p) – d(p-2)/ (p-(p-2)),…, m _n = d(p) – d(p-n)/(p-(p-n))and

where the first line is along the distance curve from the starting pointd(l)The median slope of the first point of (1)m _aAnd a second line is defined by the starting point as being along the distance curved(l)In the end point of (1)Slope of valuem _bIs defined by the inflection pointl _pTo show a distance curved(l)A projection of an intersection between a first line and a second line on the first line, and if an inflection point is presentl _pIf not an integer, then the optimal minimum number of clusters is found byL _min：

。

In yet another embodiment, disclosed is a non-transitory computer readable medium storing a program that when executed by a processor causes the processor to execute via a patient glucose monitoring system having a physiological data input device that acquires a plurality of physiological measurements of a patient over a time window to generate at least one time window dataset of a collected unsupervised daily monitoring atlas and that is in communication with the processor, such that the processor receives the generated at least one time window dataset and is in communication with the memory, an unsupervised daily monitoring atlas clustering algorithm causes the processor to automatically: the method includes preprocessing a data set to control an amount of bias/aggression according to a collected unsupervised daily monitoring atlas to generate a preprocessed data set, building a similarity matrix from the preprocessed data set, and outputting an optimal number of similarity clusters. In another embodiment of the non-transitory computer readable medium, the CGM map or the insulin map is at least one time window dataset from the patient and comprises raw data, transformed data, raw data associated with a related data tag, transformed data associated with a related data tag, or a combination thereof.

In yet another embodiment, disclosed is a method for identifying day(s) of inadequate diabetes control therapy for a patient using a monitoring system that includes a display device, a physiological data input device, and a processor. The method includes automatically receiving a plurality of physiological measurements of a patient over a time window from a physiological data input device to generate at least one time window dataset of a collected unsupervised daily monitoring atlas; and executing the stored unsupervised daily monitoring graph clustering algorithm from the memory and causing the processor to automatically: the method includes preprocessing the data set to control an amount of bias/aggression in accordance with the collected unsupervised daily monitoring atlas, thereby generating a preprocessed data set, building a similarity matrix from the preprocessed data set, and outputting on a display an optimal number of similarity clusters found by the processor from the similarity matrix.

Although several devices and their components have been discussed in detail above, it should be understood that the components, features, configurations, and methods of use of the devices discussed are not limited to the context provided above. In particular, the components, features, configurations and methods of use described in the context of one of the devices may be incorporated into any of the other devices. Furthermore, without being limited to the further description provided below, additional and alternative suitable components, features, configurations and methods of using the devices, as well as various ways in which the teachings herein may be combined and interchanged will be apparent to those of ordinary skill in the art in view of the teachings herein.

Having shown and described various versions in this disclosure, further adaptations of the methods and systems described herein may be accomplished by appropriate modifications by one of ordinary skill in the art without departing from the scope of the invention. Several of such potential modifications have been mentioned, and others will be apparent to those skilled in the art. For example, the examples, versions, geometries, materials, dimensions, ratios, steps, and the like discussed above are illustrative and not required. Therefore, the scope of the present invention should be considered in terms of the following claims and is understood not to be limited to the details of structure and operation shown and described in the specification and drawings.

Claims

1. A patient diabetes monitoring system for a patient, comprising:

a physiological data input device that acquires a plurality of physiological measurements of a patient over a time window to generate at least one time window dataset of a collected unsupervised daily monitoring atlas;

a memory storing an unsupervised daily monitoring graph clustering algorithm; and

a processor in communication with the input device to receive the generated at least one time window dataset and in communication with the memory to execute the unsupervised daily monitoring atlas clustering algorithm,

wherein the unsupervised daily monitoring graph clustering algorithm, when executed by the processor, causes the processor to automatically:

pre-processing the data set to control the amount of bias/aggression in accordance with the collected unsupervised daily monitoring atlas to generate a pre-processed data set,

establishing a similarity matrix from the preprocessed data set, an

The best number of similarity clusters found by the processor from the similarity matrix is output.

2. The system of claim 1, wherein the pre-processing of the data set controls the amount of bias/aggressiveness via a data transformation of the data set that symmetrizes the pre-processed data set for retrospective analysis.

3. The system of claim 2, wherein the data transformation for retrospective analysis is utilized by the data setObtained by processing a defined risk function, where the parametersAnd parameter ofWherein T is_cIs the center of the transformation space, D_rIs the minimum defined glucose level, G_tIs transformed data of a blood glucose concentration measurement provided in the data set, and "g" is an initial glucose level value of the blood glucose concentration measurement provided in the data set and measured in millimoles per liter.

4. The system of claim 1, wherein after the pre-processing of the data set, the pre-processed data set is then processed to establish a similarity matrix that takes into account time series dynamics in the pre-processed data set.

5. The system of claim 4, wherein the time series dynamics in the pre-processed data set are considered by a distance matrix that considers glucose value levels in real space or transformed space and calculates the distance between each pair of similar data series of the data present in the pre-processed data set via the rate of change of glucose value levels.

6. The system of claim 5, wherein the system is comprised ofTo define a matrix of distances that is,

herein, theX _iIs a first time seriesXTime of (1)iThe value of the glucose level at (c) is,Y _iis the second time seriesYTime of (1)iThe value of the glucose in the (c) fraction,kis a weight factor that is a function of,m _xis directed to a first time sequenceX _iTime ofiThe slope of (d); and ism _yIs directed to time seriesX _iTime ofiThe slope of (d).

7. The system of claim 6, used in the elastic alignment procedure in the first and second time seriesXAndYthe sum of the distances between to account for the varying time response/displacement in the preprocessed data set.

8. The system of claim 7, wherein the elastic alignment procedure is a dynamic time warping process that allows passing of locally compressed or extended first and second time series along a time axisXAndYis matched with the elasticity of the rubber.

9. The system of claim 8, wherein the dynamic time warping process results in being added to the first and second time seriesXAndYany penalty of the sum of the distances between.

10. The system according to claim 9, wherein the first and second time series are CGM curves.

11. The system of claim 9, wherein the first and second time series of preprocessed data sets are processed with a penalty by the processor as followsXAndY：

(e) starting at the origin, the distance between the curves of the first time series X and the second time series Y is: x (1,1) = Y (1, 1);

(f) maintaining the first row at a constant distance by X (i,1) = X (i-1,1) + Y (i, 1);

(g) the first column is kept constant by X (1, j) = X (1, j-1) + Y (1, j); and

(h) proceed to the end of the search space of the preprocessed dataset as defined by X (i, j) = min (X (i, j-1), X (i-1, j)) + Y (i, j) for the next row and the next column.

12. The system of claim 1, wherein the output of the process of establishing the similarity matrix is examined with respect to one or more conditions to evaluate whether the determined alignment path is a valid path, the one or more conditions being: monotonicity, continuity, boundary conditions, search window and slope.

13. The system of claim 12, wherein the output of the similarity matrix process is then used in an aggregate clustering process to output similarity clusters, the aggregate clustering process having the following pseudo code:

(a) calculating a distance matrix between the output data points;

(b) making each of the data points a cluster;

(c) the following are repeated:

i. merging the two closest clusters, an

Updating the distance matrix; and

(d) the iteration is performed until only a single cluster remains.

14. The system of claim 13, wherein inflection points in the distance matrix are computed by the processor to find an optimal minimum number of clusters.

15. The system of claim 14, wherein ifd(l)Is a distance curve in the distance matrix,is the first derivative of the distance curve, andis the second derivative of the distance curve, and ifIf present, the processor will follow the curved(l)Is calculated as the optimal minimum number of clustersPoint of (2)l。

16. The system of claim 14, wherein the processor calculates the inflection point as follows:

(e) let is provided withpDistance curve of pointsd(l)At the beginning ofkThe points are 1,2, ….kAnd find the slope:

(f) calculating the median of the slopes according to step (a):m _a = median(m ₁ , m ₂ … m _k );

(g) let is provided withpDistance curve of pointsd(l)Last of (3)nIs dotted asp-n,…,p-1,pAnd find the slope:

(h) calculating the median of the slopes according to step (c):m _b = median(m ₁ , m ₂ … m _n )，

where the first line is along the distance curve from the starting pointd(l)The median slope of the first point of (1)m _aAnd a second line is defined by the starting point as being along the distance curved(l)Median slope of the endpoint of (1)m _bIs defined by the inflection pointl _pTo show a distance curved(l)A projection of an intersection between a first line and a second line on the first line, and if an inflection point is presentl _pIf not an integer, then the optimal minimum number of clusters is found byL _min：

。

17. The system according to claim 1, wherein the physiological data input device is a CGM.

18. A non-transitory computer readable medium storing a program that when executed by a processor causes the processor to execute via a patient glucose monitoring system having a physiological data input device that acquires a plurality of physiological measurements of a patient over a time window to generate at least one time window dataset of a collected unsupervised daily monitoring profile and that is in communication with the processor, such that the processor receives the generated at least one time window dataset and is in communication with the memory, an unsupervised daily monitoring profile clustering algorithm that causes the processor to automatically:

establishing a similarity matrix from the preprocessed data set, an

And outputting the optimal number of similarity clusters.

19. The non-transitory computer readable medium of claim 18, wherein the CGM map or insulin map is at least one time window dataset from the patient and comprises raw data, transformed data, raw data associated with related data tags, transformed data associated with related data tags, or a combination thereof.

20. A method for identifying day(s) of inadequate diabetes control therapy for a patient using a monitoring system comprising a display device, a physiological data input device, and a processor, the method comprising:

automatically receiving a plurality of physiological measurements of a patient over a time window from a physiological data input device to generate at least one time window dataset of a collected unsupervised daily monitoring atlas; and

executing a stored unsupervised daily monitoring graph clustering algorithm from a memory and causing a processor to automatically:

pre-processing the data set to control the amount of bias/aggression in accordance with the collected unsupervised daily monitoring atlas, thereby generating a pre-processed data set,

establishing a similarity matrix from the preprocessed data set, an

The optimal number of similarity clusters found by the processor from the similarity matrix is output on the display.