Elastic apnea detection during sleep for arrhythmic heartbeats

Abstract

A method for generating sleep apnea information for a portion of body sensor data, the method relying on determining the presence or absence of arrhythmia in the portion of body sensor data, wherein a first sleep apnea detection technique is used in response to detecting the absence of arrhythmia in the portion of body sensor data, and a second, different sleep apnea detection technique is used in response to detecting the presence of arrhythmia in the portion of body sensor data.

Description

Technical Field

[0001] The present invention relates to the field of apnea detection during sleep.

Background Art

[0002] Sleep apnea is a common term used to refer to two types of sleep-disordered breathing, namely obstructive sleep apnea (OSA) and central sleep apnea (CSA). A combination of these two disorders is usually referred to as "mixed apnea". Both of these disorders are associated with a decrease in airflow (hypopnea) or a complete cessation (apnea). This leads to a decrease in blood oxygen saturation, a final arousal of the cerebral cortex, and a burst of related sympathetic effects, as well as an increase in heart rate and blood pressure.

[0003] Sleep apnea is often underdiagnosed due to the limited usefulness of sleep laboratories and the high cost associated with sleep tests. It is widely recognized that underdiagnosed sleep apnea is an important risk factor for the development of cardiovascular diseases, cognitive impairment, and diabetes. Therefore, there is a need to provide an effective and low-cost method for identifying the occurrence of sleep apnea.

[0004] The latest sleep apnea detection methods have steps of processing heart rate information such as heart rate variability to detect the occurrence of sleep apnea. In particular, a so-called apnea-hypopnea index (AHI) indicating the occurrence and severity of apnea events during sleep can be generated.

[0005] WO2016 / 142793A1 discloses a portable electronic device for processing signals acquired from a living body. This enables the automatic detection of biological parameters representing sleep behavior disorders (e.g., those of the RBD type) and, preferably, sleep apnea for large-scale screening applications.

[0006] US Patent Application No. 2019 / 282180A1 discloses a method and apparatus for obtaining respiratory information about a subject.

Summary of the Invention

Problems to be Solved by the Invention

[0007] There is a continuous desire to improve the accuracy and robustness of apnea detection during sleep.

Means for Solving the Problems

[0008] The present invention is defined by the claims.

[0009] According to an embodiment according to an aspect of the present invention, there is provided a method implemented by a computer for generating apnea information during sleep for a subject.

[0010] The method implemented by the computer includes the steps of: obtaining a portion of body sensor data corresponding to the occurrence of apnea and the occurrence of arrhythmic heartbeats from at least one body sensor that monitors the subject; processing the portion of the body sensor data using an arrhythmia detection technique to determine whether the portion of the body sensor data indicates the presence or absence of an arrhythmic heartbeat; in response to determining that the portion of the body sensor data indicates the absence of an arrhythmic heartbeat, processing at least the portion of the body sensor data using a first apnea detection technique during sleep, thereby generating apnea information during sleep; and in response to determining that the portion of the body sensor data indicates the presence of an arrhythmic heartbeat, processing the portion of the body sensor data using a second different apnea detection technique during sleep, thereby generating apnea information during sleep.

[0011] The present invention recognizes that the accuracy of apnea detection techniques during sleep is negatively affected during the subject's arrhythmic heartbeat, for example, when the subject has atrial fibrillation. In particular, it has been identified that arrhythmic heartbeats can cause normal apnea detection techniques during sleep to inaccurately classify a portion of the body sensor data (for patients suffering from apnea during sleep) as indicating the absence of apnea during sleep (i.e., generating false negatives).

[0012] The present invention proposes removing a portion or segment of body sensor data while an arrhythmic heartbeat is detected, thereby avoiding such a portion being processed by a first sleep apnea detection technique. As a result, the false negative rate of the sleep apnea detection technique is reduced.

[0013] This concept also allows for the use of a sleep apnea detection technique trained based on the world population (e.g., not necessarily suffering from an arrhythmic heart condition) to increase the accuracy of detecting the occurrence of sleep apnea. In other words, the first sleep apnea detection technique is trained using sensor data from the world population that does not suffer from arrhythmic heartbeats.

[0014] The proposed method enables the generation of sleep apnea information with a reduced false negative rate.

[0015] Sleep apnea information indicates whether a portion of the body sensor data indicates the presence of sleep apnea, the absence of sleep apnea, or whether the presence / absence of sleep apnea cannot be reliably determined (i.e., the sleep apnea status is unknown). In other words, the sleep apnea information provides three possible pieces of information, namely, "sleep apnea is detected", "sleep apnea is not detected", or "the sleep apnea state is unknown".

[0016] The present invention further proposes processing a portion of the body sensor data using a different sleep apnea detection technique, for example, a technique trained using sensor data in which an arrhythmic heartbeat is present, when an arrhythmic heartbeat is detected.

[0017] The use of two separate sleep apnea detection techniques makes it possible to significantly increase the accuracy of generating sleep apnea information.

[0018] At least one body sensor includes an electrocardiogram (ECG) sensor, a photoplethysmography (PPG) sensor, an accelerometer, and / or a camera. Such body sensors are particularly advantageous for long-term monitoring of a subject due to their minimal invasiveness and responsiveness to both sleep apnea and arrhythmias. In some embodiments, at least one body sensor includes an electroencephalogram (EEG) sensor.

[0019] The method implemented by the computer further includes obtaining, from a motion sensor that monitors the motion of the subject, a portion of the motion data that temporally corresponds to a portion of the body sensor data, and the step of processing a portion of the body sensor data using a first sleep apnea detection technique includes processing a portion of the motion data using a motion artifact detection technique to determine whether the portion of the motion data indicates the presence or absence of motion artifacts; and in response to determining that the portion of the motion data indicates the absence of motion artifacts, processing a corresponding portion of the body sensor data using a sleep apnea classification technique to generate sleep apnea information indicating whether the portion of the body sensor data indicates the occurrence or non-occurrence of sleep apnea.

[0020] It is also recognized herein that the motion of the subject can also affect the accuracy of the sleep apnea classification technique. Accordingly, this embodiment monitors the motion of the subject and determines whether the motion of the subject has affected the body sensor data (and thus the accuracy of the sleep apnea classification technique).

[0021] In some embodiments, in response to determining that a portion of the motion data indicates the presence of motion artifacts, the method includes generating sleep apnea information indicating that the presence or absence of sleep apnea cannot be reliably determined.

[0022] The method implemented by a computer further includes the step of obtaining, from a motion sensor that monitors the motion of a subject, a portion of motion data that temporally corresponds to a portion of body sensor data, and the first sleep apnea detection technique includes the step of using a sleep apnea classification technique that processes a portion of the body sensor data and a portion of the motion data to generate sleep apnea information indicating whether a portion of the body sensor data indicates the occurrence or non-occurrence of sleep apnea.

[0023] The motion of the subject indicates the occurrence or non-occurrence of sleep apnea. The combination of the motion data and the body sensor data is particularly advantageous for improving the accuracy of detecting the occurrence or non-occurrence of sleep apnea.

[0024] The embodiment further includes the step of obtaining the subject characteristics of the subject, and the step of processing a portion of the body sensor data using the first sleep apnea detection technique includes the step of processing at least a portion of the body sensor data and the subject characteristics using the first sleep apnea detection technique.

[0025] It is also specified that the use of the subject's characteristics can also improve the accuracy of the sleep apnea detection technique. In particular, it is specified that the subject's characteristics can affect the susceptibility or tendency of that subject to suffer from sleep apnea.

[0026] In at least one embodiment, the subject characteristics include the physical or artificial statistical characteristics of the subject and / or the patient medical history of the subject.

[0027] In particular, the subject characteristics identified as being related to sleep apnea include body mass index, neck circumference, age, weight, height, and gender, etc. The patient medical history indicates the past symptoms / signs / diagnoses of the subject identified as being related to sleep apnea, such as loud snoring, daytime sleepiness, and / or fatigue.

[0028] In some embodiments, the step of obtaining a portion of the body sensor data from at least one body sensor for monitoring a subject includes obtaining a portion of the raw sensor data from the at least one body sensor; and processing the portion of the raw sensor data using a low-pass filter to obtain a portion of the body sensor data.

[0029] Preprocessing the raw body sensor data using a low-pass filter can assist in removing baseline variations and artifacts from the raw body sensor data, thereby improving the detection of the presence or absence of sleep apnea.

[0030] Preferably, the arrhythmia detection technique is adapted to detect the presence or absence of atrial fibrillation. Atrial fibrillation has been identified as having a particularly large impact on the accuracy of sleep apnea detection techniques, and in particular, increasing the proportion of false negatives in sleep apnea detection techniques.

[0031] The body sensor data preferably includes heart rate variability data.

[0032] A computer program product including computer program code is also proposed, the computer program code causing a processing system to perform all the steps of any of the methods described herein when executed on a computing device having the processing system.

[0033] A sleep apnea detection module for generating sleep apnea information about a subject is also proposed. The sleep apnea detection module obtains a portion of body sensor data corresponding to the occurrence of sleep apnea and the occurrence of arrhythmic heartbeats from at least one body sensor that monitors the subject; processes the portion of the body sensor data using an arrhythmia detection technique to determine whether the portion of the body sensor data indicates the presence or absence of an arrhythmic heartbeat; and processes the portion of the body sensor data using a first sleep apnea detection technique in response to determining that the portion of the body sensor data indicates the absence of an arrhythmic heartbeat, thereby generating sleep apnea information.

[0034] A sleep apnea detection system is also proposed that includes any sleep apnea detection module described herein and one or more body sensors adapted to monitor a subject and obtain body sensor data corresponding to the occurrence of sleep apnea and the occurrence of arrhythmic heartbeats.

[0035] These and other aspects of the invention will be apparent from, and will be elucidated with reference to, the embodiments described hereinafter.

[0036] For a better understanding of the present invention, and to more clearly show how the present invention is practiced, the accompanying drawings are referred to for purposes of illustration only.

Brief Description of the Drawings

[0037]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

DETAILED DESCRIPTION OF THE INVENTION

[0038] The present invention will be described with reference to the drawings.

[0039] The detailed description and specific examples illustrate exemplary embodiments of the apparatus, system, and method, but are intended for illustration only and are not intended to limit the scope of the present invention. It should be understood that these and other features, aspects, and advantages of the apparatus, system, and method of the present invention will be better understood from the following description, the appended claims, and the accompanying drawings. It should be understood that the drawings are merely schematic and not drawn to scale. It should also be understood that the same reference numbers are used throughout the drawings to indicate the same or similar parts.

[0040] Embodiments of the present invention provide a method for generating apnea information during sleep for a portion of body sensor data. The method for generating apnea information during sleep depends on determining the presence or absence of arrhythmia in a portion of the body sensor data. In response to detecting the absence of arrhythmia in a portion of the body sensor data, a first apnea detection technique during sleep is used.

[0041] The present invention relates to the concept of generating apnea information during sleep for a portion of body sensor data. The purpose of the apnea information during sleep is to enable identification of the presence or absence of apnea during sleep in the subject during the period included in a portion of the body sensor data.

[0042] The present invention recognizes that there is a strong priority in avoiding false positives and / or false negatives in the determination of the presence / absence of sleep apnea. This is particularly important to prevent misdiagnosis or treatment errors by clinicians relying on any generated sleep apnea information.

[0043] The present invention proposes a method for reducing at least the occurrence of false negative determinations of sleep apnea by recognizing that an arrhythmia in a subject increases the false negative rate of current state-of-the-art sleep apnea detection systems.

[0044] The present invention proposes using different techniques to evaluate the presence / absence of sleep apnea if the reliability of conventional sleep apnea detection systems is low, relying on the basic recognition that the presence of arrhythmia affects the accuracy of sleep apnea detection techniques, and thus proposes using arrhythmia detection to control the method of generating sleep apnea information.

[0045] Both FIG. 1 and FIG. 2 are used to illustrate the basic recognition of the present invention.

[0046] FIG. 1 shows first body sensor data 110 (here, heart rate variability data) obtained by monitoring a first subject / individual with a body sensor. The first subject does not have an arrhythmic heartbeat but is suffering from sleep apnea.

[0047] FIG. 1 also shows a result 120 of processing the first body sensor data 110 using a conventional sleep apnea classification algorithm designed to show whether sleep apnea was detected (indicated by symbol Y) or not detected (indicated by symbol N) in the heart rate variability data 110.

[0048] Figure 2 shows second body sensor data 210, which is also heart rate variability data obtained by monitoring a second subject / person with a body sensor. The second subject has an arrhythmic heartbeat caused by, for example, atrial fibrillation (AF) and suffers from sleep apnea.

[0049] Figure 2 also shows result 220 of processing second body sensor data 110 using the same conventional sleep apnea classification algorithm designed to indicate whether sleep apnea was detected (indicated by symbol Y) or not detected (indicated by symbol N) in heart rate variability data 110.

[0050] As can be seen from Figure 2, the false negative rate of sleep apnea detection for the second subject is higher than that for the first subject. Therefore, the accuracy of the sleep apnea detection system is affected by the presence of an arrhythmic heartbeat.

[0051] The present invention recognizes that the simultaneous presence of sleep apnea and an arrhythmic heartbeat, such as atrial fibrillation, affects body sensor data such that accurate detection of sleep apnea is no longer possible. In particular, the false negative rate for sleep apnea detection is affected, leading to an incorrect estimate of the AHI index (if calculated).

[0052] This effect is particularly pronounced in heart rate variability (HRV) data. In HRV data, periodic oscillations are characteristic of sleep apnea. Therefore, sleep apnea can be identified by looking for periodic oscillations in HRV data. However, the presence of an arrhythmic heartbeat disrupts the regularity of the HRV data, thus causing the sleep apnea detection method to falsely detect the absence of sleep apnea.

[0053] The present invention proposes a method for dealing with the conflict between sleep apnea and the presence of an arrhythmic heartbeat, such as atrial fibrillation.

[0054] Figure 3 shows a sleep apnea detection system 300 according to an embodiment. The sleep apnea detection system includes a sleep apnea detection module 310, a set 320 of one or more sensors 321, 322, 323, and an (optional) user interface 330. The sleep apnea detection module 310 is considered to be an embodiment of the present invention itself.

[0055] The sleep apnea detection module 310 is adapted to obtain one or more portions of body sensor data from a set 320 of one or more sensors. The sleep apnea detection module 310 processes a portion of the body sensor data to detect the presence or absence of arrhythmia. In response to detecting the absence of arrhythmia, the sleep apnea detection module processes a portion of the body sensor data using a first sleep apnea detection technique to generate sleep apnea information.

[0056] A more complete example of the operation of the sleep apnea detection module will be provided later when discussing the method according to an embodiment of the present invention.

[0057] The set 320 of one or more sensors 321, 322, 323 is adapted to generate body sensor data corresponding to both arrhythmia and sleep apnea. This data is obtained during the subject's sleep and includes, for example, heart rate or heart rate variability data, accelerometer data, ECG / EEG data, etc. Appropriate examples of sensors are shown and include one or more of an electrocardiogram (ECG) sensor 321, a photoplethysmograph (PPG) sensor 322, an accelerometer, an EEG system, and / or a camera 323. The camera monitors the subject's color change, for example, to derive heart rate information. Other appropriate examples will be well known to those skilled in the art.

[0058] The set 320 of one or more sensors is further adapted to generate additional data that is not, for example, responsive to both arrhythmias and sleep apnea. This can include, for example, respiratory data that can be obtained using a camera, accelerometer, or other sensor (which forms part of the set 320 of one or more sensors). As will be explained later, this information is used to improve the specificity of sleep apnea detection.

[0059] The user interface 330 is used to display information regarding whether sleep apnea has been detected by the sleep apnea detection module 310. This includes, for example, displaying a report on the development of sleep apnea information generated during the subject's sleep.

[0060] Figure 4 shows a method 400 according to an embodiment of the present invention.

[0061] Method 400 has a step 401 of obtaining a portion of the body sensor data 490. The body sensor data is obtained directly from one or more sensors monitoring the subject or from a database storing information obtained from one or more sensors. As explained above, a portion of the body sensor data is responsive to the presence (or absence) of both arrhythmias and sleep apnea.

[0062] It will be apparent that a portion of the body sensor data corresponds to body sensor data obtained over a period of time, i.e., is different over time.

[0063] Of course, a portion of the body sensor data includes information extracted from different information sources (e.g., heart rate information and respiratory information). If at least some of the body sensor data (e.g., just the heart rate information) is responsive to arrhythmias and sleep apnea, not all portions of the body sensor data need to be responsive to both arrhythmias and sleep apnea.

[0064] A portion of the body sensor data is derived, for example, from a window or period of the raw body sensor data (e.g., the most recently acquired portion of the raw body sensor data).

[0065] In some embodiments, step 401 includes obtaining a portion of the raw sensor data from at least one body sensor and processing the portion of the raw sensor data using a (digital) low-pass filter to obtain a portion of the body sensor data. In other words, step 401 includes preprocessing a portion of the body sensor data using a (digital) low-pass filter. This reduces the influence of baseline variations and artifacts in the portion of the body sensor data.

[0066] Method 400 further includes step 402 of processing a portion of the body sensor data using an arrhythmia detection technique. The method then includes, in decision step 403, determining whether the portion of the body sensor data indicates the presence or absence of an arrhythmic heartbeat (as a result of the processing).

[0067] The step of processing a portion of the body sensor data is performed using a machine learning method, such as a trained classifier. The classifier is adapted to predict whether a portion of the body sensor data indicates the presence (i.e., an arrhythmia) of an arrhythmic heartbeat.

[0068] In response to the portion of the body sensor data indicating the absence of an arrhythmic heartbeat, the method moves to step 404 of processing a portion of the body sensor data using a first sleep apnea detection technique, thereby generating sleep apnea information.

[0069] In particular, step 404 has a step of processing a portion of the body sensor data using a first apnea detection technique during sleep. Next, the method performs a decision step 405 of determining whether the portion of the body sensor data indicates the presence or absence of apnea during sleep. In response to detecting the presence of apnea during sleep, the method performs a step 406 of generating apnea information during sleep indicating that the portion indicates the occurrence of apnea during sleep, for example, by providing an output of "present" as apnea information during sleep. In response to detecting the absence of apnea during sleep, the method performs a step 407 of generating apnea information during sleep indicating that the portion indicates the absence of apnea during sleep, for example, by providing an output of "absent" as apnea information during sleep.

[0070] Other forms of apnea information during sleep indicating the presence or absence of apnea during sleep may be generated, for example, generating the possibility of the presence of apnea during sleep.

[0071] When step 405 is performed after step 404, step 405 uses the information obtained from the first apnea detection technique during sleep to make a determination.

[0072] In response to steps 402 - 403 determining that a portion of the body sensor data indicates the presence of arrhythmic heartbeats, several different actions are also performed, and suitable examples of these will be described later. However, it should be clear that the process performed in response to the presence of arrhythmic heartbeats is different from the process performed in response to the absence of arrhythmic heartbeats, for example, certain steps including additional steps may be omitted, or another process may also be performed together.

[0073] In one embodiment, in response to determining that a portion of the body sensor data indicates the presence of arrhythmic heartbeats, method 400 performs step 408 of processing a portion of the body sensor data using a second different sleep apnea detection technique, thereby generating sleep apnea information. The second sleep apnea detection technique is different from the first sleep apnea detection technique and, for example, was trained using different data and / or utilizes different features or elements of a portion of the body sensor data, and so on.

[0074] In this embodiment, a portion of the body sensor data including arrhythmic heartbeats is used to determine the type of arrhythmia. A certain type of arrhythmia classifier is trained by using machine learning techniques such as deep learning and / or feature-based approaches. A portion of the body sensor data and / or statistical or morphological features extracted from the body sensor data are used as inputs to this type of arrhythmia classifier. The types of arrhythmia include any of atrial fibrillation, atrial flutter, ventricular fibrillation, premature atrial contraction, supraventricular tachycardia, ventricular tachycardia, sinoatrial node syndrome, supraventricular arrhythmia, premature ventricular waveform. The second sleep apnea detection technique includes different approaches for each type of arrhythmia. Based on information on whether a sleep apnea event is detected in the remaining body sensor data, a portion of the body sensor data including arrhythmic heartbeats is selected and labeled. A portion of the body sensor data or the extracted features are combined with their labels (with sleep apnea - without sleep apnea) and provided as inputs to a machine learning algorithm. This machine learning algorithm is based on a supervised learning approach that learns whether a segment of the body sensor data hides a sleep apnea event. After determining which type of arrhythmia exists in a portion of the body sensor data, the corresponding algorithm is implemented by the second sleep apnea detection technique. The step of determining the presence or absence of a condition (such as arrhythmic heartbeats, sleep apnea, or type of arrhythmia) includes the step of determining the likelihood of the presence of the condition. The likelihood of the presence of sleep apnea is determined based on the determined likelihood of the presence of arrhythmia and / or the type of arrhythmia. Information about the determined likelihood of the presence of the condition is provided to the user, for example, by being included in the sleep apnea information.

[0075] Next, method 400 moves from step 408 to step 405. When step 405 is performed after step 408, step 405 uses information obtained from the second sleep apnea detection technique to determine whether a sleep apnea has been detected.

[0076] One or more of steps 402 and 404 are performed by deriving one or more (statistical) features from a portion of the body sensor data and processing this feature using a pre-trained classification model to determine the presence or absence of arrhythmia or sleep apnea (where appropriate).

[0077] In some embodiments, the feature is directly extracted from a portion of the body sensor data. In other embodiments, the feature is extracted from information derived from a portion of the body sensor data.

[0078] For example, consider an example where, in step 401, a portion of the body sensor data is derived from a portion of the heart rate data measured by an ECG or PPG sensor and obtained from a subject.

[0079] Derivation of the feature has the step of detecting heartbeats in a portion of the heart rate data and using a heartbeat detector to identify the inter-beat interval (IBI), i.e., to generate an IBI time series. In the case of an ECG sensor, this can be performed using a QRS detector, which identifies the R peak and locates its position. In the case of a PPG sensor, this detector can instead detect individual heartbeats by finding the peak amplitude or the position of the base of the increasing pulse. Next, the (statistical) feature is extracted from the series of IBIs, whereby the feature is obtained from a portion of the body sensor data.

[0080] The pre-trained classification model for detecting arrhythmia and / or sleep apnea is created by implementing a machine learning technique and is used to process (statistical) features derived from a portion of the body sensor data, e.g., (statistical) features extracted from an IBI time series.

[0081] In one embodiment, the arrhythmia classification model includes a logistic regression classifier. This is particularly advantageous in terms of its simplicity and interpretability. However, any suitable machine learning algorithm can be used. Examples of suitable machine learning algorithms include decision tree algorithms and artificial neural networks. Other machine learning algorithms such as support vector machines, random forests or naive Bayes models are suitable alternatives.

[0082] Further examples of suitable machine learning algorithms will be described later.

[0083] The features extracted in any of steps 402 and 404 need not be identical to each other. Thus, the features used to detect sleep apnea are different from the features used to detect arrhythmia. Similarly, the features used to detect sleep apnea vary based on whether an arrhythmia is detected or not.

[0084] As a mere example, a portion of the body data includes some information (such as respiratory data etc.) that is responsive to sleep apnea but not to arrhythmia. The features extracted for the detection of sleep apnea correspond to the features of such information.

[0085] As another example, the training of the machine learning algorithm results in the identification of different features for use in the detection of sleep apnea and the detection of arrhythmia.

[0086] Returning to FIG. 4, in some embodiments, the (first and optionally second) sleep apnea detection technique uses additional information about the subject to detect the presence or absence of sleep apnea.

[0087] As an example, the characteristics of subject 290 are used to improve the apnea detection technique during sleep. The characteristics of the subject are provided as an input to the apnea detection technique during sleep, or can be used to select which of a plurality of possible apnea detection techniques during sleep to use to process a portion of the body sensor data.

[0088] Suitable subject characteristics include body mass index, neck circumference, age, gender, height and / or weight, etc. Any suitable parameters derived using the above characteristics (for example, the body mass index calculated from height, gender and weight) are also used.

[0089] The patient's medical history is used in a similar manner to the subject characteristics. The patient's medical history indicates the subject's past symptoms / signs / diagnoses, and in particular, those identified as related to sleep apnea, such as, for example, loud snoring, daytime sleepiness and / or fatigue. This information can be used to improve the detection of sleep apnea using a portion of the body sensor data.

[0090] As will be apparent to those skilled in the art, additional information obtained from a set of one or more sensors that monitor the patient and generate, for example, respiratory information, is used to improve the detection of the presence or absence of sleep apnea.

[0091] As another example, motion data is used to improve the prediction of sleep apnea. For example, the subject's motion can affect the accuracy of the detection of the occurrence of sleep apnea by introducing artifacts into the body sensor data. Furthermore, the subject's motion represents sleep apnea, for example, due to the subject waking up temporarily due to a sleep apnea event.

[0092] Motion data is obtained from a set of one or more sensors, such as a set of accelerometers or pressure sensors connected to the subject or around the subject.

[0093] The movement data of the subject is also used to determine whether the sleep apnea detection technique can reliably detect the presence or absence of sleep apnea.

[0094] Thus, in some further embodiments, the method comprises obtaining movement data corresponding in time to a portion of the body sensor data, and processing the obtained movement data to identify the presence or absence of movement artifacts, which can affect a portion of the body sensor data and thus the accuracy / reliability of sleep apnea detection.

[0095] In response to detecting the presence of movement artifacts, the method comprises generating apnea information indicating that the presence or absence of sleep apnea cannot be reliably determined. Otherwise, the method proceeds as normal (i.e., as if the step of determining the presence or absence of movement artifacts did not occur).

[0096] The process of identifying whether movement artifacts affect the function of generating accurate sleep apnea information is performed at any stage within the overall method. Preferably, this is performed before the step of determining whether a portion of the body sensor data indicates the presence or absence of arrhythmia, in order to avoid unnecessarily complex processing of a portion of the body sensor data.

[0097] Alternatively, this process can be integrated into the first and / or second sleep apnea detection techniques.

[0098] FIG. 5 shows one embodiment for processing a portion of the body sensor data using a sleep apnea detection technique. As shown, this process is at least partially integrated into steps 404 and / or 408 (if present) of method 400.

[0099] The process has a step 501 of obtaining a portion of motion data 590 from a motion sensor that monitors the motion of a subject. The portion of the motion data temporally corresponds to a portion of the body sensor data (i.e., corresponds to the same period as a portion of the body sensor data).

[0100] The process further has a step 502 of processing the portion of the motion data using an artifact detection technique to determine whether the portion of the motion data indicates the presence or absence of motion artifacts. Thus, the portion of the motion data is processed to determine whether the motion data affects the detection of sleep apnea.

[0101] Decision step 503 determines whether a motion artifact has been detected.

[0102] In response to the fact that no motion artifact has been detected, or that the detected motion artifact is negligible, e.g., the number of detected motion artifacts is lower than a predetermined threshold, in step 504, a portion of the body sensor data is processed using a sleep apnea classification technique to generate sleep apnea information. The sleep apnea classification technique includes any suitable sleep apnea processing technique such as those described above with reference to step 404 or step 408.

[0103] In response to determining that a portion of the motion data indicates the presence of a motion artifact (i.e., a non-negligible amount of motion artifacts is present), the process performs a step 505 of generating sleep apnea information indicating that the presence or absence of sleep apnea cannot be reliably determined.

[0104] Thus, the embodiment proposes to check whether arrhythmia exists and / or whether motion artifacts exist before implementing the sleep apnea detection technique. In response to the presence of either of these, the method moves to a step of generating sleep apnea information indicating that the presence or absence of sleep apnea cannot be reliably determined.

[0105] In the illustrated embodiment, the process is implemented by first checking for arrhythmias and then checking for motion artifacts. However, in other embodiments of the present invention, this process may be in the reverse order.

[0106] Thus, there is provided a method implemented by a computer for generating sleep apnea information regarding a subject, the method implemented by a computer comprising the steps of: obtaining a portion of the subject's motion data; processing the portion of the motion data using a motion artifact detection technique to determine whether the portion of the motion data indicates the presence or absence of a motion artifact; obtaining, in response to determining that the portion of the motion data indicates the absence of a motion artifact, a portion of body sensor data corresponding in time to the portion of the motion data from at least one body sensor monitoring the subject, the portion of body sensor data being responsive to the occurrence of sleep apnea and the occurrence of arrhythmic heartbeats; processing the portion of the body sensor data using an arrhythmia detection technique to determine whether the portion of the body sensor data indicates the presence or absence of an arrhythmic heartbeat; and processing the portion of the body sensor data using a first sleep apnea detection technique in response to determining that the portion of the body sensor data indicates the absence of an arrhythmic heartbeat, thereby generating sleep apnea information.

[0107] Of course, the step of obtaining a portion of the body data need not be responsive to the determination of the portion of the motion data and may be implemented, for example, in parallel with the step of obtaining the corresponding portion of the motion data, at any event.

[0108] In some further embodiments, the method comprises the step of determining the quality of a portion of the body sensor data. This is implemented using any suitable quality determination process and, for example, comprises the step of processing a portion of the body sensor data using a machine learning method for evaluating the quality of the data.

[0109] This has the step of calculating the number of heartbeats in a portion of the body sensor data in a scenario where a portion of the body sensor data includes heart rate data or accelerometer data. The number of heartbeats reflects the quality of a portion of the body sensor data.

[0110] In some embodiments, this process has the step of determining the noise level of a portion of the body part data, such as the signal-to-noise ratio (SNR). This makes it possible to derive the quality of a portion of the body sensor data.

[0111] In response to a portion of the body sensor data not meeting a predetermined threshold or criterion (e.g., the calculated number of heartbeats or SNR being less than a predetermined value), the method has the step of generating apnea information indicating that the presence or absence of sleep apnea cannot be reliably determined. Otherwise, the method proceeds as normal (i.e., as if the step of determining the quality of a portion of the body sensor data did not occur).

[0112] This step may be performed at any time during the process, but preferably is performed before the step of determining whether a portion of the body sensor data indicates the presence or absence of arrhythmia.

[0113] Thus, the embodiments propose checking for the presence of arrhythmia and optionally movement artifacts, and checking for appropriate data quality. The process of generating sleep apnea information is dependent on and / or varies according to the presence and / or absence of this / these elements.

[0114] The process of generating sleep apnea information is performed iteratively, for example, on different portions of the body sensor data. This makes it possible to process a sequence or series of portions of the body sensor data, thereby making it possible to derive sleep apnea information over a long period (e.g., during nighttime sleep).

[0115] In a preferred embodiment, a stream of body sensor data is acquired and a sliding window is applied to the stream of body sensor data. A portion of the body sensor data for each window forms the portion of the body sensor data to be subsequently processed.

[0116] In one example, the length of each window ranges from 40 to 80 seconds (e.g., 60 seconds), and 20 to 40 seconds (e.g., 30 seconds) overlap between windows. In another example, the length of each window ranges from 15 to 30 seconds (e.g., 20 seconds), and 0.5 to 2 seconds (e.g., 1 second) overlap between windows. This latter embodiment enables the identification of the presence or absence of sleep apnea at a finer granularity resolution.

[0117] In particular, the portion of the processed body sensor data is a portion of the most recently acquirable body sensor data (e.g., representing a predetermined period).

[0118] This process is shown in FIG. 6, which illustrates a method 600 for processing body sensing data 690 (a stream thereof).

[0119] Method 600 has a step 601 of acquiring a most recent portion of body sensing data 690. The method then implements method 400 for generating sleep apnea information for the acquired portion of the body sensing data.

[0120] Steps 601 and method 400 are repeatedly iterated until a stop condition is reached, as determined in step 602. The stop condition is, for example, reaching a certain number of iterations, elapsing of a predetermined period, or receiving user input.

[0121] The method then moves to an (optional) step 603 of generating a report on the body sensing data, which report indicates apnea information during sleep over time. Of course, a live report may be generated to indicate apnea information during sleep up to the current period, without the need for a stop condition to be met.

[0122] Step 603 additionally or alternatively has a step of determining an apnea hypopnea index using apnea information generated for multiple portions of the body sensor data.

[0123] In particular, the apnea hypopnea index is generated by excluding periods related to portions of the body sensor data where the presence or absence of apnea during sleep cannot be determined, i.e., periods where the apnea status during sleep is "unknown".

[0124] This has a step of dividing the number of times the presence of apnea during sleep is detected by the length of time during which the absence of apnea during sleep is detected (as opposed to, for example, the presence or absence of apnea during sleep being unknown or not reliably calculable).

[0125] Thus, in a scenario where 20 apnea events during sleep are detected over a period of 480 minutes, and 210 of those minutes are related to portions of the body sensor data where the occurrence or non-occurrence of apnea during sleep could not be accurately detected, the apnea hypopnea index is calculated to be 4.4.

[0126] Figure 7 shows a suitable example of a report 700, which is graphically represented. The report provides apnea information during sleep over a period of time.

[0127] The dotted (i.e., dot - marked) sections of the report indicate the time when the absence of sleep apnea was detected. The hatched - square sections indicate the time when the presence of sleep apnea was detected. The horizontally - hatched sections indicate the time when the presence or absence of sleep apnea could not be accurately detected due to the occurrence of arrhythmia. The diagonally - hatched sections indicate the time when the presence or absence of sleep apnea could not be reliably detected due to other reasons (e.g., low - quality data).

[0128] The display properties shown are merely illustrative and can be replaced by any other suitable indicator (e.g., color or other pattern) for distinguishing between sections.

[0129] Of course, in embodiments where a second sleep - apnea detection method is used when an arrhythmia is detected, the "horizontally - hatched" section may be replaced by a section appropriately marked to indicate the time when the presence or absence of sleep apnea was detected. Such a section includes additional indicators (e.g., a specific color or saturation compared to other sections) of the occurrence of atrial fibrillation.

[0130] In any of the aforementioned embodiments, the step of determining the presence or absence of a condition (such as an arrhythmic heartbeat or sleep apnea) includes the step of determining the likelihood of the presence of the condition. The steps carried out in response to determining the presence of the condition are carried out in response to the determined likelihood exceeding a predetermined threshold (e.g., 50% or 70%). The steps carried out in response to determining the absence of the condition are carried out in response to the determined likelihood being lower than a predetermined threshold (e.g., lower than 50%). Information about the determined likelihood of the presence of the condition is included, for example, in the sleep - apnea information and provided to the user.

[0131] The above embodiments have described how a machine learning algorithm can be used to process data in order to generate or predict output data. In particular, the machine learning algorithm can be used to process a portion of the body sensor data to determine the presence or absence of sleep apnea or arrhythmia, to process a portion of the motion data to determine the presence or absence of motion artifacts, or to process a portion of the body sensor data to determine the quality of a portion of the body sensor data.

[0132] A machine learning algorithm is any self-training algorithm that processes input data to generate or predict output data.

[0133] Suitable machine learning algorithms for use in the present invention will be apparent to those skilled in the art. Examples of suitable machine learning algorithms include decision tree algorithms and artificial neural networks. Other machine learning algorithms such as logistic regression, support vector machines, or naive Bayes models are suitable alternatives.

[0134] The structure of an artificial neural network (or simply, neural network) is inspired by the human brain. A neural network is composed of layers, and each layer contains a plurality of neurons. Each neuron involves mathematical operations. In particular, each neuron includes a combination of differently weighted conversions of a single type (e.g., the same type of conversion, such as sigmoid, but with different weights). In the process of processing input data, the mathematical operations of each neuron are performed on the input data to produce a numerical output, and the output of each layer in the neural network is sequentially supplied to the next layer. The last layer provides the output.

[0135] Methods for training machine learning algorithms are well known. Typically, such methods have a step of obtaining a training data set that includes training input data entries and corresponding training output data entries. To generate predicted output data entries, a machine learning algorithm initialized with each input data entry is applied. To modify the machine learning algorithm, the error between the predicted output data entry and the corresponding training output data entry is used. This process can be repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry (e.g., ±1%). This is commonly known as supervised learning technology.

[0136] For example, if the machine learning algorithm is formed from a neural network, the mathematical operations (weightings) of each neuron can be modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation algorithms, and the like.

[0137] Using a machine learning method has steps of deriving or extracting one or more (statistical) features from the data provided as input and processing the derived features to determine the output.

[0138] Different machine learning methods use different features of the same input data to generate their output. This is true even if the machine learning methods are intended to provide the same information (e.g., determination of the presence or absence of sleep apnea) as output.

[0139] Embodiments of the present invention propose using different sleep apnea detection techniques, e.g., machine learning algorithms trained using different data sets, depending on the determination of the presence or absence of arrhythmia. This includes using different machine learning methods to process a portion of the body sensor data to determine the presence or absence of sleep apnea.

[0140] One skilled in the art will be able to easily develop a processing system for implementing any method described herein. Therefore, each step of the flowchart may represent different actions to be performed by the processing system and may be implemented by respective modules of the processing system.

[0141] Accordingly, embodiments may utilize a processing system. The processing system can be implemented in many ways by software and / or hardware to perform the various required functions. A processor is an example of a processing system that uses one or more microprocessors programmed with software (e.g., microcode) to perform the required functions. However, the processing system may be implemented with or without using a processor and may be implemented as a combination of dedicated hardware for performing some functions and a processor (e.g., one or more programmed microprocessors and associated circuitry) for performing other functions.

[0142] Examples of processing system components used in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs).

[0143] In various embodiments, a processor or processing system is associated with one or more storage media such as volatile and non-volatile computer memories such as RAM, PROM, EPROM, and EEPROM (registered trademark). The storage media is encoded by one or more programs that perform the required functions when executed on one or more processors and / or processing systems. The various storage media may be fixed within the processor or processing system or may be transportable such that one or more programs stored thereon can be loaded into the processor or processing system.

[0144] It will be appreciated that the disclosed method is preferably a method implemented by a computer. Therefore, the concept of a computer program is also proposed, which comprises code means for implementing any of the described methods when the program is executed on a processing system such as a computer. Thus, for implementing any of the methods described herein, various portions, lines, or blocks of the code of a computer program according to an embodiment are executed by a processing system or a computer. In some alternative embodiments, the functions described in the block diagrams or flowcharts may occur in an order different from the order described in the figures. For example, two blocks shown in succession may in fact be executed substantially simultaneously, or these blocks may sometimes be executed in the reverse order depending on the functions involved.

[0145] In practicing the claimed invention, variations to the disclosed embodiments can be understood and effected by those skilled in the art in view of the drawings, this disclosure, and the appended claims. In the claims, the terms "comprising," "including," and "having" do not exclude other elements or steps, and the singular forms do not exclude the plural. A single processor or other unit may perform the functions of several items recited in the claims. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be advantageously used. Where a computer program was discussed above, it may be stored / distributed on a suitable medium, such as an optical storage medium or a solid state medium, supplied together with or as part of other hardware, but it may also be distributed in other forms, such as via the Internet or other wired or wireless remote communication systems. It should be noted that when the term "adapted to" is used in the claims or the description, the term "adapted to" is intended to be equivalent to the term "configured to." Any reference signs in the claims should not be construed as limiting the scope.

Claims

1. A method implemented by a computer for generating sleep apnea information regarding a subject, the method implemented by the computer comprising: obtaining a portion of body sensor data corresponding to the occurrence of sleep apnea and the occurrence of arrhythmic heartbeat from at least one body sensor monitoring the subject; processing the portion of the body sensor data using a machine learning method to determine whether the portion of the body sensor data indicates the presence or absence of an arrhythmic heartbeat; in response to determining that the portion of the body sensor data indicates the absence of an arrhythmic heartbeat, processing the portion of the body sensor data using a first sleep apnea detection technique, thereby generating sleep apnea information; in response to determining that the portion of the body sensor data indicates the presence of an arrhythmic heartbeat, processing the portion of the body sensor data using a second different sleep apnea detection technique, thereby generating sleep apnea information; wherein the first sleep apnea detection technique includes a machine learning algorithm; wherein the second different sleep apnea detection technique includes a machine learning algorithm trained using a dataset different from the first sleep apnea detection technique; wherein the different dataset includes sensor data indicating the presence of an arrhythmic heartbeat, a method implemented by a computer.

2. The method implemented by a computer according to claim 1, wherein the at least one body sensor includes an electrocardiogram (ECG) sensor, an accelerometer, an electroencephalogram (EEG) sensor, a photoplethysmography (PPG) sensor, and / or a camera.

3. further comprising obtaining a portion of motion data temporally corresponding to the portion of the body sensor data from a motion sensor monitoring the motion of the subject; The step of processing a portion of the body sensor data using the first apnea detection technique during sleep is processing a portion of the motion data using a motion artifact detection technique to determine whether the portion of the motion data indicates the presence or absence of motion artifacts; and processing a corresponding portion of the body sensor data using an apnea classification technique during sleep to generate apnea information during sleep indicating whether the portion of the body sensor data indicates the occurrence or non-occurrence of apnea during sleep, in response to determining that the portion of the motion data indicates the absence of motion artifacts. The method implemented by a computer according to claim 1 or 2, comprising: **Claim 4** Generating apnea information during sleep indicating that the presence or absence of apnea during sleep cannot be reliably determined, in response to determining that the portion of the motion data indicates the presence of motion artifacts. The method implemented by a computer according to claim 3. **Claim 5** obtaining a portion of the motion data corresponding in time to a portion of the body sensor data from a motion sensor that monitors the motion of the subject; The method implemented by a computer according to any one of claims 1 to 4, wherein the first apnea detection technique during sleep includes using an apnea classification technique during sleep to process the portion of the body sensor data and the portion of the motion data to generate apnea information during sleep indicating whether the portion of the body sensor data indicates the occurrence or non-occurrence of apnea during sleep. **Claim 6** obtaining the subject characteristics of the subject; The method implemented by a computer according to any one of claims 1 to 5, wherein the step of processing a portion of the body sensor data using the first apnea detection technique during sleep includes processing at least the portion of the body sensor data and the subject characteristics using the first apnea detection technique during sleep. **Claim 7** The method according to claim 6, wherein the subject characteristics include physical or anthropometric characteristics of the subject and / or the patient medical history of the subject, and is implemented by a computer.

8. The step of obtaining a portion of the body sensor data from at least one body sensor for monitoring the subject comprises: obtaining a portion of the raw sensor data from the at least one body sensor; and processing the portion of the raw sensor data using a low-pass filter to obtain the portion of the body sensor data. The method according to any one of claims 1 to 7, implemented by a computer.

9. The method according to any one of claims 1 to 8, wherein the machine learning method detects the presence or absence of atrial fibrillation, and is implemented by a computer.

10. The method according to any one of claims 1 to 9, wherein the body sensor data includes heartbeat fluctuation data, and is implemented by a computer.

11. A computer program comprising computer program code which, when executed on a computing device having a processing system, causes the processing system to perform all the steps of the method according to any one of claims 1 to 10, implemented by a computer.

12. A sleep apnea detection module for generating sleep apnea information for a subject, the sleep apnea detection module comprising: obtaining a portion of the body sensor data corresponding to the occurrence of sleep apnea and the occurrence of arrhythmic heartbeats from at least one body sensor for monitoring the subject; and processing the portion of the body sensor data using a machine learning method to determine whether the portion of the body sensor data indicates the presence or absence of an arrhythmic heartbeat. In response to determining that a portion of the body sensor data indicates the absence of arrhythmic heartbeats, processing a portion of the body sensor data using a first sleep apnea detection technique, thereby generating sleep apnea information; In response to determining that a portion of the body sensor data indicates the presence of arrhythmic heartbeats, processing a portion of the body sensor data using a second, different sleep apnea detection technique, thereby generating sleep apnea information; The first sleep apnea detection technique includes a machine learning algorithm; The second, different sleep apnea detection technique includes a machine learning algorithm trained using a data set different from that used for the first sleep apnea detection technique; The different data set includes sensor data indicating the presence of arrhythmic heartbeats, a sleep apnea detection module.

13. The sleep apnea detection module according to claim 12; One or more body sensors that monitor the subject and acquire body sensor data corresponding to the occurrence of sleep apnea and the occurrence of arrhythmic heartbeats; A sleep apnea detection system comprising:

Images