CN110301911A

CN110301911A - Passive arrhythmia detection device and method based on photoplethysmogra

Info

Publication number: CN110301911A
Application number: CN201910211570.8A
Authority: CN
Inventors: 苗丽峰; 张满满; M.威金斯
Original assignee: Samsung Electronics Co Ltd
Current assignee: Samsung Electronics Co Ltd
Priority date: 2018-03-27
Filing date: 2019-03-20
Publication date: 2019-10-08
Also published as: KR20190113552A; DE102019104798A1

Abstract

A kind of event detecting method in user's wearable device, the first sensor including realizing from user's wearable device receive photoplethysmogra (PPG) signal；PPG signal is handled at processor to obtain PPG sample of signal；The heartbeat in PPG sample of signal is detected at processor；PPG sample of signal is divided into PPG signal segment；At least one heartbeat interval (IBI) feature is extracted in each PPG signal segment；Classify at processor to each PPG signal segment using extracted IBI feature associated with PPG signal segment and using machine learning model；In response to classification, the event prediction result of PPG signal segment is generated based on extracted IBI feature at processor；And event prediction result is shown on user's wearable device.In another embodiment, this method further includes extracting the feature based on form.

Description

Passive arrhythmia detection device and method based on photoplethysmogra

Cross reference to related applications

This application claims entitled " the PASSIVE ARRHYTHMIAS DETECTION submitted on March 27th, 2018 The U.S. of BASED ON PHOTOPLETHYSMOGRAM (PPG) INTER-BEAT INTERVALS AND MORPHOLOGY " faces When patent application No.62/648,821 priority is incorporated herein by reference for all purposes.

The application is entitled " the METHOD AND APPARATUS FOR TRIAGE AND submitted on May 3rd, 2016 The U.S. Patent application No.15/ of SUBSEQUENT ESCALATION BASED ON BIOSIGNALS OR BIOMETRICS " 145,356 part continuation application, is incorporated herein by reference for all purposes.

Technical field

This disclosure relates to a kind of medical monitoring device and method thereof, more particularly to it is a kind of for arrhythmia cordis system detected System and method.

Background technique

Arrhythmia cordis (heart arrhythmia), also referred to as arrhythmia or cardiac arrhythmia, be one group of irregular heartbeats, Too fast or too slow illness.Although most types of arrhythmia cordis is not serious, some people are susceptible to suffer from such as apoplexy or mental and physical efforts The complication such as failure.Other people may cause sudden cardiac arrest.For example, auricular fibrillation (Atrial Fibrillation, AFib) is One of most common arrhythmia cordis, and the presence of AFib may potentially result in serious health risk.Traditionally, pass through the heart Electrograph (ECG) or Hotler monitor detect arrhythmia cordis.

ECG measurement needs complicated detection device, has the multiple electrodes for being attached to patient and needs positive people Work participates in.In general, ECG measurement only uses after symptom occurs in patient for diagnostic purpose.Photoplethysmogra (Photoplethysmogram, PPG) has been described as the alternative solution of ECG in arrhythmia detection.However, some be used for the heart It restrains not normal conventional PPG measuring technique and depends on heartbeat or heart rate detection using average heart rate, between such as 5 to 20 seconds Window.Average heart rate variation is not the reliable markers of arrhythmia cordis in itself.

Other methods based on PPG for detecting arrhythmia cordis have muting sensitivity and specificity.In addition, certain methods A large amount of continuous PPG signals are needed, such as 30 seconds continuous P PG signals measure to realize.

The example embodiment of the disclosure can provide the arrhythmia detection device and method based on PPG, be able to carry out Passively with asymptomatic arrhythmia detection.

Summary of the invention

Present disclosure disclose it is a kind of as more completely illustrate in claim for substantially as shown below and/ Or the arrhythmia cordis device and method detected of (such as in conjunction at least width attached drawing) described below.

From the following description and drawings will be more fully understood the disclosure these and other advantages, in terms of and novel feature, And the details of its illustrated embodiment.

In one embodiment, a kind of event detecting method in user's wearable device includes: from can in user The first sensor realized in wearable device receives photoplethysmogra (PPG) signal；At processor handle PPG signal with Obtain PPG sample of signal；The heartbeat in PPG sample of signal is detected at processor；PPG sample of signal is divided into PPG signal Section；At least one heartbeat interval (IBI) feature is extracted in each PPG signal segment；Using extracted related to PPG signal segment The IBI feature of connection simultaneously classifies to each PPG signal segment at processor using machine learning model；In response to classifying, The event prediction result of PPG signal segment is generated at processor based on extracted IBI feature；And on user's wearable device Show event prediction result.

In another embodiment, a kind of event detecting method in user's wearable device includes: from can in user The first sensor realized in wearable device receives photoplethysmogra (PPG) signal；At processor handle PPG signal with Obtain PPG sample of signal；The heartbeat in PPG sample of signal is detected at processor；PPG sample of signal is divided into PPG signal Section；At least one feature based on form is extracted in each PPG signal segment, the feature based on form and PPG sample of signal Statistical property or the waveform characteristic of PPG sample of signal are associated；Using extracted associated with PPG signal segment based on form Feature and classified at processor to each PPG signal segment using machine learning model；In response to classification, in processor Place generates event prediction result based on the extracted feature based on form；And show that event is pre- on user's wearable device Survey result.

In another embodiment, a kind of device includes: sensor module, which includes being configured as measurement light The first sensor of electric photoplethysmogram (PPG) signal；And processor, the processor include: data processing module, are configured For processing PPG signal to obtain PPG sample of signal, the heartbeat in detection PPG sample of signal and PPG sample of signal divided For PPG signal segment；Heartbeat interval detection module is configured as extracting at least one heartbeat interval in each PPG signal segment (IBI) feature；Morphology observation module is configured as extracting feature of at least one of each PPG signal segment based on form；With And categorization module, it is configured with extracted IBI feature and the extracted feature based on form associated with segmentation And classified using machine learning model to each PPG signal segment, categorization module is additionally configured to special based on extracted IBI The extracted feature based on form of seeking peace generates event prediction result.

The effect of the disclosure

Arrhythmia detection device disclosed herein based on PPG, which can be configured as using short PPG signal segment, to be extracted Accurate arrhythmia cordis information.Therefore, it is passively examined with asymptomatic arrhythmia cordis according to an embodiment of the invention, being able to carry out It surveys.

Detailed description of the invention

The various embodiments of the disclosure are disclosed in features as discussed above.

Fig. 1 (a) and Fig. 1 (b) show the electronic equipment according to the embodiment of the present disclosure.

Fig. 2 shows the block diagrams according to the user wearable device of the embodiment of the present disclosure.

Fig. 3 is the stream of the arrhythmia detection method in the user's wearable device shown in embodiment of the disclosure Cheng Tu.

Fig. 4 shows the exemplary signal waveforms of ECG signal and PPG signal.

Fig. 5 shows the exemplary signal waveforms of PPG sequence of signal samples.

Fig. 6 shows the exemplary signal waveforms of the PPG sequence of signal samples with auricular fibrillation.

Fig. 7 be show the PPG signal with normal sinus rhythm and the PPG signal with auricular fibrillation IBI it is continuous The curve graph of the histogram of the root mean square (Root Mean Square) of difference.

Fig. 8 is the more rulers for showing the IBI of the PPG signal with normal sinus rhythm and the PPG signal with auricular fibrillation Spend the curve graph of the histogram of Sample Entropy analysis.

Fig. 9 is the curve for showing the histogram of the standard deviation of area under the curve of normal sinus rhythm and auricular fibrillation Figure.

Figure 10 shows the exemplary signal waveforms of the PPG sequence of signal samples of display long-tail feature.

Figure 11 shows user's wearable device with mobile device communication in embodiment of the disclosure.

Figure 12 shows the user's wearable device communicated in embodiment of the disclosure with Cloud Server.

Specific embodiment

The disclosure can realize in many ways, including as process；Device；System；The composition of substance；Included in calculating Computer program product on machine readable storage medium storing program for executing；And/or processor, such as hardware processor or processor device are matched It is set to execution and is stored in the instruction that on the memory for being coupled to processor and/or the memory by being coupled to processor provides.? In this specification, any other form that these are realized or the disclosure can use can be referred to as technology.In general, can be at this Change the order of the steps of the disclosed process in scope of disclosure.Unless otherwise stated, being such as described as being configured It may be implemented as provisional configuration for the component of the processor or memory that execute task as in the logical of given time execution task With component or it is manufactured to the specific components of execution task.As it is used herein, term " processor " refers to being configured For one or more equipment, circuit and/or the processing core of the data of processing such as computer program instructions.

The following provide the detailed description to one or more other embodiments of the present disclosure and the principles for showing the disclosure Attached drawing.The disclosure is described in conjunction with these embodiments, but the present disclosure is not limited to any embodiments.The disclosure is replaced comprising many Generation, modification and equivalent.Numerous specific details are set forth in the following description, in order to provide the thorough understanding to the disclosure.It mentions It is that for exemplary purposes, and the disclosure can be practiced according to claim for these details, it is specific thin without these Some or all of section.For purposes of clarity, it is not described in known skill in technical field relevant to the disclosure Art material, so that the disclosure will not be obscured unnecessarily.

In embodiment of the disclosure, the arrhythmia detection system and method realized in user's wearable device pass through Analyze heartbeat interval (inter-beat interval, the IBI) feature and/or hop-by-hop shape of photoplethysmogra (PPG) signal State feature provides high accuracy arrhythmia detection using PPG signal.In some embodiments, arrhythmia detection system and Method handles PPG signal segment to analyze the heartbeat interval irregular at random and/or hop-by-hop morphological feature of the PPG waveform of user To detect arrhythmia cordis.In one embodiment, arrhythmia detection system and method are observed between the heartbeat in short PPG signal segment Phase (IBI) information and/or hop-by-hop shape information.Arrhythmia detection system and method based on PPG are configured as from PPG sections short Accurate arrhythmia cordis information is extracted, including the PPG signal message being usually rejected in conventional PPG method for measuring heart rate.? In some embodiments, the arrhythmia detection system and method based on PPG are provided better than the conventional detection side based on heart rate Accuracy in detection, the sensitivity and specificity of method.Importantly, the arrhythmia detection system and method based on PPG can be real Now passively with asymptomatic arrhythmia detection.That is, the disease of instruction arrhythmia cordis can be shown or undergone in user Arrhythmia detection is executed before shape.

Specifically, traditional arrhythmia detection techniques place one's entire reliance upon the statistical of the heartbeat interval in heartbeat message Cloth, and independent of shape information that may be present in heartbeat signal.It is lost in the rhythm of the heart of some aspects of the disclosure, the disclosure Normal detection system and method combine the analysis of the morphological feature in PPG signal, and therefore benefit from available in PPG signal Shape information exists in physiological condition.

In the present specification, photoplethysmogra (PPG) refers to the plethysmogram optically obtained, is organ Volume measurement.PPG is usually obtained by using pulse oximetry, pulse oximetry irradiation skin simultaneously measures light absorption Variation.Traditional pulse oximetry monitors the filling of corium and subcutaneous tissue of the result as cardiac cycle, blood to skin Note, is pumped into periphery for blood in cardiac cycle cardiac, to generate pressure pulse at skin.By utilizing come self-luminous The light of diode (LED) irradiates skin, then measures transmission or reflection and detects to the light quantity of photodiode by pressure pulse Caused volume change.The advantage of detection based on PPG is easily to record and monitor from consumer's grade wearable device PPG signal, the positive effort without participant.This advantage is mutually tied with economical and practical wearable device and smart phone It closes, the monitoring and detection of passively cardiac arrhythmia may be implemented.

In embodiment of the disclosure, the arrhythmia detection system and method based on PPG of the disclosure use PPG signal Short discrete segments to detect heart or arrhythmia cordis, and one or more signals analyses are executed to short PPG signal segment.At one In embodiment, the statistical regularity of the distribution of the heartbeat interval for the PPG heartbeat for including in each PPG signal segment of network analysis.? In another embodiment, the feature based on form for the PPG heartbeat for including in each PPG signal segment of network analysis.Show some In example, system can analyze morphological character statistical distribution or from each PPG signal segment extracted adjacent PPG heartbeat based on Similitude between the feature of form.In yet another embodiment, system execute two group analysis — that is, heartbeat interval characteristic and point Both morphological features are analysed to detect arrhythmia cordis.

Therefore, in some embodiments, the arrhythmia detection system and method based on PPG are by only analyzing PPG signal Heartbeat interval (IBI) feature of section provides the high accuracy heart based on the feature of hop-by-hop form by analysis PPG signal segment Restrain not normal detection.In some cases, information related with the hop-by-hop morphological feature of PPG signal segment is added based on IBI feature Into analysis to improve accuracy in detection.

In the present specification, the term " feature " used in " IBI feature " or " feature based on form " refers to waveform Shape, waveform characteristic, waveform quality and statistical property or attribute, the feature of measurement or attribute or derived character or attribute two Person.For example, IBI feature may include the statistical distribution of heartbeat interval.In another example, morphological feature may include adjacent Waveform similarity between PPG heartbeat.Morphological feature can also include the statistical distribution of certain morphological characters.

In alternative embodiments, the arrhythmia detection system and method based on PPG of the disclosure use electrocardiogram (ECG) signal adjusts PPG signal to further enhance accuracy in detection.In one embodiment, ECG signal is used in IBI The time interval of the heartbeat of detection is adaptively adjusted in signature analysis.In another embodiment, ECG signal was used in the classification phase Between be adaptively adjusted decision threshold to increase accuracy in detection.

Compared with the conventional arrhythmia detection method for using PPG, the arrhythmia detection system based on PPG of the disclosure Many advantages are realized with method.For example, implement the arrhythmia detection system based on PPG of the disclosure, so as to point It analyses many short PPG signal segments and does not need PG sections of long continuous P.From multiple PPG signal segments (for example, length is 2 to 15 Second) information/feature can be aggregated to be used for statistical estimation and classification.Short PPG signal segment is more required than in some legacy systems 30 seconds signal segments more frequently can be used.Secondly, average heart rate is used only in the arrhythmia detection system based on PPG of the disclosure Information provides detection more more accurate than legacy system.The arrhythmia detection system based on PPG of the disclosure uses heartbeat interval With hop-by-hop morphological feature, they can preferably indicate the characteristic of arrhythmia cordis and lead to the significant improvement of detection performance.

In an exemplary embodiment of the disclosure, the arrhythmia detection system and method based on PPG are sensed comprising PPG It is realized in the wearable device based on wrist of device.User or subject persistently dress the wearable device based on wrist, and System provides the notice of arrhythmia cordis detected.In some embodiments, the wearable device based on wrist includes acceleration Meter, is used to continuously determine whether wearer to be static.In one embodiment, when user is static, PPG optical sensor quilt Activation is to measure PPG signal.One group of key index of PPG signal is observed, to determine whether wearer suffers from arrhythmia cordis.Can In alternative embodiment, PPG optical sensor is continuously measured, and accelerometer is to the arrhythmia detection system based on PPG System provides movement indication signal.It arrhythmia detection system processing PPG signal based on PPG and can abandon and high degree of motion The part of associated PPG signal, the high degree of motion may influence the accuracy of PPG signal.

Fig. 1 (a) and Fig. 1 (b) show the electronic equipment according to the embodiment of the present disclosure.With reference to Fig. 1 (a) and Fig. 1 (b), electricity Sub- equipment 100 (can be user's wearable device) has display 160, processor 130, sensor module 150, battery (not Show), band 140 and retaining ring 142.Band 140 can wind in wrist, and user's wearable device 100 can be by making It is maintained in wrist with retaining ring 142.Sensor module 150 includes one or more sensors 152,156 and native processor 154 (not shown).Native processor 154 realizes the control function of sensor module, and the place of sensing signal can also be performed Reason or pretreatment.Processor 130 realizes the control function for being used for user's wearable device, and can also execute to sensing signal Further signal processing function.Native processor 154 or processor 130 can also be known as diagnostic processor.

Although user's wearable device 100 can be worn in wrist, the various embodiments of the disclosure are not necessary Limitation.User's wearable device 100 may be designed to be worn in the other parts of body, such as on arm (around forearm, ancon or upper arm), on leg, on chest, as headband on head, as " choker (choker) " one Sample is in throat and on ear.User's wearable device 100 can be communicated with other electronic equipments, other electricity The various Medical Devices of sub- equipment such as smart phone, laptop computer or hospital or doctor's office.

Display 160 can export the monitored physiological signal from user's body, for user and/or other people see It examines.Monitored physiological signal is sometimes referred to as bio signal or biological attribute data.The bio signal of monitoring can be for example Heart (pulse) rate, pulse form (shape), pulse interval (heartbeat interval), breathing (breathing) rate and blood pressure.For example, Display 160 can also be when using user's wearable device 100 or using other measuring devices and state and diagnostic result To user or other people output orders.

The signal that processor 130 receives monitoring from the sensor in sensor module 150 or sensed.For example, working as user When dressing user's wearable device 100, sensor 152,156 obtains signal from the wrist of user.In embodiment of the disclosure, Sensor module 150 includes the sensor 152 as biological physiology sensor.In one embodiment, biological physiology sensor It is photoplethysmogra (PPG) sensor.In other embodiments, sensor module 150 further includes passing as inertia measurement The second sensor 156 of sensor.In one embodiment, inertia measurement sensor is accelerometer.Sensor module 150 can be with Including processor 154, for controlling sensor 152,156, and it is also used to handle the signal sensed by sensor.For example, Processor 154 can decompose the signal monitored by sensor 152,156, then rebuild decomposed signal.The disclosure it is various Embodiment can make processor 130 also execute the function of processor 154.The various embodiments of the disclosure can also have different numbers The sensor of amount.

In some embodiments, sensor 152 is PPG sensor, for constantly or periodically monitoring the heart of user Dirty relevant physiologic information, such as heart pulse rate rate or heart pulse rate shape.Meanwhile sensor 156 is for continuously or all Monitor to phase property the accelerometer of the motion information of user.Sensor module 150 may include other sensors, such as use In the thermometer for obtaining user's body temperature.

User's wearable device 100 realizes the arrhythmia detection system based on PPG of the disclosure in processor 130. In some embodiments, the arrhythmia detection system based on PPG includes for being short PPG signal segment by PPG signal subsection Signal processing module, and further include machine learning network, it is used to assess PPG signal segment and estimates exist in monitored signal The probability of arrhythmia cordis.

Fig. 2 shows the block diagrams according to the user wearable device of the embodiment of the present disclosure.With reference to Fig. 2, user is wearable to be set Standby 100 include sensor module 150, processor 130, display 160 and for providing the battery of electric power to other assemblies 170.Processor 130 controls the output provided on display 160.Display 160 can also include input equipment (not shown), Such as button, dial, touch sensitive screen and microphone.

In embodiment of the disclosure, sensor module 150 includes life of the biological physiology sensor 152 to measure user Object signal.In the present embodiment, biological physiology sensor 152 is PPG sensor.Sensor module 150 may also include inertia survey Quantity sensor 156 is to measure the motor message of user.In the present embodiment, inertia measurement sensor 156 is accelerometer, such as Three axis accelerometer.Native processor 154 has can be set in sensor module 150, for controlling sensor 152,156, and It is also used to handle the bio signal and motor message sensed by sensor 152,156 respectively.In some embodiments, Ke Yi Signal processing operations are realized at native processor 154 and/or processor 130.Alternatively, native processor 154 can execute signal A part of processing, the signal processing such as signal specific pre-process, and processor 130 is realized and determined for biometric Or other signal processing algorithms of other function.In embodiment of the disclosure, for executing biometric signal Processing Algorithm Par-ticular processor it is not important for practicing for the disclosure.

In embodiment of the disclosure, processor 130 be configured as control user's wearable device in sensing operation, adopt Sample scheduling, signal processing operations and equipment communication event and other equipment specific function.In the present embodiment, processor 130 Including CPU 132, memory 134, input/output (I/O) interface 182, communication interface 184 and detection module 190.Although place Reason device 130 is described as including these various assemblies, other that different function is differently grouped can be used in other embodiments Framework.For example, can be grouped in different IC chips.Or grouping can such as will connect in I/O interface 182 and communication The different elements of mouth 184 are combined.

Processor 130 combines detection module 190 to execute arrhythmia cordis to the bio signal (such as PPG signal) sensed Detection.In embodiment of the disclosure, detection module 190 includes data processing module 192, IBI detection module 194, form inspection Survey module 196 and categorization module 198.Signal processing module 192 is configured as pre- to the bio signal execution signal sensed Processing.For example, data processing module 192 can execute the PPG signal sensed, baseline is removed or DC signal level removes.? In other embodiments, the PPG signal that data processing module 192 can execute signal subsection will be sensed is divided into short PPG letter Number section.For example, each PPG signal segment can be between 2 to 15 seconds.Alternatively, each PPG signal segment may include n detected Heartbeat.In one example, n is between 40 and 70.That is, each PPG signal segment may include 40 to 70 heartbeats.

IBI detection module 194 realizes the analysis to the heartbeat interval in the heartbeat of PPG signal segment detected.Form inspection Survey analysis of the realization of module 196 to the feature based on form of the heartbeat of PPG signal detected.In some embodiments, it examines Surveying module 190 may include either one or two module in IBI detection module and Morphology observation module.

Categorization module 198 realizes the classification of PPG signal segment to detect the presence of arrhythmia cordis in PPG signal segment.Classification mould Block 198 uses the analysis result from IBI detection module 194 and/or the analysis result from Morphology observation module 196.Classification Module 198 predicts that there are the probability of arrhythmia cordis in PPG signal segment.

In alternative embodiments, the arrhythmia detection system of the disclosure can with the user comprising sensor module It is realized in the electronic equipment of wearable device communication.For example, electronic equipment can be mobile device, such as smart phone or plate Equipment.According to the arrhythmia detection method of the disclosure, as shown in figure 11, can be moved by the PPG signal sensed and such as Other of signal are supplied to electronic equipment for signal processing and arrhythmia detection with signal.Electronic equipment can provide inspection Result is surveyed to be shown on user's wearable device.

In yet another embodiment, the arrhythmia detection system of the disclosure can arrangement over data networks and with packet It is realized in the Cloud Server of user's wearable device communication containing sensor module.As shown in figure 12, the PPG signal that is sensed and Other of such as motor message can be supplied to Cloud Server by data network, for the heart according to the disclosure with signal Restrain the signal processing and arrhythmia detection of not normal detection method.Cloud Server, which can provide, to be shown on user's wearable device The testing result shown.In the present specification, Cloud Server, which refers to, passes through cloud computing platform by data network (such as internet) The logical server of building, trustship and transmitting.For example, Cloud Server possesses and shows the performance and function similar with traditional server Can, but can be remotely accessed from cloud service provider.

Fig. 3 is the stream of the arrhythmia detection method in the user's wearable device shown in embodiment of the disclosure Cheng Tu.It in some embodiments, can be in the processor (use in such as Fig. 1 (a) and Fig. 1 (b) and Fig. 2 in wearable device The processor 130 of family wearable device 100) in implementation method 200.Alternatively, can be set in the movement communicated with wearable device Standby middle implementation method 200.In yet another embodiment, can in the Cloud Server communicated with wearable device implementation method 200.With reference to Fig. 3, method 200 receives biological signal data from the first sensor realized in user's wearable device (202) The channel of signal.For example, biological signal data signal can be PPG signal.In the present embodiment, method 200 receives original number According to sample, that is, the not yet data sample of processing or bottom line processing.

At 204, method 200 executes the processing of PPG signal to obtain PPG sample of signal.In some instances, PPG believes Number processing may include remove DC baseline signal level.In other examples, processing may include other signal processings to increase Strong signal level.As processing as a result, generating PPG sample of signal.In one embodiment, by user's wearable device 100 Processor 130 detection module 190 in data processing module 192 execute processing step.

Fig. 4 shows the exemplary signal waveforms of ECG signal and PPG signal.Specifically, for measuring the rhythm of the heart or heartbeat PPG signal has specific waveform profiles and is different from the waveform profiles of ECG signal.The electricity of heart is measured with reference to Fig. 4, ECG Activity and ECG signal (curve 250) include the prominent features of referred to as QRS complex (QRS complex), and the QRS is compound Wave indicates that the main pumping of heart is shunk.The peak R in ECG signal is occurred between each pulse heartbeat by algorithm of heart rate for measuring Time quantum.Duration between each peak R is known as the interval RR.

Meanwhile PPG measurement enters the pressurization pulse of the blood of the artery of body, this causes artery returning to its original state It slightly expands before.PPG signal is optical signal, and wherein the amplitude of optical signal is directly proportional to pulse pressure.PPG signal (curve 252) include pulse paracycle, there is peak and valley, can serve to indicate that the periodicity of signal waveform, to allow to carry out heart rate Estimation.Specifically, the duration between the peak of two adjacent pulses or the paddy of two adjacent pulses is referred to as heartbeat interval (inter-beat interval, IBI), may be used as the instruction of heart rate.In some cases, PPG signal shows dicrotic pulse Incisura (dicrotic notch).Dicrotic notch is that observe in the down stroke of arterial pressure waveform small deflects down.It Represent the intersection of the primary and reflected pressure wave that are superimposed in arterial tree.

Fig. 3 is returned to, at 206, method 200 detects the heartbeat in PPG sample of signal.In one embodiment, method 200 The peaks or valleys in the signal waveform of PPG signal are detected, and indicate the heartbeat in PPG sample of signal using peaks or valleys detected Position.Fig. 5 shows the exemplary signal waveforms of PPG sequence of signal samples.With reference to Fig. 5, in the present embodiment, method 200 Paddy in detection signal waveform is to determine the heartbeat in PPG sample of signal or the position of heartbeat.Therefore, method 200 is by PPG The boundary marking of each pulse in sample of signal is heartbeat.

Fig. 3 is returned to, at 208, PPG sample of signal is divided into PPG signal segment by method 200.In one embodiment, side PPG sample of signal is divided into the PPG signal segment with given duration (such as t seconds) by method 200.For example, each PPG signal Section can be 2 to 15 seconds.In another embodiment, PPG sample of signal is divided into the PPG section of n heartbeat by method 200.For example, Each PPG signal segment may include 40 to 70 heartbeats.

In embodiment of the disclosure, method 200 collects the heartbeat for giving the duration to form PPG signal segment, wherein Heartbeat in PPG signal segment can in time continuously or discontinuously.In the other embodiments of the disclosure, method 200 is collected The heartbeat of given quantity is to form PPG signal segment, and wherein the heartbeat in PPG signal segment can in time continuously or discontinuously. Method 200 can collect some heartbeats in PPG sample of signal, then abandon some heartbeats, then restore with collect it is some its His heartbeat is to form PPG signal segment.In one embodiment, remaining PPG signal sample after method 200 can be abandoned by division Originally PPG sample of signal was divided into PPG signal segment, the remaining PPG data sample is in the discontinuous of PPG sample of signal It is collected in duration.

At 210, method 200 extracts the heartbeat interval feature in each PPG signal segment.Specifically, method 200 is assessed PPG signal segment is to analyze the statistical regularity of the distribution of the heartbeat interval for the PPG heartbeat for including in each PPG signal segment.With this Kind mode, method 200 can extract irregular heartbeat interval characteristic at random.In one embodiment, it can be worn by user The IBI detection module 194 worn in the detection module 190 of the processor 130 of equipment 100 executes IBI characteristic extraction step.

For normal PPG pulse, since breathing and other long-term sympathetic nerve are reacted, the time interval between individual heartbeat Changed in a manner of quite predictable.However, when individual suffers from arrhythmia cordis, due in tissue there are abnormal activation mode, Heartbeat interval becomes very irregular, so that interval is obvious more unstable and statistically more unpredictable.By will be normal PPG pulse is compared with the PPG pulse with arrhythmia cordis, it can be observed that irregular heartbeat interval.

Specifically, Fig. 5 shows the PPG sample of signal of the collection of the subject from normal sinus rhythm.The IBI duration Second variation from 1.04 seconds to 1.12.Although the IBI duration changes on PPG pulse, the IBI duration be consistent and Change in a predictive manner.Fig. 6 shows the exemplary signal waveforms of the PPG sequence of signal samples with auricular fibrillation.? PPG sample of signal is deposited in the case of arrhythmias, the IBI duration PPG pulse collection close variation it is very big.In this example In, second variation from 0.56 second to 1.14 of IBI duration.This irregular scrambling is arrhythmia cordis or auricular fibrillation Instruction.

In embodiment of the disclosure, method 200 analyzes each PPG signal segment and extracts the " nothing of the IBI in PPG signal Regularly scrambling " feature.In some embodiments, it is counted using the one or more of distribution of heartbeat interval (IBI) amount Measure the scrambling in the IBI to characterize PPG signal.In some embodiments, method 200 implements one or more statistics surveys Amount is to assess the IBI duration.In one example, statistical measurement may include standard deviation (standard Deviation), the degree of bias (skewness), kurtosis (kurtosis), comentropy (information entropy), IBI company Continue root mean square (Root Mean Square of successive differences, RMSSD), the turning point ratio of difference (Turning point ratio) and multiple dimensioned Sample Entropy.Statistical measurement is used to extract the feature of PPG signal segment, wherein special Sign can indicate to deviate normal sinus rhythm.

Fig. 7 is show PPG signal with normal sinus rhythm and the IBI of the PPG signal with auricular fibrillation continuous The curve graph of the histogram of the root mean square of difference.It can be easily by using the root mean square analysis of continuous difference with reference to Fig. 7 Distinguish the PPG signal with normal IBI and the PPG signal with abnormal IBI.

Fig. 8 is the more rulers for showing PPG signal and the IBI of the PPG signal with auricular fibrillation with normal sinus rhythm Spend the curve graph of the histogram of Sample Entropy analysis.With reference to Fig. 8, is analyzed, can be easily distinguished by using multiple dimensioned Sample entropy PPG signal with normal IBI and the PPG signal with abnormal IBI.

Fig. 3 is returned to, at 212, method 200 extracts the feature based on form in each PPG signal segment.Specifically, side Method 200 assesses feature based in the form of of the PPG signal segment to analyze the PPG heartbeat for including in each PPG signal segment.Based on form Feature may include PPG heartbeat in PPG signal segment statistical property, measurement feature or export feature.Spy based on form Sign can also include waveform shape, waveform characteristic and the waveform quality of PPG heartbeat in PPG signal segment.In one embodiment, side Method 200 analyzes the statistical distribution of morphological feature and between the morphological feature of the extracted adjacent PPG heartbeat of each PPG signal segment Similitude.In one embodiment, it is examined by the form in the detection module 190 of the processor 130 of user's wearable device 100 It surveys module 196 and executes morphological feature extraction step.

When being in normal sinus rhythm, the adjacent PPG heartbeat of the subject from normal sinus rhythm is in waveform/form Upper height is similar, as shown in the PPG pulse wave in Fig. 5.Not but the case where this is not arrhythmia cordis, such as the PPG pulse wave in Fig. 6 Shown in shape.For example, the heartbeat formerly irregularly reached is generally not allowed the blood of previous heartbeats to be completely dispersed in skin, generate DC signal rises, and it can be seen that similar negative base linc motion in the PPG heartbeat of rear arrival.Similarly, due to myocardial abnormality It shrinks, and blood is differently sprayed, so the perfusion of PPG waveforms detection also can have the shape and pressure different from periphery Power reflection, constitutes the additivity reflection configuration being superimposed upon on reset pressure pulse.Therefore, morphological feature may be used as distinguishing normal sinus The good measure of the property rhythm of the heart and arrhythmia cordis.

In embodiment of the disclosure, arrhythmia detection method 200 extracts the feature based on form, including PPG signal The standard deviation of the area under the curve (AUC) of PPG heartbeat in section.For example, Fig. 9 is to show normal sinus rhythm and auricular fibrillation Area under the curve standard deviation histogram curve graph.It can clearly be observed that normal sinus rhythm is straight in Fig. 9 Morphological character difference between side's figure and auricular fibrillation histogram.Method 200 is assessed in PPG signal segment under the curve of PPG heartbeat The standard deviation of area is to detect the instruction of possible arrhythmia cordis.

In other embodiments, it includes in PPG signal segment that arrhythmia detection method 200, which extracts the feature based on form, The waveform similarity of adjacent PPG heartbeat.Crosscorrelation or similarity measurement can be used to assess waveform similarity.Method 200 The waveform similarity in PPG heartbeat in assessment PPG signal segment is measured to detect the instruction of possible arrhythmia cordis.

In other embodiments, arrhythmia detection method 200 is extracted in the feature based on form, including PPG signal segment The ratio of PPG heartbeat with long-tail.For example, the detection of method 200 has the generation percentage of the PPG heartbeat of long-tail.In this theory In bright book, the PPG heartbeat with long-tail refers to the PPG heartbeat waveform with the down slope extended.That is, long-tail feature Refer to the extended down stroke of arterial pressure waveform.Figure 10 shows the example of the PPG sequence of signal samples of display long-tail feature Property signal waveform.The PPG pulse with long-tail is indicated with reference to Figure 10, label LT.Method 200 assesses the PPG heart in PPG signal segment It jumps to detect the quantity of the heartbeat with long-tail feature.The appearance percentage of long-tail feature is a kind of feature based on form, It may be used to indicate possible arrhythmia cordis.In some embodiments, long-tail feature is detected by using mode identification technology.

In other embodiments, arrhythmia detection method 200 is extracted in the feature based on form, including PPG signal segment The ratio of PPG heartbeat with abnormal incisura (notch).For example, the detection of method 200 has the hair of the PPG heartbeat of abnormal incisura Raw percentage.Abnormal incisura is different from previously described dicrotic notch, because abnormal incisura represents two unfinished abnormal hearts It jumps, and dicrotic notch represents a normal heartbeat.The PPG pulse with abnormal incisura is indicated with reference to Figure 10, label A N.Method PPG heartbeat in 200 assessment PPG signal segments, to detect the quantity of the heartbeat with abnormal incisura feature.Abnormal incisura feature It is a kind of feature based on form that percentage, which occurs, can serve to indicate that possible arrhythmia cordis.In some embodiments, lead to Use pattern identification technology is crossed to detect abnormal incisura feature.

In other embodiments, arrhythmia detection method 200 is extracted in the feature based on form, including PPG signal segment PPG heartbeat rising edge exchange (AC) component standard deviation and failing edge AC component standard deviation.PPG waveform packet Include exchange (AC) component and direct current (DC) component.AC component corresponds to the variation with the blood volume of heartbeat synchronization.DC component comes from It is determined by Tissue reflectance or the optical signalling of transmission, and by institutional framework and vein and volume of arterial blood.DC component shows The minor change breathed out.The basic frequency of AC component changes with heart rate, and is superimposed upon on DC baseline.In one embodiment In, method 200 calculates the positive area and heart failing edge of the AC amplitude of heartbeat rising edge as the first derivative of PPG waveform Negative product of the AC amplitude as the first derivative of PPG waveform.In another embodiment, PPG exchanges (AC) pulse wave profile Characteristic can be pulse wave sample of signal, shrink peak-to-peak amplitude, diastole peak-to-peak amplitude, second dervative extreme value.In the disclosure In embodiment, method 200 assesses the standard deviation of the AC component of PPG heartbeat in PPG signal segment, to detect possible arrhythmia cordis Instruction.

At 214,200 use of method extracted IBI feature associated with each PPG signal segment and/or extracted Morphological feature classify to PPG signal segment.Method 200 classifies to PPG signal segment using machine learning model, the machine Device learning model previously based on from one or more groups of arrhythmia cordis training datas signal and arrhythmia cordis annotation be trained. In one embodiment, method 200 classifies to PPG signal segment using the combination of IBI feature and morphological feature.Specifically, Certain morphological features indicate arrhythmia cordis.Therefore, the combination of the feature using IBI and based on form can contribute to increase the rhythm of the heart A possibility that not normal prediction.In one embodiment, in the detection module 190 by the processor 130 of user's wearable device 100 Categorization module 198 execute classifying step.

In some embodiments, method 200 is come using random forest grader (random forest classifier) Execute classification.Random forest is a kind of integrated approach, by combining several different isolated footing classifier/decision trees come structure It builds.Using each Individual classifier of sampled data set training, and replaced from original training data collection.There is maximum information to increase for selection The feature of benefit is to be split.Optimal segmentation feature is identified from the random subset of available feature.This pack (bagging)/ Bootstrapping polymerization (bootstrap aggregation) has the advantages that reduce overfitting, therefore model can be generalized to more Big group, while reducing error rate.In one embodiment, in order to be realized in embedded system, random forest mould 3 decision trees are used only in type, and the depth of each tree is 5.In this way, method 200 realizes the reality of arrhythmic events When measure and prediction.

As it is used herein, term " machine learning model ", which refers to, can be used training to provide the classification of Accurate classification Model.In practice, the training of disaggregated model is executed on high power computer, trained model, which is then deployed in execution, makes In the equipment inferred with the model.In some embodiments, it can be held using any machine learning and/or sorting technique The classification of the above-mentioned PPG feature of row.In brief, embodiment of the disclosure is related to arrhythmia detection or thing using machine learning Part prediction, the machine learning can be incrementally improved based on expert's input.In at least one of various embodiments, it can incite somebody to action Data, which are supplied to, has used multiple classifiers (index, label or annotation) and one or more groups of training datas and/or test number According to the machine learning model being trained.

At 216, method 200 is based on extracted IBI feature associated with each PPG signal segment and/or is extracted Morphological feature generate event prediction result.In one embodiment, method 200 based on the IBI feature for using PPG sections and/or The classification of the PPG section of morphological feature generates arrhythmia detection result.

In some embodiments, in response to detecting the presence of arrhythmia cordis, arrhythmia detection method 200 is sent to user Notice.It is notified for example, can be sent via the application program in mobile device and/or wearable device.

Using above-mentioned analysis, the arrhythmia detection system based on PPG is using multiple short discrete segments of PPG signal as defeated Enter, using from these section of extracted heartbeat interval feature and/or morphological feature, and provides testing result.In this way, real Having showed can the passive detection system used in entire one day of user.

In some embodiments, method 200 receives motion information (218) associated with user's wearable device.For example, Motion information can be obtained from second sensor (such as inertia measurement sensor or accelerometer).Method 200 can be believed in PPG Abandoned during number division step (208) using motion information associated with high degree of motion --- therefore may be unreliable --- PPG sample of signal.Alternatively, PPG sensor can close during the high degree of motion period, therefore at these There is no PPG signal available during section.Therefore, in embodiment of the disclosure, the PPG signal segment thus generated need not be in time It is continuous, but can be incoherent PPG sample of signal.Arrhythmia detection method 200 can be to short PPG signal sample This section is operated, and wherein each of PPG sample of signal section can be discontinuous in time.

In some embodiments, method 200 receives ECG signal (220).ECG signal can be used to adjust in method 200 Heartbeat detected in PPG signal segment.In one example, method 200 obtains the information about heart rate using ECG signal, It can be used for being adaptively adjusted the time interval between decision-making value and heartbeat to improve accuracy in detection.It is noticeable It is that method 200 is used only ECG signal and carries out adjusting thresholds, and determines whether there is arrhythmia cordis using only PPG signal.

It in one example, can if sufficiently there is arrhythmia cordis in the given period of PPG during stationary state To require user to carry out ECG measurement, sensor can be co-located on wearable device.The combination of PPG and ECG measurement can be with The presence of arrhythmia cordis is interpreted as by doctor or ECG parser.Then the determination can be presented to the user or is stored and be used for The cumulative analysis in future is to identify chronic disease trend.

In the arrhythmia detection method 200 above described in Fig. 3 based on PPG, method 200 is shown for the heart Restrain the combination of the IBI feature and the feature based on form of not normal detection.Although using IBI feature and the group of the feature based on form Conjunction improves accuracy in detection, but IBI feature or only can be used only in the arrhythmia detection method based on PPG of the disclosure It is realized using based on the feature of form.The instruction of arrhythmia cordis is provided separately in IBI feature or feature based on form, can For use as the basis of accurate arrhythmia detection.

All aspects of this disclosure are described herein with reference to flow chart diagram or block diagram, wherein any combination of each frame or frame It can be realized by computer program instructions.General purpose computer, special purpose computer or other programmable numbers can be provided instructions to According to the processor of processing unit to realize machine or product, and when being executed by a processor, instruction creation is for realizing scheming In each frame or frame combination in function, the method for movement or event specified.

In this regard, each frame in flowchart or block diagram can correspond to include for realizing specified logic function The module, section of one or more executable instructions or a part of code.It shall yet further be noted that in some alternative embodiments, with Any associated function of frame can not be occurred by sequence shown in figure.For example, two frames continuously shown can actually It substantially simultaneously executes, or frame can be executed in reverse order sometimes.

It will be appreciated by the skilled addressee that all aspects of this disclosure can be presented as equipment, system, method or calculating Machine program product.Therefore, all aspects of this disclosure (being herein commonly referred to as circuit, module, component or system) can embody For any combination of hardware, software (including firmware, resident software, microcode etc.) or software and hardware, including embody on it Computer program product in non-transitory computer-readable medium with computer readable program code.

The specific embodiment being discussed in detail above be in order to illustrate the disclosure is provided, rather than it is restrictive.In this public affairs Many modifications and variations in the range of opening are possible.

Claims

1. the event detecting method in a kind of user's wearable device, comprising:

The first sensor realized from user's wearable device receives photoplethysmogra (PPG) signal；

The PPG signal is handled at processor to obtain PPG sample of signal；

The heartbeat in the PPG sample of signal is detected at the processor；

The PPG sample of signal is divided into PPG signal segment；

At least one heartbeat interval (IBI) feature is extracted in each PPG signal segment；

At the processor, using extracted IBI feature associated with each PPG signal segment and engineering is used Model is practised, is classified to each PPG signal segment；

In response to the classification, generated based on extracted IBI feature for the PPG signal segment at the processor The event prediction result of PPG signal segment；And

The event prediction result is shown on user's wearable device.

2. according to the method described in claim 1, further include:

At least one feature based on form is extracted in each PPG signal segment；

Using with each associated extracted IBI feature of PPG signal segment and the extracted feature based on form simultaneously Classified at the processor to each PPG signal segment using the machine learning model；And

In response to the classification, based on extracted IBI feature and the extracted feature based on form at the processor Generate the event prediction result.

3. according to the method described in claim 1, wherein, at least one described heartbeat is extracted in each PPG signal segment Interphase (IBI) feature includes:

By using standard deviation, the root mean square and multiple dimensioned sample of the degree of bias, kurtosis, comentropy, turning point ratio, continuous difference One or more of this entropy analyzes duration of the heartbeat interval of heartbeat detected in the PPG signal segment, to mention Take the IBI feature.

4. according to the method described in claim 2, wherein, at least one is based on described in extraction in each PPG signal segment The feature of form includes:

By analyzing between the statistical distribution of morphological character and the morphological feature of adjacent cardiac in each PPG signal segment One or both in similitude extracts the feature based on form.

5. one or both described based on form to extract by analyzing according to the method described in claim 4, wherein Feature includes:

By analysis one of the following or multiple extract the feature based on form: institute in each PPG signal segment The standard deviation of the area under the curve of the heartbeat of detection；The wave-form similarity of adjacent cardiac described in each PPG signal segment； The ratio of PPG heartbeat in each PPG signal segment with long-tail；There is abnormal incisura in each PPG signal segment The ratio of PPG heartbeat；And the standard deviation of exchange (AC) component of the rising edge of the PPG heartbeat in each PPG signal segment The standard deviation of the AC component of difference and failing edge.

6. according to the method described in claim 1, wherein, it includes inciting somebody to action that the PPG sample of signal, which is divided into the PPG signal segment, The PPG sample of signal is divided into the PPG signal segment with given duration t seconds.

7. according to the method described in claim 1, wherein, it includes inciting somebody to action that the PPG sample of signal, which is divided into the PPG signal segment, The PPG sample of signal is divided into the PPG signal segment by beats, and each PPG signal segment has n heartbeat.

8. according to the method described in claim 1, wherein, the PPG sample of signal in one or more PPG signal segments includes The PPG sample of signal collected within the discontinuous duration.

9. according to the method described in claim 8, further include:

The second sensor realized from user's wearable device receives the fortune indicated at user's wearable device Move movable motor message；

In response to the motor message, PPG signal sample associated with the high degree of motion activity of the PPG sample of signal is removed This；And

The PPG sample of signal is divided into the PPG signal segment by dividing remaining PPG sample of signal, it is described remaining PPG data sample was collected within the discontinuous duration of the PPG sample of signal.

10. according to the method described in claim 9, wherein, the first sensor includes photoplethysmogra (PPG) sensing Device, and the second sensor includes accelerometer.

11. according to the method described in claim 1, further include:

Electrocardiogram (ECG) signal is received at the processor；

The PPG sample of signal in each PPG signal segment is adjusted using ECG signal；And

Decision-making value is adjusted during classifying to each PPG signal segment to improve accuracy in detection.

12. according to the method described in claim 1, wherein, being generated based on extracted IBI feature and being used for the PPG signal segment The event prediction result of the PPG signal segment include: to generate the event prediction as a result, the event prediction result refers to Show that there are the probability of arrhythmia cordis in the given PPG signal segment of the PPG signal segment.

13. according to the method described in claim 1, further include:

In response to indicating that there are the event predictions of the high probability of event in the given PPG signal segment of the PPG signal segment As a result, providing notice on user's wearable device.

14. according to the method described in claim 1, wherein, it is described to obtain that the PPG signal is handled at the processor PPG sample of signal includes: that the PPG signal is handled at the processor realized in user's wearable device to obtain The PPG sample of signal.

15. according to the method described in claim 1, wherein, it is described to obtain that the PPG signal is handled at the processor PPG sample of signal includes: to handle at the processor realized in the mobile device communicated with user's wearable device The PPG signal is to obtain the PPG sample of signal.

16. according to the method described in claim 1, wherein, it is described to obtain that the PPG signal is handled at the processor PPG sample of signal includes: to handle at the processor realized in the Cloud Server communicated with user's wearable device The PPG signal is to obtain the PPG sample of signal.

17. the event detecting method in a kind of user's wearable device, comprising:

The PPG signal is handled at processor to obtain PPG sample of signal；

The heartbeat in the PPG sample of signal is detected at the processor；

The PPG sample of signal is divided into PPG signal segment；

Extract at least one feature based on form in each PPG signal segment, it is described based on the feature of form with it is described The statistical property of PPG sample of signal or the waveform characteristic of the PPG sample of signal are associated；

Existed using the extracted feature based on form associated with each PPG signal segment and using machine learning model Classify at the processor to each PPG signal segment；

In response to the classification, event prediction knot is generated based on the extracted feature based on form at the processor Fruit；And

The event prediction result is shown at user's wearable device.

18. according to the method for claim 17, wherein extract at least one described base in each PPG signal segment Include: in the feature of form

19. according to the method for claim 18, wherein one or both described based on form to extract by analyzing Feature include:

20. a kind of device, comprising:

Sensor module, the first sensor including being configured to measurement photoplethysmogra (PPG) signal；And

Processor, comprising:

Data processing module is configured as processing PPG signal to obtain PPG sample of signal, detects in the PPG sample of signal Heartbeat, and PPG signal segment is divided by PPG sample of signal is implemented；

Heartbeat interval detection module, at least one heartbeat interval (IBI) for being configured as extracting in each PPG signal segment are special Sign；

Morphology observation module is configured as extracting at least one in each PPG signal segment the feature based on form；And

Categorization module is configured with extracted IBI feature associated with the PPG signal segment and extracted is based on The feature of form simultaneously uses machine learning model, classifies to each PPG signal segment, the categorization module is also configured To generate event prediction result based on extracted IBI feature and the extracted feature based on form.

21. device according to claim 20, wherein the heartbeat interval detection module is configured as by using standard Deviation, the degree of bias, kurtosis, comentropy, turning point ratio, the root mean square of continuous difference and one or more in multiple dimensioned Sample Entropy It is a to extract the IBI feature to analyze the duration of the heartbeat interval of heartbeat detected in the PPG signal segment.

22. device according to claim 20, wherein the Morphology observation module is configured as each described by analyzing The one or both in similitude between the statistical distribution of morphological character in PPG signal segment and the morphological feature of adjacent cardiac To extract the feature based on form.

23. device according to claim 22, wherein the Morphology observation module is configured as by analyzing in following One or more extracts the feature based on form: in each PPG signal segment below the curve of heartbeat detected Long-pending standard deviation；The wave-form similarity of adjacent cardiac described in each PPG signal segment；Have in the PPG signal segment The ratio of the PPG heartbeat of long-tail；There is the ratio of the PPG heartbeat of abnormal incisura in each PPG signal segment；And each institute That states the standard deviation of exchange (AC) component of the rising edge of the PPG heartbeat in PPG signal segment and failing edge exchanges (AC) component Standard deviation.