KR20190113552A - Passive arrhythmias detection apparatus and method based on photoplethysmogram(ppg) inter-beat intervals and morphology - Google Patents

Abstract

A method for event detection in a user-wearable device includes the steps of: receiving, from a first sensor implemented in the user-wearable device, photoplethysmogram (PPG) signals; processing, at a processor, the PPG signals to obtain PPG signal samples; detecting, at the processor, beats in the PPG signal samples; dividing the PPG signal samples into PPG signal segments; extracting at least one inter-beat interval (IBI) feature in each PPG signal segment; classifying, at the processor, each PPG signal segment using the extracted IBI feature associated with the PPG signal segment and using a machine learning model; in response to the classifying, generating, at the processor, an event prediction result for the PPG signal segment based on the extracted IBI feature; and displaying the event prediction result at the user-wearable device. In another embodiment, the method further includes extracting morphology-based features.

Description

PASSIVE ARRHYTHMIAS DETECTION APPARATUS AND METHOD BASED ON PHOTOPLETHYSMOGRAM (PPG) INTER-BEAT INTERVALS AND MORPHOLOGY}

The present invention relates to a medical monitoring apparatus and method, and more particularly to a system and method for detecting arrhythmia.

Cardiac arrhythmias, also known as arrhythmias or irregular heartbeats, are groups in which the heartbeat is irregular, too fast or too slow. Most arrhythmia is not serious, but some things can make a person more prone to complications such as stroke or heart failure. Others can cause heart attacks. For example, atrial fibrillation (AFib) is one of the most common cardiac arrhythmias, and the presence of cardiac fibrillation can potentially pose serious health risks. Traditionally, cardiac arrhythmias are detected by electrocardiograms (ECGs) or holter monitors.

ECG measurements require sophisticated sensing devices with multiple electrodes attached to the patient and require active human participation. In general, ECG measurements are performed only for diagnostic purposes after the patient has exhibited symptoms. Photoplethysmogram (PPG) has been described as an alternative to ECG in arrhythmia detection. However, some conventional PPG measurement techniques for cardiac arrhythmias rely primarily on heart rate or heart rate detection using average heart rate, such as a window between 5 and 20 seconds. Changes in the average heart rate itself are not reliable signals for arrhythmia.

Other PPG based methods for detecting cardiac arrhythmias have low sensitivity and specificity. In addition, some methods require large continuous chunks of PPG signals, such as 30 seconds of continuous PPG signals, to perform measurements.

SUMMARY OF THE INVENTION The present invention has been made to solve the above technical problem, and an object of the present invention is to provide a PPG-based arrhythmia detection apparatus and method that enables manual and asymptomatic arrhythmia detection.

The present invention discloses an apparatus and invention for arrhythmia detection, for example as described or shown below in connection with at least one of the figures. Apparatus and invention for arrhythmia detection are described more fully in the claims.

Advantages, aspects and novel features of the present invention as well as the detailed description of the embodiments shown will be more fully understood from the following description and drawings.

According to an embodiment of the present invention, a method for event detection in a user wearable device includes receiving photoplethysmogram (PPG) signals from a first sensor implemented in the user wearable device, and processing the PPG signals to obtain PPG signal samples in a processor. Detecting a pulsation in the PPG signal samples at the processor, dividing the PPG signal samples into PPG signal segments, extracting at least one inter-beat interval (IBI) feature from each of the PPG signal segments, At the processor, using the extracted IBI feature associated with each of the PPG signal segments and classifying each of the PPG signal segments using a machine learning model, in response to the classifying, at the processor, based on the extracted IBI feature Generates event prediction results for one PPG signal segment in one of the PPG signal segments. In step and user wearable device, which comprises a step of displaying the event predictions.

According to another embodiment of the present invention, a method for event detection in a user wearable device includes receiving PPG signals from a first sensor implemented in the user wearable device, and processing the PPG signals to obtain PPG signal samples in a processor. Detecting a pulsation in the PPG signal samples at the processor, dividing the PPG signal samples into PPG signal segments, statistical properties of the PPG signal samples or waveform characteristics of the PPG signal samples in each of the PPG signal segments. Extracting at least one morphology-based feature associated with the processor; classifying, classifying each of the PPG signal segments using the extracted morphology-based feature associated with each of the PPG signal segments and using a machine learning model In response, at the processor, the extracted polo Based on the basis of the characteristic includes the step of displaying the predicted result on the event step and a user wearable device, which generates an event predictions.

An apparatus according to another embodiment of the present invention includes a sensor module and a processor including a first sensor configured to measure PPG signals. The processor is configured to process PPG signals to obtain PPG signal samples, detect a pulsation in the PPG signal samples, and divide the PPG signal samples into PPG signal segments, at least one in each of the PPG signal segments. Using an IBI sensing module configured to extract features of the IBI, a morphology sensing module configured to extract at least one morphology based feature in each of the PPG signal segments and an extracted IBI feature and an extracted morphology based feature associated with the segment And a classification module configured to classify each of the PPG signal segments using a machine learning model, the classification module generating event prediction results based on the extracted IBI features and the extracted morphology based features.

PPG-based arrhythmia detection apparatus according to an embodiment of the present invention is configured to extract the accurate arrhythmia information using short PPG signal segments. Therefore, according to an embodiment of the present invention, passive arrhythmias and asymptomatic arrhythmias can be detected.

Various embodiments of the invention are disclosed in the following description and the accompanying drawings.
1 illustrates an electronic device according to an embodiment of the present disclosure.
2 is a block diagram of a user wearable device according to an exemplary embodiment.
3 is a flowchart illustrating a method of detecting arrhythmia in a user wearable device according to an exemplary embodiment of the present invention.
4 shows exemplary signal waveforms of an ECG signal and a PPG signal.
5 shows an example signal waveform of a sequence of PPG signal samples.
6 shows an example signal waveform of a series of PPG signal samples with atrial fibrillation.
FIG. 7 is a plot showing the histogram of the root mean square of the successive differences in the IBI of PPG signals with atrial fibrillation and PPG signals with normal orbital nodal rhythm.
8 is a plot showing a histogram of multiscale sample entropy analysis of IBI of atrial fibrillation PPG signals and PPG signals with normal orbital nodule rhythm.
9 is a plot showing the histogram of the standard deviation of the area under the curve for normal isotopes and atrial fibrillation.
10 shows an example signal waveform of a series of PPG signal samples showing long tail features.
11 illustrates a user wearable device communicating with a mobile device according to an embodiment of the present disclosure.
12 illustrates a user wearable device communicating with a cloud server according to an embodiment of the present disclosure.

The present invention is a process; Device; system; Composition of the substance; A computer program product implemented on a computer readable storage medium; And / or a processor, such as a processor device or a hardware processor, configured to execute instructions stored in memory coupled to the processor or provided by the memory coupled to the processor. In this specification, such implementations or any other form disclosed herein may be referred to as techniques. In general, the order of the steps of the disclosed process may be changed within the scope of the present invention. Unless stated otherwise, a component such as a processor or memory that is described as being configured to perform a task may be implemented as a general component that is temporarily configured to perform a task at a given time or as a specific component produced to perform a task. Can be. The term processor, as used herein, may refer to one or more devices, circuits, and / or processing cores configured for processing data, such as computer program instructions.

DETAILED DESCRIPTION A detailed description of one or more embodiments of the invention is provided below in conjunction with the accompanying drawings that illustrate the principles of the invention. The invention is described in connection with these embodiments, but the invention is not limited to any embodiment. The invention includes many alternatives, modifications, and equivalents. Numerous specific details are set forth in the following description to provide a thorough understanding of the present invention. These details are provided for the purpose of illustration, and the invention may be practiced according to the claims without some or all of these specific details. For the purpose of clarity, technical material known in the technical area related to the present invention has not been described in detail so as not to unnecessarily obscure the present invention.

In embodiments of the present invention, an arrhythmia detection system and method implemented in a user wearable device may be in the form of inter-beat interval (IBI) characteristics of photoplethysmogram (PPG) signals and / or beat-by-beat. Analyzing the medical features provides high precision cardiac arrhythmia detection using PPG signals. In some embodiments, the arrhythmia detection system and method processes segments of PPG signals to detect arrhythmia and to irregularly analyze irregular IBI features and / or morphological features of the pulsation change of the user's PPG waveform. In one embodiment, the arrhythmia detection system and method observes IBI information and / or morphological information of pulsating changes in short PPG signal segments. PPG-based arrhythmia detection systems and methods are configured to extract accurate arrhythmia information from short PPG signal segments, including PPG signal information that is generally rejected in conventional PPG heart rate measurement methods. In some embodiments, PPG-based arrhythmia detection systems and methods provide improved detection accuracy, sensitivity, and specificity compared to conventional heart rate-based detection methods. Importantly, PPG-based arrhythmia detection systems and methods enable manual and asymptomatic arrhythmia detection. That is, arrhythmia detection may be performed before the user exhibits or experiences symptoms that indicate arrhythmia.

In particular, conventional arrhythmia detection techniques rely solely on the statistical distribution of interbeat intervals of heartbeat information and not on the morphological information that may be present in the heartbeat signal. In some aspects of the present disclosure, the arrhythmia monitoring system and method of the present invention includes an analysis of the morphological features of PPG signals, and thus benefits from the morphological information available in the PPG signal that is also present in the physiological environment.

In the present invention, PPG refers to an optically obtained blood flow meter. Blood meters are volumetric meters in body organs. PPG is often obtained using pulse oximeters that light up the skin and measure changes in light absorption. Conventional pulse oximeters monitor the perfusion of blood into the dermis and subcutaneous tissue of the skin as a result of the cardiac cycle where the heart pumps blood to the periphery, causing a pressure pulse on the skin. Volume changes due to pressure pulses are detected by illuminating the skin with light from the light emitting diodes and then measuring the amount of light that is transmitted or reflected to the photodiode. The advantage of PPG-based detection is that PPG signals can be easily recorded and monitored on consumer-level wearable devices without the active effort of participants. This advantage may enable manual cardiac arrhythmia monitoring and detection along with available wearable devices and smartphones.

In embodiments of the present invention, the PPG-based arrhythmia detection system and method of the present invention uses short discrete segments of PPG signals and performs one or more signal analysis on short PPG signal segments to remove cardiac or heart arrhythmia. Detect. In one embodiment, the system analyzes the statistical regularity of the distribution of the IBI of the PPG beats contained in each PPG signal segment. In another embodiment, the system analyzes the morphology based features of the PPG beats included in each PPG signal segment. In some embodiments, the system can analyze the statistical distribution of similarity or morphology characteristics between morphology based features of adjacent PPG beats extracted from each PPG signal segment. In another embodiment, the system performs analysis of both sets, i.e., IBI characteristics and morphology characteristics, which are analyzed to detect arrhythmia.

Thus, in some embodiments, the PPG based arrhythmia detection system and method provides high precision cardiac arrhythmia detection by only analyzing the IBI characteristics of the PPG signal segments or analyzing the pulsating interval morphology based features of the PPG signal segment. In some examples, information related to the morphological features of the beat interval of PPG signal segments may be added to an analysis based on IBI features to increase sensing accuracy.

In the present description, the term 'feature' used in 'IBI features' or 'morphology based features' refers to waveform shape, waveform characteristics, waveform quality and statistical characteristics or properties, measured features or Refers to all attributes or derived features or attributes. As an example, the IBI characteristics may comprise a statistical distribution of the IBI. As another example, morphology features may include waveform similarity between adjacent PPG beats. Morphological features may also include a statistical distribution of certain morphological characteristics.

In other embodiments, the PPG-based arrhythmia detection system and method of the present invention uses an electrocardiogram (ECG) signal to adjust the PPG signal to improve detection accuracy. In one embodiment, the ECG signal is used to adaptively adjust the time interval of the beats detected in the IBI feature analysis. In another embodiment, the ECG signal is used to adaptively adjust decision thresholds during classification to increase detection accuracy.

The PPG-based arrhythmia detection system and method of the present invention realizes many advantages over conventional arrhythmia detection methods using PPG. For example, the PPG-based arrhythmia detection system of the present invention can be implemented to analyze many short PPG signal segments and not to require long continuous PPG segments. As an example, information / features of various PPG signal segments that are 2-15 seconds in length may be aggregated for statistical evaluation and classification. Short PPG signal segments may be used much more frequently than the 30 second signal segment required in some conventional systems. Second, the PPG-based arrhythmia detection system of the present invention provides more accurate detection than conventional systems using only average heart rate information. The PPG-based arrhythmia detection system of the present invention employs IBI and rhythm interval morphology features that can better represent the characteristics of the arrhythmia and significantly improve detection performance.

In an exemplary embodiment of the present invention, a PPG-based arrhythmia detection system and method is implemented in a wrist-based wearable device that includes a PPG sensor. The user or subject wears a wristband wearable device continuously and the system provides notification of the detected arrhythmia. In some embodiments, the wrist based wearable device includes an accelerometer that is used to continuously determine whether the wearer is stationary. In one embodiment, the PPG light sensor is activated to measure the PPG signal when the user is at a standstill. PPG signals are observed in a series of key indicators for determining whether a wearer suffers from arrhythmia. In other embodiments, the PPG optical sensor makes continuous measurements and the accelerometer provides a motion indication signal to the PPG based arrhythmia detection system. PPG-based arrhythmia detection systems can process PPG signals and eliminate some of the PPG signals associated with high intensity motion that can affect the accuracy of the PPG signals.

1 illustrates an electronic device according to an embodiment of the present disclosure. Referring to FIG. 1, the electronic device 100, which may be a user wearable device, includes a display 160, a processor 130, a sensor module 150, a battery (not shown), a band 140, and a latch 150. Has The band 140 may be wrapped around the wrist, and the user wearable device 100 may be fixed to the wrist using the clasp 142. Sensor module 150 may include one or more sensors 152, 156 and a local processor 154. The local processor 154 may implement a control function for the sensor module and perform processing or preprocessing of the detected signal. The processor 130 may implement a control function for the user wearable device and perform an additional signal processing function on the detected signal. Local processor 154 or processor 130 may also be referred to as a diagnostic processor.

Although the user wearable device 100 may be worn on the wrist, various embodiments of the present disclosure need not be limited thereto. In addition, the user wearable device 100 is designed to be worn on a portion of the body, for example, on the arm (forearm, elbow or upper arm), on the legs, on the chest, on the head like a headband, on the neck, like a choker, on the ear Can be. The user wearable device 100 may communicate with other electronic devices, such as, for example, a smartphone, laptop or various medical devices in a hospital or doctor's office.

Display 160 may output monitored physiological signals from the user's body for the user and / or others to see. Monitored physiological signals are often referred to as bio signals or biometric data. Monitored biosignals can be, for example, heart rate (pulse rate), pulse morphology (type), pulse interval (IBI), breathing rate and blood pressure. The display 160 may also output commands to the user or others, for example in the use of the user wearable device 100 or the use of other measurement devices, as well as status and diagnostic results.

The processor 130 receives signals monitored or detected from sensors in the sensor module 150. For example, the sensors 152, 156 acquire a signal from the wrist of the user when the user wearable device 100 is worn by the user. In an embodiment of the invention, the sensor module 150 includes a sensor 152 that is a biophysiological sensor. In one embodiment, the biophysiological sensor is a PPG sensor. In another embodiment, the sensor module 150 further includes a second sensor 156 that is an inertial measurement sensor. In one embodiment, the inertial measurement sensor is an accelerometer. The sensor module 150 may include a processor 154 for controlling the sensors 152 and 156 and for processing the signals sensed by the sensors. For example, processor 154 may disassemble signals monitored by sensors 152 and 156 and then reconstruct the disassembled signals. Various embodiments of the present disclosure may also have a processor 130 that performs the functions of the processor 154. Various embodiments of the present invention may also have different numbers of sensors.

In some embodiments, sensor 152 is a PPG sensor used to continuously or periodically monitor cardiac related physiological information such as a user's heart rate or heart pulse form. On the other hand, the sensor 156 is an accelerometer used for continuously or periodically monitoring the motion information of the user. Sensor module 150 may include other sensors, such as, for example, a thermometer for measuring a user's temperature.

The user wearable device 100 implements the PPG-based arrhythmia detection system of the present invention in the processor 130. In some embodiments, the PPG-based arrhythmia detection system includes a signal processing module for dividing the PPG signals into short PPG signal segments, and a machine for measuring the probability of the presence of cardiac arrhythmias in the monitored signal and evaluating the PPG signal segments. It includes a running network.

2 is a block diagram of a user wearable device according to an exemplary embodiment. Referring to FIG. 2, the user wearable device 100 includes a battery 170 for supplying power to the sensor module 150, the processor 130, the display 160, and other components. Processor 130 controls the output provided on display 160. Display 160 may also include input devices (not shown) such as, for example, buttons, dials, a touch sensitive screen, and a microphone.

In embodiments of the present disclosure, the sensor module 150 includes a biophysiological sensor 152 for measuring a biosignal of a user. In this embodiment, the biophysiological sensor 152 is a PPG sensor. The sensor module 150 may further include an inertial measurement sensor 156 for measuring a motion signal of the user. In this embodiment, the inertial measurement sensor 156 is an accelerometer, such as a three axis accelerometer. The sensor module 150 may be provided with a local processor 154 for controlling the sensors 152 and 156 and for processing biosignals and motion signals sensed by the sensors 152 and 156, respectively. In some embodiments, the signal processing processor may be implemented in local processor 154 and / or processor 130. Alternatively, local processor 154 may perform some portion of signal processing, such as specific signal preprocessing, and processor 130 may implement other signal processing algorithms for biometric determination or other functions. In embodiments of the present invention, the particular processor used to execute the biometric signal processing algorithm is not critical to the practice of the present disclosure.

In embodiments of the present invention, the processor 130 is configured to control a sensing operation, a sampling schedule, a signal processing operation, and a device communication event and other device specific functions in the user wearable device. In this embodiment, the processor 130 includes a CPU 132, a memory 134, an input / output interface 182, a communication interface 184, and a sensing module 190. The processor 130 is described as including several elements, but other embodiments may use other structures in which different functions are grouped in different ways. For example, the grouping may be in different integrated circuit chips. Alternatively, grouping may combine different components, such as input / output interface 182 and communication interface 184 together.

The processor 130 may integrate the sensing module 190 to perform arrhythmia detection on the sensed biosignal such as the PPG signal. In embodiments of the present invention, the sensing module 190 includes a data processing module 192, an IBI sensing module 194, a morphology sensing module 196, and a classification module 198. The signal processing module 192 is configured to perform signal processing on the sensed biosignal. For example, the data processing module 192 may perform baseline removal or direct current (DC) signal level removal on the detected PPG signal. In other embodiments, the data processing module 192 may perform signal splitting to split the sensed PPG signals into short PPG signal segments. For example, each PPG signal segment may be between 2 and 15 seconds. Alternatively, each PPG signal segment may include n sensed beats. In one embodiment, n is between 40 and 70. That is, each PPG signal segment may include 40 to 70 beats.

IBI sensing module 194 implements the analysis of the IBI in the sensed beats of the PPG signal segment. Morphology sensing module 196 implements analysis of morphology based features of the sensed pulsation of a PPG signal. In some embodiments, the sensing module 190 may include one or both of an IBI sensing module and a morphology sensing module.

The classification module 198 implements classification of PPG signal segments to detect the presence of arrhythmia in the PPG signal segments. The classification module 198 uses the analysis results from the IBI sensing module 194 and / or the analysis results from the morphology sensing module 196. The classification module 198 predicts the presence of arrhythmia in the PPG signal segment.

In other embodiments, the arrhythmia detection system of the present invention may be implemented in an electronic device that communicates with a user wearable device including a sensor module. For example, the electronic device may be a mobile device such as a smart phone or tablet device. Other accompanying signals such as sensed PPG signals and motion signals may be provided to the electronic device for signal processing and arrhythmia detection according to the arrhythmia detection method of the present invention as shown in FIG. 11. The electronic device provides the detection result to be displayed on the user wearable device.

In still other embodiments, the arrhythmia detection system of the present invention may be implemented in a cloud server disposed on a data network and may communicate with a user wearable device including a sensor module. Other accompanying signals such as the detected PPG signal and the motion signal may be provided to the cloud server through the data network for signal processing and arrhythmia detection according to the arrhythmia detection method of the present invention as shown in FIG. 12. The cloud server may provide a detection result to be displayed on the user wearable device. In this description, a cloud server refers to a logical server that is built, hosted and delivered via a cloud computing platform over a data network, such as the Internet. For example, a cloud server has similar performance and features as a normal server but can be accessed remotely from a cloud service provider.

3 is a flowchart illustrating a method of detecting arrhythmia in a user wearable device in embodiments of the present invention. In some embodiments, the method 200 may be implemented in a processor in a wearable device, such as the processor 130 of the user wearable device 100 of FIGS. 1 and 2. Alternatively, the method 200 may be implemented in a mobile device in communication with the wearable device. In yet another embodiment, the method 200 may be implemented in a cloud server in communication with a wearable device. Referring to FIG. 3, the method 200 receives a channel of biosignal data signals from a first sensor implemented in the user wearable device 202. For example, the biosignal data signal may be a PPG signal. In this embodiment, the method 200 receives raw data samples, that is, unprocessed or minimally processed data samples.

At operation 204, the method 200 performs processing of PPG signals to obtain PPG signal samples. In some examples, the processing of the PPG signals may include removing the direct current baseline signal level. In other examples, the processing may include other signal processing that enhances the signal level. As a result of the processing, PPG signal samples are generated. In one embodiment, the processing step is performed by the data processing module 192 in the sensing module 190 of the processor 130 of the user wearable device 100.

4 shows exemplary signal waveforms of an ECG signal and a PPG signal. In particular, the PPG signal for measuring heart rhythm or heart rate has a contour of a specific waveform and is different from the contour of the waveform of an ECG signal. Referring to FIG. 4, the ECG measures the electrical activity of the heart and the ECG signal (curve 250) includes a salient feature known as the QRS complex that represents the major pumping contraction of the heart. The R peak in the ECG signal is used by the heart rate algorithm to measure the time that occurs between pulsating pulses. The duration between each R peak is referred to as the RR interval.

PPG, on the other hand, measures the pressurized pulse acting on the arteries of the body, which slightly expands the artery before the artery returns to its previous state. A PPG signal is an optical signal in which the amplitude of the optical signal is directly proportional to the pulse pressure. The PPG signal (curve 252) can be used to indicate the periodicity of the signal waveform and include quasi-periodic pulses with peaks and valleys that allow estimation of the heart rate. In particular, the duration between the peaks of two adjacent pulses or between the minimum points of two adjacent pulses is referred to as Inter-beat Interval (IBI), which can be used as an indicator of heart rate. In some cases, the PPG signal exhibits redundant scars. Redundant cuts are small downward breaks observed in the downstroke of the arterial pressure waveform. This represents the intersection of the primary and reflected pressure waves superimposed in the artery tree.

Referring again to FIG. 3, at operation 206, the method 200 detects a rhythm in the PPG signal samples. In one embodiment, the method 200 detects the peaks and troughs in the signal waveform of the PPG signal and uses the sensed peaks or troughs to indicate the location of the pulsation in the PPG signal samples. 5 shows an example signal waveform of a sequence of PPG signal samples. Referring to FIG. 5, in this embodiment, the method 200 senses the lowest points of the signal waveform to determine the location of the rhythm or heartbeat in the PPG signal samples. Thus, the method 200 identifies the boundary of each pulse in PPG signal samples as a beat.

Referring again to FIG. 3, at operation 208, the method 200 splits the PPG signal samples into PPG signal segments. In one embodiment, the method 200 partitions PPG signal samples into PPG signal segments having a given time duration, such as t seconds. For example, each PPG signal segment may be between 2 and 15 seconds. In another embodiment, the method 200 splits the PPG signal samples into n beats of PPG segments. For example, each PPG signal segment may include a beat between 40 and 70 times.

In embodiments of the present disclosure, the method 200 collects beats for a given time duration to form a PPG signal segment, and the beats of the PPG signal segments may or may not be continuous in time. In other embodiments of the invention, the method 200 collects a given number of beats to form a PPG signal segment, and the beats of the PPG signal segment may or may not be continuous in time. The method 200 may collect some beats in the PPG signal samples, discard some beats, and then resume collecting other beats to form a PPG signal segment.

At operation 210, the method 200 extracts IBI features from each PPG signal segment. In particular, the method 200 evaluates the PPG signal segments to analyze the statistical regularity of the distribution of the IBI of the PPG beats included in each PPG signal segment. In this way, the method 200 may extract irregularly irregular IBI characteristics. In one embodiment, the IBI feature extraction step is performed by the IBI sensing module 194 in the sensing module 190 of the processor 130 of the user wearable device 100.

For normal PPG pulses, the time interval between an individual's heartbeat changes in a fairly predictable manner due to breathing and other long-term sympathetic responses. However, when an individual suffers from arrhythmia, the IBI becomes very irregular due to abnormal activation patterns present in the tissues, making the intervals clearly statistically more irregular and less predictable. Abnormal IBI can be observed by comparing normal PPG pulses with PPG pulses with arrhythmia.

In particular, FIG. 5 shows PPG signal samples collected from subjects of normal orbital nodule rhythm. IBI durations vary from 1.04 seconds to 1.12 seconds. The IBI duration varies with the PPG pulses, but the IBI duration varies in a constant and predictable manner. 6 shows an example signal waveform of a series of PPG signal samples with atrial fibrillation. For PPG signal samples with arrhythmia, the IBI duration varies greatly with the PPG pulse set. In the examples, the IBI duration varies from 0.56 seconds to 1.14 seconds. These irregular irregularities are indicative of arrhythmia or atrial fibrillation.

In embodiments of the present disclosure, the method 200 analyzes each PPG signal segment and extracts the 'irregular irregularity' feature of the IBI from the PPG signals. In some embodiments, the irregularities of the IBI of the PPG signal are characterized using one or more statistical measures of the distribution of IBI amounts. In some embodiments, the method 200 implements one or more statistical measures to assess the IBI duration. In one example, the statistical measures may include standard deviation, asymmetry, kurtosis, information entropy, root mean square root of successive difference of IBI (RMSSD), turning point ratio, and multiscale sample entropy. Statistical measures are used to extract features of the PPG signal segment that represent deviations from normal orbital nodal rhythms.

FIG. 7 is a plot showing the histogram of the root mean square of the successive difference in IBI of PPG signals with atrial fibrillation and PPG signals with normal orbital nodule rhythm. Referring to FIG. 7, by using the square root mean square of the continuous difference analysis, PPG signals with abnormal IBI and PPG signals with normal IBI can be easily distinguished.

8 is a plot showing a histogram of multiscale sample entropy analysis of IBI of atrial fibrillation PPG signals and PPG signals with normal orbital nodule rhythm. Referring back to FIG. 8, by using multiscale sample entropy analysis, PPG signals with abnormal IBI and PPG signals with normal IBI can be easily distinguished.

Returning to FIG. 3, at operation 212, the method 200 extracts morphology based features from each PPG signal segment. In particular, the method 200 evaluates PPG signal segments to analyze the morphology based features of the PPG beats included in each PPG signal segment. Morphology-based features may include statistical features, measured features or features derived from PPG rhythm of PPG signal segments. Morphology-based features may include waveform shape, waveform characteristics, and waveform quality of the PPG beats in the PPG signal segment. In one embodiment, the method 200 analyzes the statistical distribution of similarity and morphology characteristics between the morphology features of adjacent PPG beats extracted from each PPG signal segment. In one embodiment, the morphology feature extraction step is performed by the morphology detection module 196 in the sensing module 190 of the processor 130 of the user wearable device 100.

As shown by the PPG pulse waveforms in FIG. 5, in normal isotropic rhythms, adjacent PPG beats from the subjects of normal isotropic rhythms are very similar in waveform shape / shape. However, as shown by the PPG pulse waveform in FIG. 6, this is not the case for arrhythmia. For example, irregularly arriving rhythms generally prevent the blood of the previous rhythm from being fully dispersed in the skin to cause a DC signal rise, while similar negative baseline motion is seen in late arriving PPG rhythms. Similarly, the perfusion detected by the PPG waveform may have a different shape and pressure reflection from the surroundings because the heart muscle is abnormally contracted and the blood is ejected differently, making up an additional reflection waveform superimposed on the original pressure pulse. As a result, morphological features can be used as a good means of distinguishing normal isotonic nodes from cardiac arrhythmias.

In embodiments of the invention, the arrhythmia detection method 200 extracts morphology based features including a standard deviation of the area (AUC) under the curve of the PPG beats in the PPG signal segment. For example, FIG. 9 is a plot showing the histogram of the standard deviation of the area under the curve for normal isotopes and atrial fibrillation. The morphological characteristic difference between the histogram of normal orbital nodule and the histogram of atrial fibrillation can be clearly seen in FIG. 9. The method 200 assesses the standard deviation of the area under the curve of the PPG beats in the PPG signal segment to detect signs of cardiac arrhythmias.

In other embodiments, the arrhythmia detection method 200 extracts morphology based features including waveform similarity of adjacent PPG beats in the PPG signal segment. Waveform similarity can be assessed using cross correlation or similarity measures. The method 200 evaluates the waveform similarity measure of the PPG beats in the PPG signal segment to detect signs of possible arrhythmia.

In other embodiments, the arrhythmia detection method 200 extracts morphology based features that include the ratio of PPG beats with long tails in the PPG signal segment. For example, the method 200 detects the incidence of PPG beats with long tails. In this description, a PPG beat with a long tail refers to a PPG beat waveform with an extended downward slope. That is, the long tail feature refers to an extended downstroke of the arterial pressure waveform. 10 shows an example signal waveform of a series of PPG signal samples showing long tail features. Referring to FIG. 10, the label LT indicates PPG pulses with long tails. The method 200 evaluates the PPG beats in the PPG signal segment to detect the number of beats with the long tail feature. The incidence of long tail features is one morphology-based feature that can be used to indicate the likelihood of arrhythmia. In some embodiments, the long tail feature is detected by using pattern recognition technology.

In other embodiments, the arrhythmia detection method 200 extracts morphology based features that include a PPG beat rate with an abnormal notch in the PPG signal segment. For example, the method 200 detects the incidence of PPG beats with abnormal notches. An abnormal notch differs from the previously described overlapping notch in that an abnormal notch represents two incomplete abnormal heartbeats, while the redundant notch represents one normal heartbeat. Referring to FIG. 10, the label AN shows PPG pulses with abnormal notches. The method 200 evaluates the PPG beats in the PPG signal segment to detect the number of beats with abnormal notch features. The incidence of abnormal notch features is one morphology based feature that can be used to indicate the likelihood of arrhythmia. In some embodiments, abnormal notch features are detected by using a pattern recognition technique.

In other embodiments, the arrhythmia detection method 200 extracts morphology based features that include a standard deviation of alternating current (AC) components of rising edges of the PPG rhythm and a standard deviation of falling components of falling edges in the PPG signal segment. PPG waveforms include an alternating current component and a direct current component. The alternating current component is synchronized with the heart rate to match the change in blood volume. The direct current component arises from the optical signals reflected or transmitted by the tissue and is determined by the tissue structure as well as the venous and arterial blood volume. The DC component shows a slight change in breathing. The fundamental frequency of the alternating current component varies with heart rate and is superimposed on a direct current reference line. In one embodiment, the method 200 calculates the alternating amplitude of the rising edge of the heartbeat as the positive region of the first derivative of the PPG waveform, and the alternating amplitude of the falling edge of the heartbeat is negative of the first derivative of the PPG waveform. Calculate as the area of. In another embodiment, the PPG alternating pulse waveform contour may be characterized by signal samples, systolic peak amplitude, diastolic peak amplitude, second derivative limit of the pulse waveform. In embodiments of the invention, the method 200 evaluates the standard deviation of the alternating current components of the PPG rhythm in the PPG signal segment to detect signs of arrhythmia.

At operation 214, the method 200 classifies PPG signal segments using the extracted IBI features and / or respective extracted morphology features associated with each PPG signal segment. The method 200 classifies the PPG signal segment using a previously trained machine learning model based on signals and arrhythmia annotation from one or more sets of arrhythmia training data. In one embodiment, the method 200 uses a combination of IBI features and morphology features to classify the PPG signal segment. In particular, certain morphological features indicate cardiac arrhythmias. Thus, the combination of IBI and morphology-based features can play an important role in increasing the likelihood of arrhythmia prediction. In one embodiment, the classification step is performed by the classification module 198 in the sensing module 190 of the processor 130 of the user wearable device 100.

In some embodiments, method 200 uses a random forest classifier to perform the classification. Random Forest is an ensemble method constructed by combining each different and independent base classifiers / decision trees. Each independent classifier is trained using the data set sampled in place of the original training data set. Features with the maximum information gain are selected to be separated. The best partitioning functions are identified from a random subset of the available functions. This backing / bootstrap integration has the advantage of reducing overfitting, allowing the model to be generalized to a larger population while reducing error rates. In one embodiment, to enable implementation on an embedded system, the random forest model uses only three decision trees and each tree has a depth of five. In this way, the method 200 enables real time measurement and prediction of arrhythmia events.

The term 'machine learning model' as used herein refers to classification models that can be used to train to provide accurate classification. Indeed, training of the classification model is performed on high power computers and the trained model is placed on the device where inference using the model is performed. In some embodiments, any machine learning and / or classification techniques may be used to perform the classification of the PPG features described above. Briefly described, an embodiment of the present invention relates to arrhythmia detection or event prediction using machine learning that can be incrementally improved based on expert input. In at least one of the various embodiments, the data may be provided to a machine learning model trained using a plurality of classifiers (index, labels or annotations) and one or more sets of training data and / or test data.

At operation 216, the method 200 generates an event prediction result based on the extracted IBI features and / or the extracted morphology features associated with each PPG signal segment. In one embodiment, the method 200 uses the IBI features and / or morphology features of the PPG segments to generate an arrhythmia detection result based on the classification of the PPG segments.

In some embodiments, in response to detecting the presence of arrhythmia, the arrhythmia detection method 200 sends a notification to the user. For example, the notification may be sent via an application on the mobile device and / or wearable device.

Using the foregoing analysis, the PPG based arrhythmia detection system takes as input a number of short discrete segments of PPG signals using IBI features and / or morphology features extracted from these segments and provides a detection result. In this way, a passive sensing system that can be used for the user's day is realized.

In some embodiments, method 200 receives motion information associated with user wearable device 218. For example, motion information may be obtained from a second sensor, such as an inertial measurement sensor or an accelerometer. The method 200 may use motion information during the PPG signal segmentation step 208 to discard unreliable PPG signal samples associated with high intensity motion. Or, because the PPG sensor is turned off during high-intensity motion, the PPG signal cannot be used during this period. Thus, in embodiments of the present invention, the PPG signal segments thus generated may be PPG signal samples that are not necessarily contiguous in time but are not connected. Arrhythmia detection method 200 may operate on short segments of PPG signal samples where each segment of PPG signal samples may be discontinuous in time.

In some embodiments, method 200 receives (220) ECG signals. The method 200 may use the ECG signal to adjust the sensed beats in the PPG signal segments. In one embodiment, the method 200 uses an ECG signal for information about heart rate, which can be used to adaptively adjust the time interval between decision thresholds and beats to improve detection accuracy. It should be noted that the method 200 uses the ECG signal only for threshold adjustment, and the determination of the presence or absence of arrhythmia is made using the PPG signal alone.

In one embodiment, if the arrhythmia is sufficiently present in a given period of PPG during a stationary state, the user may be asked to perform an ECG measurement, and the user's sensors may be placed together on the wearable device. The combination of PPG and ECG measurements can be interpreted for the presence of arrhythmia by physicians or ECG analysis algorithms. The determination can then be presented to the user or stored for future cumulative analysis to identify chronic disease trends.

In the PPG based arrhythmia detection method 200 described above in FIG. 3, the method 200 illustrates the use of a combination of IBI features and morphology based features for arrhythmia detection. Using a combination of IBI features and morphology-based features improves detection accuracy, but the PPG-based arrhythmia detection method of the present invention can be implemented using only IBI features or using morphology-based features. IBI features or morphology based features provide an indication of arrhythmia that can be used as the basis for accurate and accurate arrhythmia detection.

Aspects of the present disclosure are described herein with reference to flowcharts or block diagrams, where each block or combination of arbitrary blocks can be implemented by computer program instructions. The instructions may be provided to a processor of a general purpose computer, special purpose computer or other programmable data processing devices to cause a machine or article of manufacture, wherein when executed by the processor the instructions may be assigned to each block or combination of blocks in the diagram, It is possible to create means for implementing an action or event.

In this regard, each block in the flowchart or block diagram may correspond to a module, segment or code portion that includes one or more executable instructions for implementing a particular logical function (s). It should also be noted that in some alternative implementations, functionality related to any block may occur out of the order noted in the figures. For example, two blocks shown in succession may in fact be executed substantially concurrently or the blocks may sometimes be executed in the reverse order.

Those skilled in the art will appreciate that aspects of the present specification may be implemented as an apparatus, system, method or computer program product. Accordingly, aspects of the present disclosure, generally referred to herein as circuits, modules, components, or systems, relate to a computer program product contained on a non-transitory computer readable medium having computer readable program code implemented thereon. Hardware, software (including firmware, resident software, micro-code, etc.), or any combination of software and hardware.

The foregoing detailed description is provided by way of explanation of particular embodiments of the invention and is not intended to be limiting. Many modifications and variations are possible within the scope of the invention.

Claims (23)

In the method for event detection in a user wearable device,
Receiving photoplethysmogram (PPG) signals from a first sensor implemented in the user wearable device;
Processing the PPG signals to obtain PPG signal samples at a processor;
Detecting a pulsation in the PPG signal samples at the processor;
Dividing the PPG signal samples into PPG signal segments;
Extracting at least one inter-beat interval (IBI) feature from each of the PPG signal segments;
In the processor, using the extracted IBI feature associated with each of the PPG signal segments and classifying each of the PPG signal segments using a machine learning model;
In response to the classifying, generating, at the processor, an event prediction result for one PPG signal segment of the PPG signal segments based on the extracted IBI feature; And
Displaying the event prediction result at the user wearable device. The method of claim 1,
Extracting at least one morphology based feature in each of the PPG signal segments;
At the processor, using the extracted IBI feature and the extracted morphology based feature associated with each of the PPG signal segments and using the machine learning model. Classifying each of the PPG signal segments; And
And in response to the categorizing, generating, at the processor, the event prediction result based on the extracted IBI feature and the extracted morphology based feature. The method of claim 1,
Extracting the at least one IBI feature from each of the PPG signal segments is:
Duration of the IBI of the detected beat in the PPG signal segments using one or more of standard deviation, asymmetry, kurtosis, information entropy, turning poing rate, root mean square of consecutive differences, and multiscale sample entropy Extracting the IBI feature by analyzing time. The method of claim 2,
Extracting the at least one morphology based feature in each of the PPG signal segments is:
Extracting the morphology based feature by analyzing one or both of the statistical distribution of morphological characteristics of adjacent beats in each of the PPG signal segments and the similarity between the morphology features. The method of claim 4, wherein
Extracting the morphology based feature by analyzing the one or all of:
A standard deviation of the area under the curve of the sensed beat in each of the PPG signal segments, the waveform similarity of the adjacent beat in the PPG signal segments, and a PPG with a long tail in the PPG signal segments One or more of the rate of beats, the rate of PPG beats with abnormal notches in the PPG signal segments, and the standard deviation of the alternating components of the rising and falling edges of the beats within the PPG signal segments. Extracting the morphology based feature by analyzing. The method of claim 1,
Dividing the PPG signal samples into the PPG signal segments comprises dividing the PPG signal samples into the PPG signal segments having a given duration of t seconds. The method of claim 1,
Dividing the PPG signal samples into the PPG signal segments comprises dividing the PPG signal samples into the PPG signal segments having n number of beats according to the number of beats. The method of claim 1,
Wherein the PPG signal samples in one or more PPG signal segments comprise PPG signal samples collected for a discontinuous duration. The method of claim 8,
Receiving a motion signal representing motion activity in the user wearable device from a second sensor implemented in the user wearable device;
In response to the motion signal, removing PPG signal samples associated with the high intensity of the motion activity of the PPG signal samples; And
Dividing the PPG signal samples into the PPG signal segments by dividing the remaining PPG signal samples collected during the discontinuous duration of the PPG signal samples. The method of claim 9,
The first sensor comprises a PPG sensor,
And said second sensor comprises an accelerometer. The method of claim 1,
Receiving electrocardiogram (ECG) signals at the processor;
Adjusting the PPG signal samples in each of the PPG signal segments using the ECG signals; And
Adjusting a decision threshold while classifying each of the PPG signal segments to increase sensing accuracy. The method of claim 1,
Generating the event prediction result for the one PPG signal segment of the PPG signal segments based on the extracted IBI feature may include the event indicating a probability that arrhythmia is present in a given PPG signal segment of the PPG signal segments. Generating a prediction result. The method of claim 1,
Providing a notification on the user wearable device in response to the event prediction result indicating a high probability that an event is present in a given PPG signal segment of the PPG signal segments. The method of claim 1,
Processing the PPG signals to obtain the PPG signal samples at the processor comprises processing the PPG signals to obtain the PPG signal samples at the processor implemented in the user wearable device. The method of claim 1,
Processing the PPG signals to obtain the PPG signal samples at the processor includes processing the PPG signals to obtain the PPG signal samples at the processor implemented at a mobile device in communication with the user wearable device. Way. The method of claim 1,
Processing the PPG signals to obtain the PPG signal samples at the processor includes processing the PPG signals to obtain the PPG signal samples at the processor implemented at a cloud server in communication with the user wearable device. Way. In the method for event detection in a user wearable device,
Receiving photoplethysmogram (PPG) signals from a first sensor implemented in the user wearable device;
Processing the PPG signals to obtain PPG signal samples at a processor;
Detecting a pulsation in the PPG signal samples at the processor;
Dividing the PPG signal samples into PPG signal segments;
Extracting at least one morphology based feature associated with statistical characteristics of the PPG signal samples or waveform characteristics of the PPG signal samples in each of the PPG signal segments;
In the processor, classifying each of the PPG signal segments using the extracted morphology based feature associated with each of the PPG signal segments and using a machine learning model;
In response to the categorizing, generating, at the processor, an event prediction result based on the extracted morphology based feature; And
Displaying the event prediction result at the user wearable device. The method of claim 17,
Extracting the at least one morphology based feature from each of the PPG signal segments:
Extracting the morphology based feature by analyzing one or both of the similarity between the morphological features of adjacent beats in each of the PPG signal segments and the statistical distribution of the morphology features. The method of claim 18,
Extracting the morphology based feature by analyzing the one or all of:
A standard deviation of the area under the curve of the sensed beat in each of the PPG signal segments, the waveform similarity of the adjacent beat in the PPG signal segments, and a PPG beat with a long tail in the PPG signal segments. Analyzes one or more of a ratio of, a ratio of PPG beats with abnormal notches in the PPG signal segments, and a standard deviation of the alternating components of the rising and falling edges of the beat within the PPG signal segments. Thereby extracting the morphology based feature. A sensor module comprising a first sensor configured to measure PPG signals; And
Include processors,
The processor is:
A data processing module configured to process the PPG signals to obtain PPG signal samples, detect a pulsation in the PPG signal samples, and divide the PPG signal samples into PPG signal segments;
An IBI sensing module configured to extract a feature of at least one IBI in each of the PPG signal segments;
A morphology sensing module configured to extract at least one morphology based feature in each of the PPG signal segments; And
A classification module configured to classify each of the PPG signal segments using a feature of the extracted IBI and the extracted morphology based feature associated with the segment and using a machine learning model,
The classification module is configured to generate an event prediction result based on the extracted IBI feature and the extracted morphology based feature. The method of claim 20,
The IBI sensing module uses one or more of standard deviation, asymmetry, kurtosis, information entropy, turning poing ratio, root mean square of consecutive differences, and multiscale sample entropy to detect the detected in the PPG signal segments. And extract the IBI feature by analyzing the duration of the IBI in the beat. The method of claim 20,
And the morphology detection module is configured to extract the morphology based feature by analyzing one or both of a statistical distribution of morphological characteristics of adjacent beats in each of the PPG signal segments and similarities between the morphology features. The method of claim 22,
The morphology sensing module is configured to provide a standard deviation of the area under the curve of the sensed beat in each of the PPG signal segments, the waveform similarity of the adjacent beats in the PPG signal segments, and the long tail in the PPG signal segments. standard of the rate of PPG beats with a tail, the rate of PPG beats with abnormal notches in the PPG signal segments, and the alternating component of the rising and falling edges of the beats within the PPG signal segments. And extract the morphology based feature by analyzing one or more of the deviations.

Images