METHOD FOR DETERMINING A SET OF AT LEAST ONE CARDIO-RESPIRATORY DESCRIPTOR OF AN INDIVIDUAL DURING SLEEP AND CORRESPONDING SYSTEM.

A system using accelerometers and a PPG sensor with machine learning analyzes sleep patterns to efficiently detect sleep disorders, overcoming harmonization issues and reducing the need for lengthy human analysis.

FR3064463B1Active Publication Date: 2024-11-22FIVEFIVE
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
FR2017052688
Authority / Receiving Office
FR · FR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2017-03-30
Publication Date
2024-11-22
Estimated Expiration
2037-03-30

AI Technical Summary

Technical Problem

Existing methods for diagnosing sleep disorders such as obstructive sleep apnea and central sleep apnea are limited by the lack of harmonization in defining physiological events, requiring lengthy human analysis of polysomnography data, and lack of non-intrusive, efficient, and automated diagnostic tools.

Method used

A system using three-axis accelerometers and a PPG sensor to extract cardio-respiratory descriptors, synchronized with gyroscopes and microprocessors, applies machine learning to analyze respiratory, cardiac, and snoring patterns, and determines positional data to identify sleep disorders automatically and efficiently.

Benefits of technology

Enables rapid, automated, and non-intrusive detection of sleep disorders by analyzing multiple nights of data, providing precise cardio-respiratory descriptors and reducing the need for bulky equipment, improving diagnostic accuracy and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000026_0000
    Figure 00000026_0000
  • Figure 00000026_0001
    Figure 00000026_0001
  • Figure 00000026_0002
    Figure 00000026_0002
Patent Text Reader

Abstract

The invention relates to a method for detecting a set of events potentially indicative of sleep disorders. It is essentially characterized by: Recording (100) data beforehand in a memory: ○ From a first accelerometer in a thoracic position, ○ From a second accelerometer, synchronized with the first, in an abdominal position; ○ From a photoplethysmography sensor in a thoracic position.- filter (110) said data by extracting ○ low frequency, mid frequency and high frequency ranges from the first accelerometer, and ○ at least one low frequency range from the second accelerometer, determine (120) in at least one range, a set of features representative of a cardiorespiratory state, as well as their time of extraction, and - compare (130) said set of features with a reference model comprising a correspondence between a set of time-distributed features and a set of physiological events, and - deduce (140) a set of at least one probable corresponding event which the individual has faced.
Need to check novelty before this filing date? Find Prior Art

Description

Preferably, we plan to extract at least the respiratory component from the raw data from the second accelerometer placed in the individual's abdominal position. Advantageously, with a three-axis accelerometer arrangement X, Y, and Z as illustrated in Figure 2A and Figure 2B, with XoZ the sagittal or median plane, XoY the coronal or frontal plane, and YoZ the axial or transverse plane, we can extract: - the respiratory component by a linear combination of the 3 axes, in this case in the form of a vector; - the cardiac component by using only the Z-axis component of the signal from each accelerometer, - the snoring component by a linear combination of the 3 axes, in this case in vector form. It is also advantageous to plan to weight the extracted snoring component along at least one of the X, Y and Z axes adaptively, depending on the individual's position, the determination of the individual's position being described later. We can also plan a signal smoothing step after extracting the components in each of the frequency bands, which allows us to remove artifacts. At the PPG sensor level, the difference in absorption between hemoglobin and oxyhemoglobin under red and infrared light is calculated. An average filter with an intermediate averaging time is then used, allowing for a compromise between measurement accuracy and artifact rejection. This time is, as a non-limiting example, 1 second. Synchronization The data (raw or filtered) from the second accelerometer is synchronized with the data (raw or filtered) from the first accelerometer and the data from optional sensors, for example by an embedded processor (microcontroller) or an acquisition system (server). For example, one can plan for an acquisition system, in this case a CPU or a microcontroller, coupled with a clock and a sampler and an analog-to-digital converter. As described later, an individual's position can be determined during sleep. This position can be synchronized with specific characteristics (respiratory, cardiac, snoring) as described below. Determination of characteristics From the data from the accelerometers (raw and preferably filtered) and where appropriate in addition from at least one of a PPG sensor, a microphone, and a set of at least one gyroscope, it is possible to determine, by frequency band, a set of at least one characteristic representative of a sleep disorder which normally took place in the individual's sleep phase, as well as the moment, i.e. the instant or the range of time, at which said characteristic was extracted. Indeed, sleep-disordered breathing manifests itself through physical or physiological conditions in the individual: snoring, tossing and turning, restlessness, positioning, chest breathing (or lack thereof), abdominal breathing (or lack thereof), etc. The convergence of these events suggests a potentially pathological state. These physical or physiological conditions in turn translate into measurable characteristics, which are extracted from the signals from the accelerometers and, where applicable, from optional sensors. The "low frequency" range of accelerometers is representative of respiratory efforts, which include, for example, the individual's inhalation and exhalation. We can extract characteristics such as: - respiratory rate, that is to say both thoracic respiratory rate and abdominal respiratory rate, - respiratory amplitude, - respiratory entropy, - etc. The "medium frequency" range is representative of the individual's heart contractions. We can extract characteristics such as: - the instantaneous heart rate, at least thanks to the first accelerometer, - the time between two successive heartbeats, - etc. The "high frequencies" range is representative of the individual's snoring. We can extract characteristics such as: the times or frequency of an individual's snoring, at least thanks to the first accelerometer, - Spectral analysis of snoring (power, amplitude, ...) - etc. For the optional sensors, it is planned to extract characteristics representing the individual's oxygen saturation at the thoracic level using the PPG sensor. From at least one of the following: the first accelerometer, the second accelerometer, and at least one angular position sensor, typically a gyrometer (or gyroscope), it is possible to extract angular position characteristics that preferentially allow for algorithmic optimization for: Calculating the user's position, - The detection, in association with at least one set of accelerometers, of possible abnormal respiratory patterns (atypical angular variation during post-obstructive apnea respiratory resumption for example). The "low frequency", "mid frequency" and "high frequency" ranges can be processed in series or in parallel. It is also possible to select a set of features from the set of features determined, in particular by a Sequential type algorithm Feature Selection (SFS) or Fast Correlation-Based Filter (FCBF), which allows for optimized calculations. Sleep disorders - events The individual's sleep disorders are characterized in particular by at least one of the following syndromes: - obstructive sleep apnea syndrome (OSAS), and - Central sleep apnea syndrome (CSAS). These syndromes typically correspond to the occurrence of at least one of the following (physiological) events during the individual's sleep: - obstructive sleep apnea, - central apnea, - mixed apnea, - hypopnea, - hyperventilation and - hypoventilation. The above events can be characterized by a specific time distribution of determined characteristics. From a physiological point of view, in most cases of OSA, air stops flowing to, and from, the lungs because of a blockage in the upper airways, in the nose or in the throat. Unlike OSA, CSA can occur even when the airway is clear. CSA means that the individual stops breathing for a predetermined number of consecutive seconds during sleep, in this case at least 10 consecutive seconds. Hypopnea, or shallow breathing, is a decrease in respiratory flow rate, the limits of which are more or less well-defined, for a predetermined number of consecutive seconds during sleep. In this case, a decrease of at least 50% in respiratory flow rate for 10 seconds is considered to define hypopnea. Hypopnea can also be further defined by a decrease of 3% to 4% in oxygen saturation (oxygen in the blood). In practice, apnea / hypopnea can therefore be characterized by repeated pauses in breathing or shallow breathing for short periods during sleep. Anatomically, there is an intermittent collapse of the upper airways, which may result in a reduction of blood oxygen concentrations during sleep. Thus, a sleeping person becomes unable to breathe normally and most often wakes up with each collapse. In reality, these definitions are relative, and the boundaries are not clearly established due to differing definitions among authors and a lack of harmonization. In other words, two different experts faced with the same results may offer two different interpretations, including, for example, the consideration or exclusion of factors surrounding the individual (risk factors, medical history, etc.) in their diagnosis. The present invention makes it possible to overcome this constraint. Reference model A reference model is planned. In this case, the reference model includes a first group of physiological data relating to individuals recognized as having sleep disorders, for example a set of polysomnographies. The reference model also advantageously includes a second group of physiological data, relating to individuals recognized as not suffering from sleep disorders, for example a set of polysomnographies. In particular, it is expected that at least the first group of physiological data will include a correspondence between a set of time-distributed features and a given event (obstructive apnea, central apnea, mixed apnea, hypopnea, hyperventilation or hypoventilation). It is then possible to compare the characteristics (determined or selected) with the reference model, in this case at least with the data from the first group, to deduce the probable corresponding event(s) that the individual faced during the period of time considered, which makes it possible to detect a set of events potentially markers of sleep disorders and to carry out a form of screening for sleep disorders instead of a diagnosis. Preferably, the comparison step is implemented by machine learning, which consists of building a decision model, for example by support vector machines (SVM), by Gaussian approach, by Bayesian approach, etc., particularly for the detection and discrimination of abnormal respiratory events. The comparison is made, for example, on the following extracted characteristics: - respiratory (rate, amplitude, thoraco-abdominal synchronization index), - cardiac (rate, HRV), - snoring (spectral composition). - Thoracic oxygen saturation (PPG sensor) Learning at the signal processing level for the first and second accelerometers makes the extracted features more robust and therefore improves the performance of the comparison algorithm. For example, processing the signal from the "mid-frequency" range should result in a sufficiently accurate heart rate. In this case, the comparison step is implemented by supervised learning using the reference model. Classification and determination of cardiorespiratory descriptors From the determined characteristics, representative of a potential sleep disorder, we can determine a set of at least one cardio-respiratory descriptor of the individual, each descriptor being representative of the occurrence of event(s) representative of sleep disorder(s) during the individual's sleep; that is to say the type of disorder (apnea, hypopnea, hyper or hypoventilation), the time when the event occurs, and the number of times these events take place, as well as their frequency if they are regular. A typical cardiorespiratory descriptor is, for example, the AHI, calculated from the number and type of respiratory events detected per hour of sleep (obstructive, central or mixed apneas / obstructive or central hypopneas). Another cardiorespiratory descriptor is the thorax-abdomen desynchronization index per unit of time. This helps to refine the determination of whether apneas are central or obstructive in nature. In this case, we can compare the respiratory effort values ​​from the filtered data of the second (abdominal) accelerometer with the respiratory effort values ​​from the filtered data of the first (thoracic) accelerometer. The thorax-abdomen desynchronization index is calculated by estimating the phase shift between the signals from the first and second accelerometers. We can then estimate the delay of one relative to the other by comparing their respective zero crossings, and then by calculating a wavelet transform to compare the spectral phase. Another cardiorespiratory descriptor is heart rate variability (HRV), which is the fluctuation in the duration of the time intervals between two consecutive heartbeats. It results primarily from extrinsic regulations and determines heart rate. While heart rate can be almost stable over a given time integration period, the time between two heartbeats can vary greatly, and its informational value is more important. In this case, it is expected that the extraction of the cardiac component will also include a step of determining the variability of the heart rate, in this case by temporal and spectral analysis of the signal from the first accelerometer in the "mid-frequency" range. Another cardiorespiratory descriptor is the oxygen desaturation index over the night of measurement. It accounts for the number of desaturations greater than 3% during the night, the largest oxygen desaturation during the night, and the average profile over the night. Another cardiorespiratory descriptor is respiratory entropy per unit of time. Another cardio-respiratory descriptor is the percentage of snoring per unit of time. Advantageously, at least one of the determined cardiorespiratory descriptors is linked to the individual's position. Indeed, the occurrence of marker events for sleep disturbances, determined by the comparison step, can be associated with the individual's position at that time or within a predetermined previous time window. This allows for the determination of cardiorespiratory descriptors for which the individual's physiological data are linked to their posture in a synchronized manner. We can thus determine, for example: a set of snoring / position pairs; HRV / position; snoring / heart rate; HRV / heart rate; etc. - a set of triplets snoring / heart rate / position; etc. - a set of quadruplets snoring / heart rate / HRV / position; etc. This makes it easier to determine the chain of events that, during a predetermined time window, led to an abnormal cardiorespiratory characteristic, for example, an increase in heart rate, an increase in snoring amplitude, etc. To determine if a cardiorespiratory characteristic is "abnormal," the value of said characteristic is typically compared with a reference value, preferably included in the reference model. In addition to the determined or selected events, the decision-making model can use as input other information relating to the individual, including but not limited to: their age, weight, height, neck circumference, medical history, eating habits, consumption of products known to be likely to disrupt sleep (alcohol and cigarettes in particular). It is also advantageous to calculate the temporal correlation between at least two physiological events of the individual over the same time period (simultaneity of events) or over several time periods (diachronism of events, for example a turning onto one side which induces snoring). Determining the position The next step is to determine the individual's position during sleep by quantifying the changes in acceleration along different axes. Typically, signals from at least one accelerometer, and optionally from at least one gyroscope, are used to determine if the individual is in one of the following positions: - lying on his back, - lying on his stomach, - lying on his right side, - lying on his left side, - seated or semi-seated, and - standing. Thanks to the invention, it becomes possible to perform a statistical analysis by cross-correlation, which makes it possible to better define the conditions under which the individual presents sleep disorders, for example an acceleration of the heart rate when he snores on his left side. The data from the first and second accelerometers and, where applicable, from additional optional sensors are analyzed (in real time or a posteriori) by a succession of sliding windows whose duration is predetermined (for example, a few seconds) during a predetermined total period (for example, a few minutes) and the characteristics determined during this predetermined total period are compared to the data of the reference model. Data from the first and second accelerometers, the PPG sensor, the microphone, or the individual's position can also be used to differentiate between specific or selected events. For example, a heart rate increase combined with vertical acceleration indicates that the individual is likely standing up, and therefore awake, and corresponding events can then be distinguished from the reference model. Thanks to the present invention, it is possible to obtain a dynamic of measurements over several nights whereas the prior art only allows a single night. Analysis time according to prior art solutions is generally on the order of one to two hours to analyze a single night's sleep for an individual. In contrast, according to the invention, the computation time is that of a single processor, i.e., a few seconds. Furthermore, the calculation is automatic, meaning it is independent of any human intervention, particularly by a medical specialist. The present invention enables the determination of correlated and abnormal events, such as monitoring the treatment of sleep disorders. It is non-intrusive, requires no bulky traditional equipment (headset, EEG, EMG, oxygenation measurement device, airflow measurement device, etc.), and may require only two accelerometers, providing a level of measurement and analysis precision suitable for early screening in the general or pathological population.

Claims

CLAIMS 1. Device for determining a set of at least one cardiorespiratory descriptor of an individual during sleep, comprising: a first set of at least one accelerometer, configured to be placed in the thoracic position of the individual; and a second set of at least one accelerometer, synchronized with the first set of at least one accelerometer and configured to be placed in the abdominal position of said individual; a memory to record data from the accelerometers; characterized in that it comprises: a computer configured to execute a computer program comprising program code instructions for, when said program is executed on a computer, executing the steps of a method for detecting a set of events potentially marking sleep-disordered breathing, comprising steps consisting of: Record (100) previously in a memory data from: o a first set of at least one accelerometer, configured to be placed in the thoracic position of an individual; and o a second set of at least one accelerometer, synchronized with the first set of at least one accelerometer and configured to be placed in the abdominal position of said individual; - filtering (110) said data, the filtering step comprising: o extracting a low frequency range, a mid frequency range and a high frequency range from the data from the first set of at least one accelerometer, and o extract at least one low frequency range from the data from the second set of at least one accelerometer, - determining (120) in at least one of the extracted frequency ranges, a set of at least one characteristic representative of a cardio-respiratory state, as well as the moment at which said characteristic was extracted; and - comparing (130) said set of at least one determined characteristic with similar characteristics from a reference model, the model comprising a correspondence between a set of characteristics distributed over time and a set of given physiological events, each event being representative of a potential sleep disorder, and - deducing therefrom (140) a set of at least one probable corresponding event which the individual has faced during a predetermined period of time.

2. Device according to claim 1, wherein the method comprises at least one of: - a step consisting of determining the position of said individual during his sleep by the data from at least one of said first set of at least one accelerometer and said second set of at least one accelerometer; and - a step consisting of determining, on the basis of at least one of said events, a set of at least one cardio-respiratory descriptor of the individual.

3. Device according to any one of the preceding claims, in which: - The step (100) of pre-recording data in a memory comprises recording data from at least one of the sensors among: a photoplethysmography (PPG) sensor, a microphone, and a set of at least one angular position measurement sensor, synchronized with the first set of at least one accelerometer and configured to be placed in the thoracic position of said individual; - The step (110) of filtering said data, further comprising at least one of: A step of coupling data from a set of at least one angular position measurement sensor for extracting user position characteristics or atypical respiratory angular velocity variations; A step consisting of averaging the data from the optical photoplethysmography sensor.

4. Device according to claim 3, in which the step of determining a set of at least one cardio-respiratory descriptor of the individual comprises at least one of: - the calculation of an apnea-hypopnea index (AHI) o determining a first set of respiratory efforts in the data from the first set of at least one accelerometer, o determining a second set of respiratory efforts in the data from the second set of at least one accelerometer, and o the determination of a thorax-abdomen desynchronization index, by synchronized comparison of the first set of respiratory efforts respiratory efforts and respiratory efforts second set of respiratory efforts, o determination of heart rate variability (HRV), o the oximetry profile or oxygen saturation, o determination of respiratory entropy per unit time, o determination of the percentage of snoring per unit time; and o determination of a nocturnal profile of oxygen saturation.

5. Device according to any one of the preceding claims, further comprising a step of selecting a set of characteristics from the set of characteristics determined, for example by a Sequential Feature Selection (SFS) or Fast Correlation-Based Filter (FCBF) type algorithm, prior to the comparison step.

6. Device according to any one of the preceding claims, in which the comparison step is implemented by machine learning, optionally supervised, and preferably supervised by the reference model.

7. Device according to any one of the preceding claims, in which the method further comprises a step of calculating at least one of: - the temporal correlation between at least two events determined over the same time range or over several time ranges; - a set of pairs (event / position), - a set of pairs (event / characteristic), - a set of triplets (event / characteristic / position), - a set of quadruplets (event / event / characteristic / position).

8. Device according to any one of claims 4 to 7, further comprising a step of determining a chain of events which, during a predetermined time window, resulted in a cardio-respiratory descriptor of abnormal value.

9. Device according to any one of the preceding claims, also comprising: - at least one of a photoplethysmography (PPG) sensor, a microphone and a set of at least one gyroscope synchronized with the first set of at least one accelerometer; a reference model stored in a memory, the model comprising a correspondence between a set of time-distributed features and a set of given physiological events, each event being representative of a potential sleep disorder; and 5 a device for filtering data from the accelerometers configured for extract low frequency, mid frequency and high frequency ranges.