System and method of detecting obstructive sleep apnea using photoplethysmography signals

Abstract

SYSTEM AND METHOD OF DETECTING OBSTRUCTIVE SLEEP APNEA USING PHOTOPLETHYSMOGRAPHY SIGNALS The present invention discloses a system 100 and method 400 for detecting OSA using PPG signals. The system 100 is connected with a wearable device 110 configured to acquire PPG signals. Further, the system 100 employs a Multivariate Long Short-Term Memory - Fully Convolutional Network ML model 160 to classify apnea events. The system 100 segments the PPG signals into windows, ensuring that apnea events remain centrally positioned. The ML model 160 extracts temporal and spatial features, leveraging LSTM layers 162, convolutional layers 164, squeeze-and-excitation (SE) blocks 166, and a global average pooling (GAP) layer 168. Further, the ML model 160 classifies windows into apnea or non-apnea events using a fully connected layer 170 with a sigmoid activation function. The invention improves OSA detection accuracy, enables real-time monitoring, and provides a cost-effective alternative to polysomnography. [To be Published with Figure 1]

Description

FIELD OF INVENTION

[0001] The present invention generally relates to data computing. Morespecifically, the present invention is related to a deep-learning-based approach fordetecting obstructive sleep apnea (OSA) using single-channelphotoplethysmography (PPG).BACKGROUND OF THE INVENTION

[0002] The subject matter discussed in the background section should not beassumed to be prior art merely as a result of its mention in the background section.Similarly, a problem mentioned in the background section or associated with thesubject matter of the background section should not be assumed to have beenpreviously recognized in the prior art. The subject matter in the background sectionmerely represents different approaches, which in and of themselves may alsocorrespond to implementations of the claimed technology.

[0003] Obstructive Sleep Apnea (OSA) is a prevalent sleep disordercharacterized by repetitive obstruction or narrowing of the upper airway duringsleep, leading to reduced or completely halted airflow. The narrowing results in asignificant drop in blood oxygen levels, triggering physiological responses such asmicro-arousals and abrupt shifts in sleep patterns. Common symptoms of OSAinclude excessive daytime sleepiness, morning headaches, mood disturbances,forgetfulness, and frequent nighttime awakenings. Individuals with risk factors suchas obesity, large tonsils, or hormonal imbalances are more susceptible to developingOSA. If left untreated, OSA can lead to severe complications, including heartfailure, atrial fibrillation, other arrhythmias, and non-alcoholic steatohepatitis.

[0004] Polysomnography (PSG) is widely recognized as the gold standard fordetecting OSA, as it provides a comprehensive evaluation of sleep-disorderedbreathing using multiple physiological signals. The PSG procedure involvessimultaneous monitoring of several parameters, including airflow measurementthrough nasal pressure transducers and oronasal thermal sensors, respiratory efforttracking via thoracic and abdominal belts, oxygen saturation monitoring using pulseoximetry, and cardiac activity assessment via electrocardiography (ECG).Additionally, sleep architecture is analyzed using electroencephalography (EEG),electrooculography (EOG) for eye movements, and electromyography (EMG) formuscle tone. While PSG offers a detailed analysis, its extensive monitoringrequirements present several challenges. The PSG procedure is highly intrusive,requiring patients to be attached to multiple sensors, which can cause discomfortand disrupt natural sleep patterns. Discomfort can result in altered sleep behavior,leading to data that may not accurately reflect the sleep state of a patient.

[0005] Moreover, PSG tests are conducted in specialized sleep laboratories,requiring trained professionals to administer and interpret the results. Dependenceon dedicated facilities and professionals makes PSG expensive and limits itsaccessibility. Another major drawback is that PSG is not suitable for continuouslong-term monitoring, as it requires a controlled environment and cannot be easilyadapted for home-based or ambulatory use. This limitation makes it impractical fortracking variations in OSA severity over extended periods.

[0006] Additionally, patients with conditions that necessitate frequentmovement during sleep, such as Overactive Bladder Syndrome or Restless LegSyndrome, may experience additional challenges, as movement can disrupt signalacquisition, leading to incomplete or poor-quality data. The complex setup, highcost, and inconvenience of PSG contribute to a significant proportion ofundiagnosed OSA cases, emphasizing the need for a more accessible and userfriendly diagnostic solution.

[0007] In response to the limitations of PSG, alternative methods leveragingdeep-learning and machine-learning techniques must be explored for OSAdetection. However, a majority of such techniques may rely on multiplephysiological signals, such as nasal airflow, thoracic movement, and oxygensaturation, necessitating the use of multiple sensors. Multiple sensor requirementlimits the practicality of such techniques, as multiple sensors make them less userfriendly and more difficult to integrate into wearable or home-based monitoringsystems. Moreover, accuracy of such techniques remains lower than that of PSGdue to incomplete feature extraction from a single physiological signal.

[0008] One of the most significant limitations of these techniques is theirfailure to properly account for the inherent time delay between an apneic event andthe corresponding changes in PPG signals. Apneic events often inducephysiological changes, such as variations in heart rate and blood oxygen levels, butsuch changes do not manifest instantaneously in PPG signals. Many existingmodels do not consider this delay, leading to poor feature representation,misclassification of apnea events, and reduced overall detection accuracy.

[0009] A critical challenge in PPG-based OSA detection is determining anappropriate time window for analyzing apneic events. Traditionally, fixed-durationwindows are employed, such as 90 seconds or 5 minutes, to segment PPG data foranalysis. However, fixed-window approaches introduce significant issues. Apneicevents vary in duration and often occur at irregular intervals. When fixed timewindows are used, an event may be split between two adjacent windows, causingthe loss of crucial event data. This fragmentation affects the ability of deep-learningmodels to recognize clear patterns related to the onset, duration, and resolution ofapnea events.

[0010] Furthermore, splitting events across multiple windows createsinconsistencies in ground truth labeling, confusing machine-learning models duringtraining and reducing classification accuracy. Another issue with fixed-windowapproaches is their inability to account for the time delay between apnea onset andits corresponding physiological changes in PPG signals. If the chosen windowlength does not align properly with these delayed changes, crucial post-eventpatterns may be missed, further reducing model performance.

[0011] In light of the above-mentioned challenges, there is a long-felt need foran improved approach to OSA detection that ensures accurate event classificationwhile maintaining the feasibility and comfort of long-term monitoring.SUMMARY OF THE INVENTION

[0012] This summary is provided to introduce aspects related to a system anda method of detecting obstructive sleep apnea using photoplethysmography signals,and the aspects are further described below in the detailed description. Thissummary is not intended to identify essential features of the claimed subject matter,nor is it intended for use in determining or limiting the scope of the claimed subjectmatter.

[0013] In an embodiment, a method of detecting obstructive sleep apnea (OSA)using photoplethysmography (PPG) signals is disclosed. The method includesreceiving PPG signals from a wearable device. Further, the method includes denoising the PPG signals using a Butterworth bandpass filter. The method furtherincludes segmenting the PPG signals into one or more windows using an adaptivewindowing technique. Apnea events are centrally positioned within the one or morewindows.

[0014] Furthermore, the method includes extracting temporal and spatialfeatures from the PPG signals in the one or more windows using a machine learning(ML) model, and classifying the one or more windows into an apnea event and anon-apnea event, based on presence of an apnea event in each window. Further, themethod includes providing an output indicating presence of the apnea event withineach window.

[0015] In an aspect, segmenting further includes applying a fixed window sizeof 60 seconds to the one or more windows. Further, segmenting includespositioning an apnea event at center of each window to capture delays in a PPGsignal that happen before and after the apnea event. Furthermore, segmentingincludes adjusting the one or more windows to prevent fragmentation of the apneaevent.

[0016] In another aspect, the ML model includes a Multivariate Long ShortTerm Memory - Fully Convolutional Network (MLSTM-FCN) architecture. TheMLSTM-FCN architecture includes a long short-term memory (LSTM) layer tocapture the temporal features in the PPG signals. Further, the MLSTM-FCNarchitecture includes a fully convolutional network (FCN) branch to extract spatialfeatures using a sequence of convolutional layers. Furthermore, the MLSTM-FCNarchitecture includes one or more squeeze and excitation (SE) blocks to adjustfeature importance. The MLSTM-FCN architecture further includes a globalaverage pooling (GAP) layer to reduce dimensionality while retaining importantPPG signal features.

[0017] In another aspect, classifying further includes concatenating temporalfeatures captured by the LSTM layer and the spatial features extracted by the FCNbranch. Further, classifying includes mapping the temporal features and the spatialfeatures to an apnea classification using a fully connected layer. The fully connectedlayer includes a sigmoid activation function. Further, classifying includes utilizingthe sigmoid activation function to provide the output. The output classifies the oneor more windows into an apnea event and a non-apnea event, based on presence ofan apnea event in each window.

[0018] In yet another aspect, a 4th-order Butterworth bandpass filter is utilizedto de-noise within a frequency range of 0.5-8 Hz.

[0019] In an embodiment, a system to detect OSA using PPG signals isdisclosed. The system includes a processor and a memory. The memory is coupledwith the processor. Further, the memory stores program instructions configured toreceive PPG signals from a wearable device and de-noise the PPG signals by usinga Butterworth bandpass filter. The memory further stores program instructionsconfigured to segment the PPG signals into one or more windows. Apnea eventsare centrally positioned within the one or more windows.

[0020] Further, the memory stores program instructions configured to extracttemporal and spatial features from the PPG signals in the one or more windowsusing an ML model. The memory further stores program instructions configured toclassify the one or more windows into an apnea event and a non-apnea event, basedon presence of an apnea event in each window. Further, the memory stores programinstructions configured to provide an output indicating presence of the apnea eventwithin each window.

[0021] In an aspect, to segment the PPG signals into one or more windows, thememory further stores program instructions configured to apply a fixed windowsize of 60 seconds to the one or more windows. Further, the memory stores programinstructions configured to position an apnea event at center of each window tocapture delays in a PPG signal that happen before and after the apnea event.Furthermore, the memory stores program instructions configured to adjust windowsto prevent fragmentation of the apnea event.

[0022] In another aspect, the ML model includes a Multivariate Long ShortTerm Memory - Fully Convolutional Network (MLSTM-FCN) architecture. TheMLSTM-FCN architecture includes a long short-term memory (LSTM) layer tocapture the temporal features in the PPG signals. Further, the MLSTM-FCNarchitecture includes a fully convolutional network (FCN) branch to extract spatialfeatures using a sequence of convolutional layers. Furthermore, the MLSTM-FCNarchitecture includes one or more squeeze and excitation (SE) blocks to adjustfeature importance. Additionally, the MLSTM-FCN architecture includes a globalaverage pooling (GAP) layer reduces dimensionality while retaining important PPGsignal features.

[0023] In one aspect, to classify the one or more windows into an apnea eventand a non-apnea event, the memory further stores program instructions configuredto concatenate temporal features captured by the LSTM layer and the spatialfeatures extracted by the FCN branch and map the temporal features and the spatialfeatures to an apnea classification using a fully connected layer. The fully connectedlayer includes a sigmoid activation function. The memory further stores programinstructions configured to utilize the sigmoid activation function to provide theoutput. The output classifies the one or more windows into an apnea event and anon-apnea event, based on presence of an apnea event in each window.

[0024] In another aspect, a 4th-order Butterworth bandpass filter is utilized tode-noise within a frequency range of 0.5-8 Hz.

[0025] Other aspects and advantages of the invention will become apparentfrom the following description, taken in conjunction with the accompanyingdrawings, illustrating by way of example the principles of the invention.BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings constitute a part of the description and areused to provide a further understanding of the present disclosure. In the drawings:

[0027] Figure 1 illustrates a working architecture of a system for detectingobstructive sleep apnea (OSA) using photoplethysmography (PPG) signals, inaccordance with an embodiment of the present disclosure;

[0028] Figure 2 illustrates a block diagram of the system for detecting OSAusing PPG signals, in accordance with an embodiment of the present disclosure;

[0029] Figure 3 illustrates an architecture of a machine learning model of thesystem for detecting OSA using PPG signals, in accordance with an embodimentof the present disclosure; and

[0030] Figure 4 illustrates a flow chart of a method of detecting OSA usingPPG signals, in accordance with an embodiment of the present disclosure.

[0031] A more complete understanding of the present disclosure and itsembodiments thereof may be acquired by referring to the following description andthe accompanying drawings.DETAILED DESCRIPTION OF THE INVENTION

[0032] Exemplary embodiments now will be described with reference to theaccompanying drawings. The disclosure may, however, be embodied in manydifferent forms and should not be construed as limited to the embodiments set forthherein; rather, these embodiments are provided so that this disclosure will bethorough and complete, and will fully convey its scope to those skilled in the art.The terminology used in the detailed description of the particular exemplaryembodiments illustrated in the accompanying drawings is not intended to belimiting. In the drawings, like numbers refer to like elements.

[0033] It is to be noted, however, that the reference numerals used hereinillustrate only typical embodiments of the present subject matter, and are therefore,not to be considered for limiting its scope, for the subject matter may admit to otherequally effective embodiments.

[0034] The detailed description includes specific details for the purpose ofproviding a thorough understanding of the present invention. However, it will beapparent to those skilled in the art that the present invention may be practicedwithout these specific details.

[0035] The present invention aims to provide a solution for the abovementioned challenges by introducing a deep-learning-based method for detectingObstructive Sleep Apnea (OSA) using a single-channel Photoplethysmography(PPG) signal. Unlike traditional approaches that require multiple physiologicalsensors, this invention leverages PPG data, which is easily accessible throughwearable devices, to provide an accurate and non-intrusive method for apneadetection.

[0036] Additionally, the invention incorporates an adaptive windowingstrategy designed to preserve integrity of apnea events within a window for analysiswhile ensuring inclusion of sufficient pre-event and post-event data. Thewindowing strategy eliminates event fragmentation, enhances accuracy of groundtruth labels, and minimizes model misclassification. The present invention providesa practical and cost-effective alternative to traditional polysomnography, enablingcontinuous long-term monitoring in home environments without the need forspecialized sleep laboratories or trained personnel. By improving the feasibility andaccessibility of OSA detection, the invention has the potential to address thesignificant percentage of undiagnosed cases and contribute to better managementof sleep-disordered breathing conditions.

[0037] Figure 1 illustrates a working architecture of a system 100 for detectingobstructive sleep apnea (OSA) using photoplethysmography (PPG) signals, inaccordance with an embodiment of the present disclosure. The system 100 may beconnected to a wearable device 110 (labelled as PPG Sensor in Figure 1) via acommunication network 120. The wearable device 110 may be configured tomonitor physiological signals, particularly PPG signals of a user, for the detectionof OSA. Further, the wearable device 110 may transfer the PPG signals to thesystem 100, via the communication network 120, for detecting OSA.

[0038] The wearable device 110 may include one of, a wristband, a fingertipsensor, a smartwatch and a ring-type sensor. The wearable device 110 may beconfigured to ensure user comfort during sleep. Further, the wearable device 110may include an optical sensor (PPG sensor) that emits and detects light signals tomeasure changes in blood volume within microvascular tissue, providing a realtime PPG signal. The PPG sensor operates by emitting light into skin of the userand measuring variations in light absorption caused by pulsatile blood flow. Thevariations correspond to heart rate and blood oxygen levels of the user, which arecritical indicators for detecting apnea episodes.

[0039] In an embodiment, the wearable device 110 may capture PPG signalsat a predefined sampling rate, ensuring that even minor variations in pulsewaveform morphology associated with apneic events are accurately recorded.Additionally, the wearable device 110 may include sensors, including but notlimited to an accelerometer and a gyroscope, to track body movements and detectpotential artifacts caused by excessive motion. In an embodiment, the wearabledevice 110 may enhance the reliability of the PPG signals by enabling signalcorrection and noise reduction techniques.

[0040] In an embodiment, to ensure optimal functionality, the wearable device110 may include a processor capable of executing basic preprocessing tasks, suchas filtering raw PPG signals to remove noise and artifacts. In an embodiment,filtering may be executed using a Butterworth bandpass filter implemented in thewearable device 110. The Butterworth bandpass filter may be an Analog or Digitalfilter. In an implementation, the Butterworth bandpass filter may be located withinsignal processing circuits of the wearable device. In one embodiment, theButterworth bandpass filter may be implemented using resistors, capacitors, andoperational amplifiers (op-amps) in low-pass, high-pass, band-pass, or band-stopconfigurations.

[0041] Further, the wearable device 110 may include a memory for temporarystorage of data related to PPG signals before transmission. Additionally, thewearable device 110 may include a wireless communication module (e.g.,Bluetooth, Wi-Fi) and a battery unit.

[0042] The communication network 120 may utilize network components toestablish a connection between the system 100 and the wearable device 110. Thenetwork components may include hubs, switches, routers, bridges, and repeaters.The routers may be of different types, such as Provide Edge (PE) routers, CustomerEdge (CE) routers, and intermediate routers.

[0043] The communication network 120 may be a wired and / or a wirelessnetwork. The communication network 120 may be implemented usingcommunication techniques such as Visible Light Communication (VLC),Worldwide Interoperability for Microwave Access (WiMAX), Long TermEvolution (LTE), Wireless Local Area Network (WLAN), Infrared (IR)communication, Public Switched Telephone Network (PSTN), Radio waves, andother communication techniques known in the art.

[0044] The system 100 may be a remote processing unit for receiving, storing,analyzing, and processing PPG signals transmitted from the wearable device 110using edge computing. In one embodiment, the system 100 may be implemented asa standalone physical computing device, such as a server, or as a cloud-based serverinfrastructure, enabling scalable and distributed data processing. The system 100may utilize an adaptive windowing technique to ensure that each apnea event isfully contained within a single analysis window while capturing pre-event and postevent variations in the PPG signal. Traditional fixed-duration windowingapproaches often result in fragmented apnea events across multiple windows,leading to misclassification and reduced model performance.

[0045] Further, the system 100 may be configured to execute a machinelearning (ML) model trained to identify patterns associated with sleep apnea eventsbased on variations in the PPG signal. Furthermore, the system 100 provides abinary output to a user device 130, indicating presence of an apnea event in ananalysis window.

[0046] The user device 130 includes but is not limited to a mobile phone, alocal computer, a tablet, a laptop, a smart watch, and a smart ring. Further, the userdevice 130 may provide a user interface on a display, which can be accessed by auser 140. The user 140 may include but is not limited to healthcare professionals,researchers and the user wearing the wearable device 110. The user device 130 mayinclude real-time alerts, historical trend analysis, and data export capabilities forfurther clinical assessment, based on the output provided by the system 100.

[0047] Figure 2 illustrates a block diagram of the system 100 for detectingOSA using PPG signals, in accordance with an embodiment of the presentdisclosure. The system 100 may be a remote processing unit for receiving, storing,analyzing, and processing PPG signals transmitted from the wearable device 110(as shown in Figure 1) using edge computing. In one embodiment, the system 100may be implemented as a standalone physical computing device, such as a server,or as a cloud-based server infrastructure, enabling scalable and distributed dataprocessing.

[0048] Further, the system 100 may include one or more networkinterfaces 102 (e.g., wired, wireless, etc.), at least one processor 104, anda memory 106. The one or more network interfaces 102, the at least oneprocessor 104, and the memory 106 may be interconnected by a system bus and apower supply (not shown in the Figure 2). Further, the processor 104 iscommunicatively coupled with the memory 106.

[0049] The one or more network interfaces 102 may be used to provide inputto or fetch output from, the system 100. The one or more network interfaces 108may be implemented as a Command Line Interface (CLI) or a Graphical UserInterface (GUI). Further, Application Programming Interfaces (APIs) may also beused for remotely interacting with edge systems and cloud servers.

[0050] The processor 104 may include one or more general purpose processors(e.g., INTEL or Advanced Micro Devices (AMD) microprocessors) and / or oneor more special purpose processors (e.g., digital signal processors or XilinxSystem On Chip (SOC) Field Programmable Gate Array (FPGA) processor),MIPS / ARM-class processor, a microprocessor, a digital signal processor, anapplication specific integrated circuit, a microcontroller, a state machine, or anytype of programmable logic array.

[0051] The memory 106 may include, but is not limited to, non-transitorymachine-readable storage devices such as hard drives, magnetic tape, floppydiskettes, optical disks, Compact Disc Read-Only Memories (CD-ROMs), andmagneto-optical disks, semiconductor memories, such as ROMs, Random AccessMemories (RAMs), Programmable Read-Only Memories (PROMs), ErasablePROMs (EPROMs), Electrically Erasable PROMs (EEPROMs), flash memory,magnetic or optical cards, or other type of media / machine-readable mediumsuitable for storing electronic instructions.

[0052] The memory 106 may include a plurality of storage locations that areaddressable by the processor 104 and the network interfaces 102 for storingsoftware programs and other necessary information (program instructions andmachine learning model 160) associated with the embodiments described herein.The processor 104 detects OSA by executing program instructions stored in thememory 106. The processor 102 may comprise hardware elements or hardwarelogic adapted to execute the software programs and manipulate data structures.

[0053] It will be apparent to those skilled in the art that other processor andmemory types, including various computer-readable media, may be used to storeand execute program instructions pertaining to the techniques described herein.Also, while the description illustrates various processes, it is expresslycontemplated that various processes may be embodied as modules configured tooperate in accordance with the techniques herein (e.g., according to thefunctionality of a similar process). Further, while the processes have been shownseparately, those skilled in the art will appreciate that processes may be routines ormodules within other processes.

[0054] The processor 104 of the system 100 is configured to receive PPGsignals from the wearable device 110, via the communication network 120 (asshown in Figure 1). Further, the processor 104 utilizes the one or more networkinterfaces 102 of the system 100 to receive the PPG signals from the communicationnetwork 120.

[0055] In one embodiment, the processor 104 may be configured to de-noisethe PPG signals by using a Butterworth bandpass filter to enhance signal qualityand improve the accuracy of apnea detection. The Butterworth bandpass filter maybe a 4th-order Butterworth bandpass filter configured to de-noise within apredefined frequency range. Furthermore, the Butterworth bandpass filter may beat least one of, a hardware based Analog Butterworth filter, a software-based DigitalButterworth filter.

[0056] The predefined frequency range may be 0.5-8 Hz. The frequency rangeof 0.5-8 Hz is selected based on spectral characteristics of PPG signals related tosleep apnea detection. The lower cut-off frequency of 0.5 Hz eliminates lowfrequency drift and baseline wandering, which can result from sensor movement,respiration, or slow variations in blood volume. The upper cut-off frequency of 8Hz removes high-frequency noise, such as power line interference, muscle artifacts,and random electronic noise from the sensor. By restricting the PPG signal to thisspecific range, the filter may ensure that essential cardiovascular and respiratorycomponents relevant for apnea detection are retained while non-informative ordisruptive signal components are suppressed.

[0057] In an embodiment, the processor 104 may be configured to adaptivelyadjust filtering process based on signal quality metrics. For instance, if excessivemotion artifacts or poor-quality signals are detected, the processor 104 may applyadditional filtering techniques or request re-acquisition of the PPG signals tomaintain integrity of apnea detection.

[0058] In an embodiment, the filtering process ensures that only high-qualitysignals contribute to apnea detection by selecting minimum 6 hours of recordeddata to capture sufficient sleep patterns, and signal quality rating of at least 95%(based on 'qupleth5' parameter) to reduce noise interference, and an 'outstanding'overall rating in the 'overall5' quality metric to eliminate unreliable recordings.

[0059] Further, the processor 104 may be configured to segment the PPGsignals into one or more windows using an adaptive windowing technique tofacilitate accurate apnea detection while preserving temporal relationship betweenapnea events and corresponding changes in the PPG signal. The adaptivewindowing technique ensures that apnea events are centrally positioned within eachwindow.

[0060] Conventional windowing techniques use fixed segmentation intervals,where the PPG signal is divided into consecutive windows without regard to eventboundaries. An apnea event may be split between two consecutive windows,leading to ambiguity in classification and increased confusion for a machinelearning model. Further, changes in PPG signals due to an apnea event do not occurinstantaneously but rather exhibit a delay. If an apnea event is positioned towardsthe end of a window in a conventional approach, corresponding delayed PPGvariations may appear in next window, resulting in a misalignment between inputdata and ground-truth labels.

[0061] To address the above-mentioned challenges of conventionalapproaches, the processor 104 implements the adaptive windowing techniqueincorporating 60-second windows, ensuring that every apnea event is centrallylocated within its respective window. The rationale behind the 60-seconds durationis based on a statistical analysis of apnea events, which revealed that 41.7% of apneaevents and 10% of hypopnea events last longer than 30 seconds, with a mean apneaduration of 21.8 seconds. A 60-second window may provide a sufficient temporalbuffer before and after the apnea event, allowing the system 100 to capture theapnea event and delayed physiological changes in the PPG signal.

[0062] The processor 104 may adjust window boundaries to ensure completeapnea events remain within a single segment / window. If an apnea event of length xseconds is detected, the processor 104 calculates a pre-event and post-event bufferof y=(60-x) / 2 seconds to maintain symmetrical padding around the apnea event.The adaptive windowing technique prevents fragmentation of apnea events acrossmultiple windows and ensures that all relevant PPG variations associated withapnea are contained within the same segment / window.

[0063] By centralizing apnea events within windows, the system 100 mayimprove reliability of ground-truth labels, reducing risk of misclassification andimproving accuracy of machine-learning-based apnea detection. The adaptivewindowing technique ensures that normal windows contain only normal breathingpatterns, while apnea windows consistently feature a clearly defined apneic eventat their center, creating a well-structured dataset for training and inference ofmachine learning models.

[0064] The processor 104 may execute a machine learning (ML) model 160stored in the memory 106 to detect OSA in the one or more windows. The MLmodel 160 may be a Multivariate Long Short-Term Memory - Fully ConvolutionalNetwork (MLSTM-FCN) model 160, as illustrated in Figure 3.

[0065] In an embodiment, the ML model 160 may be trained and validatedusing Multi-Ethnic Study of Atherosclerosis (MESA) dataset. The MESA datasetcomprises PPG recordings from about 2056 subjects of diverse ethnic backgrounds,including Black, White, Hispanic, and Chinese-American individuals, agedbetween 45-84 years. To ensure the quality of data used for training, a rigorous datapre-processing pipeline may be implemented. The PPG signals from the MESAdataset may be merged with their corresponding annotation files, after which thesignals may be labeled and categorized into three classes: obstructive apnea,hypopnea, and normal. However, for the present invention, obstructive apnea andhypopnea may be treated as a single class, referred to as apnea.

[0066] The MLSTM-FCN framework is specifically designed to leveragetemporal dependencies and local spatial patterns in PPG signals, making it accuratefor apnea detection. The MLSTM-FCN model 160 employs a dual-brancharchitecture that includes an LSTM layer 162 to extract temporal features from thePPG signals. The LSTM layer 162 may identify long-term variations in the PPGsignal due to apnea-related disturbances. In an embodiment, the LSTM layer 162may be configured with 8 units, captures dependencies across time, ensuring thateven delayed responses in the PPG signal are recognized.

[0067] Further, the LSTM layer 162 may include an attention mechanism toassign higher importance to critical segments of the PPG signal, allowing the MLmodel 160 to focus on relevant patterns related to OSA events. The attentionmechanism may function by analyzing sequential data and identifying portions ofthe PPG signal that exhibit patterns indicative of apnea events. The attentionmechanism ensures that the ML model 160 focuses on most relevant time intervals,rather than treating all input data equally. Given that apneic events often result indelayed changes in PPG signals, the attention mechanism improves ability of theML model 160 to distinguish between normal and abnormal patterns by weightingcritical segments more heavily. By doing so, the attention mechanism enhances thelearning process of the MLSTM-FCN architecture, allowing it to capture bothshort-term and long-term dependencies more effectively.

[0068] Further, the MLSTM-FCN model 160 employs a 1D fullyconvolutional network branch 164 (FCN) for local spatial feature extraction. In anembodiment, the FCN 164 includes a first convolutional layer with 128 filters anda kernel size of 8, a second convolutional layer with 256 filters and a kernel size of5, and a third convolutional layer with 128 filters and a kernel size of 3.

[0069] To further refine feature selection, squeeze-and-excitation (SE) blocks166 are incorporated after each convolutional layer 164. The SE blocks 166reweight feature maps and adjusts feature importance, emphasizing mostinformative aspects of the PPG signal while suppressing less relevant noise.Following convolution, the ML model 160 may apply a global average poolinglayer 168 (GAP), which reduces feature dimensionality while retaining crucialapnea-related information to prevent overfitting.

[0070] Outputs of both the LSTM layer 162 and FCN 164 branches areconcatenated to form a comprehensive feature representation, combining long-termdependencies with localized spatial signal variations. Further, a fully connectedoutput layer 170 with a sigmoid activation function predicts and provides an outputindicating a probability of an apnea event occurring within the one or morewindows and classifies each window based on the probability, as an apnea event(A) or a normal event (N), as shown in Figure 3. In an embodiment, the ML model160 achieves a test accuracy of 93.44%, outperforming traditional methods by11.3%. The ML model 160 further records a precision of 0.94, recall of 0.91,specificity of 0.95, and an F1-score of 0.93.

[0071] The processor 104 may be further configured to provide the outputindicating presence of the apnea event within each classified window to the userdevice 130, allowing the user 140 to remotely track sleep patterns of patient. Thereal-time transmission enables early intervention, reducing health risks associatedwith undiagnosed or untreated sleep apnea.

[0072] In an embodiment, the system 100 may incorporate an adaptivethresholding mechanism. The adaptive thresholding mechanism may adjust adetection threshold based on historical sleep data of the user, improvingpersonalized accuracy over time. By continuously learning from past apneapatterns, the system 100 minimizes false positives and false negatives, making ithighly reliable for long-term use.

[0073] Figure 4 illustrates a flow chart of a method 400 of detecting OSAusing PPG signals, in accordance with an embodiment of the present invention. Inthis regard, each block may represent a module, segment, or portion of code, whichcomprises one or more executable instructions for implementing the specifiedlogical function(s). It should also be noted that in some alternative implementations,the functions noted in the blocks may occur out of the order noted in the drawings.

[0074] For example, two blocks shown in succession in Figure 4 may in factbe executed substantially concurrently or the blocks may sometimes be executed inthe reverse order, depending upon the functionality involved. Any processdescriptions or blocks in flow charts should be understood as representing modules,segments, or portions of code which include one or more executable instructionsfor implementing specific logical functions or steps in the process, and alternateimplementations are included within the scope of the example embodiments inwhich functions may be executed out of order from that shown or discussed,including substantially concurrently or in reverse order, depending on thefunctionality involved. In addition, the process descriptions or blocks in flow chartsshould be understood as representing decisions made by a hardware structure suchas a state machine.

[0075] The order in which the method is described is not intended to beconstrued as a limitation, and any number of the described method blocks can becombined in any order to implement the method. Additionally, individual blocksmay be deleted from the methods without departing from the scope of the subjectmatter described herein.

[0076] Furthermore, the method can be implemented in any suitable hardware,software, firmware, or combination thereof. Furthermore, the above-mentionedmethods may be implemented in suitable hardware, computer-readable instructions,or a combination thereof. The steps of such methods may be performed by either asystem under the instruction of machine-executable instructions stored on a nontransitory computer-readable medium or by dedicated hardware circuits,microcontrollers, or logic circuits. The method may include the following steps.

[0077] At step 402, PPG signals may be received from a wearable device at asystem for detecting OSA. The wearable device may include one of, a wristband, afingertip sensor, a smartwatch and a ring-type sensor. The wearable device may beconfigured to ensure user comfort during sleep.

[0078] Further, the wearable device may include an optical sensor (PPGsensor) that emits and detects light signals to measure changes in blood volumewithin microvascular tissue, providing a real-time PPG signal. The PPG sensoroperates by emitting light into skin of a user and measuring variations in lightabsorption caused by pulsatile blood flow. The variations correspond to heart rateand blood oxygen levels of the user, which are critical indicators for detecting apneaepisodes.

[0079] In an embodiment, the wearable device may capture PPG signals at apredefined sampling rate, ensuring that even minor variations in pulse waveformmorphology associated with apneic events are accurately recorded. Additionally,the wearable device may include sensors, including but not limited to anaccelerometer and a gyroscope, to track body movements and detect potentialartifacts caused by excessive motion. In an embodiment, the wearable device mayenhance the reliability of the PPG signals by enabling signal correction and noisereduction techniques.

[0080] Further, the wearable device may include a memory for temporarystorage of data related to PPG signals before transmission. Additionally, thewearable device may include a wireless communication module (e.g., Bluetooth,Wi-Fi) and a battery unit. Further, the wearable device may transfer the PPG signalsto the system, via the communication network, for detecting OSA.

[0081] The system may be a remote processing unit for receiving, storing,analyzing, and processing PPG signals transmitted from the wearable device usingedge computing. In one embodiment, the system may be implemented as astandalone physical computing device, such as a server, or as a cloud-based serverinfrastructure, enabling scalable and distributed data processing.

[0082] At step 404, the PPG signals may be de-noised using a Butterworthbandpass filter. In an embodiment, to ensure optimal functionality, the wearabledevice may include a processor capable of executing basic preprocessing tasks,such as filtering raw PPG signals to remove noise and artifacts. In an embodiment,filtering may be executed using a Butterworth bandpass filter implemented in thewearable device.

[0083] The Butterworth bandpass filter may be an Analog or Digital filter. Inan implementation, the Butterworth bandpass filter may be located within signalprocessing circuits of the wearable device. In one embodiment, the Butterworthbandpass filter may be implemented using resistors, capacitors, and operationalamplifiers (op-amps) in low-pass, high-pass, band-pass, or band-stopconfigurations.

[0084] At step 406, the PPG signals may be segmented into one or morewindows using an adaptive windowing technique to facilitate accurate apneadetection while preserving temporal relationship between apnea events andcorresponding changes in the PPG signal. The adaptive windowing techniqueensures that apnea events are centrally positioned within each window.

[0085] Conventional windowing techniques use fixed segmentation intervals,where the PPG signal is divided into consecutive windows without regard to eventboundaries. An apnea event may be split between two consecutive windows,leading to ambiguity in classification and increased confusion for a machinelearning model. Further, changes in PPG signals due to an apnea event do not occurinstantaneously but rather exhibit a delay. If an apnea event is positioned towardsthe end of a window in a conventional approach, corresponding delayed PPGvariations may appear in next window, resulting in a misalignment between inputdata and ground-truth labels.

[0086] To address the above-mentioned challenges of conventional windowingtechniques, the system implements the adaptive windowing techniqueincorporating 60-second windows, ensuring that every apnea event is centrallylocated within its respective window. The rationale behind the 60-seconds durationis based on a statistical analysis of apnea events, which revealed that 41.7% of apneaevents and 10% of hypopnea events last longer than 30 seconds, with a mean apneaduration of 21.8 seconds. A 60-second window may provide a sufficient temporalbuffer before and after the apnea event, allowing the system to capture the apneaevent and delayed physiological changes in the PPG signal.

[0087] The system may adjust window boundaries to ensure complete apneaevents remain within a single segment / window. If an apnea event of length xseconds is detected, the system calculates a pre-event and post-event buffer ofy=(60-x) / 2 seconds to maintain symmetrical padding around the apnea event. Theadaptive windowing technique prevents fragmentation of apnea events acrossmultiple windows and ensures that all relevant PPG variations associated withapnea are contained within the same segment / window.

[0088] At step 408, temporal and spatial features from the PPG signals in theone or more windows may be extracted using a machine learning (ML) model. TheML model may be a Multivariate Long Short-Term Memory - Fully ConvolutionalNetwork (MLSTM-FCN) model.

[0089] In an embodiment, the ML model may be trained and validated usingMulti-Ethnic Study of Atherosclerosis (MESA) dataset. The MESA datasetcomprises PPG recordings from about 2056 subjects of diverse ethnic backgrounds,including Black, White, Hispanic, and Chinese-American individuals, agedbetween 45-84 years. To ensure the quality of data used for training, a rigorous datapre-processing pipeline may be implemented. The PPG signals from the MESAdataset may be merged with their corresponding annotation files, after which thesignals may be labeled and categorized into three classes: obstructive apnea,hypopnea, and normal. However, for the present invention, obstructive apnea andhypopnea may be treated as a single class, referred to as apnea.

[0090] The MLSTM-FCN framework is specifically designed to leveragetemporal dependencies and local spatial patterns in PPG signals, making it accuratefor apnea detection. The MLSTM-FCN model employs a dual-branch architecturethat includes an LSTM layer to extract temporal features from the PPG signals. TheLSTM layer may identify long-term variations in the PPG signal due to apnearelated disturbances.

[0091] In an embodiment, the LSTM layer may be configured with 8 units,captures dependencies across time, ensuring that even delayed responses in the PPGsignal are recognized. Further, the LSTM layer may include an attention mechanismto assign higher importance to critical segments of the PPG signal, allowing the MLmodel to focus on relevant patterns related to OSA events. The attention mechanismmay function by analyzing sequential data and identifying portions of the PPGsignal that exhibit patterns indicative of apnea events. The attention mechanismensures that the ML model focuses on most relevant time intervals, rather thantreating all input data equally. Given that apneic events often result in delayedchanges in PPG signals, the attention mechanism improves ability of the ML modelto distinguish between normal and abnormal patterns by weighting criticalsegments more heavily. By doing so, the attention mechanism enhances the learningprocess of the MLSTM-FCN architecture, allowing it to capture both short-termand long-term dependencies more effectively.

[0092] Further, the MLSTM-FCN model may employ a 1D fully convolutionalnetwork branch (FCN) for local spatial feature extraction. In an embodiment, theFCN includes a first convolutional layer with 128 filters and a kernel size of 8, asecond convolutional layer with 256 filters and a kernel size of 5, and a thirdconvolutional layer with 128 filters and a kernel size of 3.

[0093] To further refine feature selection, squeeze-and-excitation (SE) blocksmay be incorporated after each convolutional layer. The SE blocks reweight featuremaps and adjusts feature importance, emphasizing most informative aspects of thePPG signal while suppressing less relevant noise. Following convolution, the MLmodel may be configured to apply a global average pooling layer (GAP), whichreduces feature dimensionality while retaining crucial apnea-related information toprevent overfitting.

[0094] At step 410, the one or more windows are classified into an apnea eventand a non-apnea event, based on presence of an apnea event in each window. Toclassify the windows, outputs of both the LSTM layer and FCN branches areconcatenated to form a comprehensive feature representation, combining long-termdependencies with localized spatial signal variations. Further, a fully connectedoutput layer with a sigmoid activation function predicts and provides an outputindicating a probability of an apnea event occurring within the one or morewindows. The output classifies each window based on the probability, as an apneaevent or a normal event.

[0095] At step 412, the system may provide the output indicating presence ofthe apnea event within each classified window to the user device, allowing the userto remotely track sleep patterns of patient. The real-time transmission enables earlyintervention, reducing health risks associated with undiagnosed or untreated sleepapnea.

[0096] The user device includes but is not limited to a mobile phone, a localcomputer, a tablet, a laptop, a smart watch, and a smart ring. Further, the user devicemay provide a user interface on a display, which can be accessed by a user. Theuser may include but is not limited to healthcare professionals, researchers and theuser wearing the wearable device. The user device may include real-time alerts,historical trend analysis, and data export capabilities for further clinical assessment,based on the output provided by the system.Technical Advancement and Economic Significance

[0097] The system and the method disclosed in the present invention ofdetecting obstructive sleep apnea (OSA) using photoplethysmography (PPG)signals may have the following advantages over conventional art:- Accurate OSA detection is achieved using a Multivariate Long Short-TermMemory - Fully Convolutional Network (MLSTM-FCN) model, improvingclassification performance compared to traditional methods.- Efficient de-noising of PPG signals is performed using a 4th-order Butterworthbandpass filter, ensuring high-quality input data for OSA detection andreduction in PPG signal processing overhead.- Adaptive windowing technique ensures that apnea events remain centrallypositioned within windows, reducing misclassification errors.- Parallel feature extraction using LSTM and convolutional layers enables themodel to capture both temporal and spatial patterns in PPG signals, enhancingdetection accuracy.- Reduced dependency on specialized sleep clinics lowers patient costs andenhances accessibility, making OSA diagnosis widely accessible.- Cost-effective solution is provided by eliminating the need for complexpolysomnography (PSG) setups, making OSA detection more accessible.

[0098] The specification may refer to "an", "another", "one" or "some"embodiment(s) in several locations.

[0099] This does not necessarily imply that each such reference is to the sameembodiment(s), or that the feature only applies to a single embodiment. Singlefeatures of different embodiments may also be combined to provide otherembodiments.

[0100] The terms "or" and "and / or" as used herein are to be interpreted asinclusive or meaning any one or any combination. Therefore, "A, B or C" or "A, Band / or C" mean "any of the following: A; B; C; A and B; A and C; B and C; A, Band C." An exception to this definition will occur only when a combination ofelements, functions, steps or acts are in some way inherently mutually exclusive.

[0101] As used herein, the singular forms "a", "an" and "the" are intended toinclude the plural forms as well, unless expressly stated otherwise. It will be furtherunderstood that the terms "includes", "comprises", "including" and / or"comprising" when used in this specification, specify the presence of statedfeatures, integers, steps, operations, elements, and / or components, but do notpreclude the presence or addition of one or more other features, integers, steps,operations, elements, components, and / or groups thereof. It will be understood thatwhen an element is referred to as being "connected" or "coupled" to anotherelement, it can be directly connected or coupled to the other element or interveningelements may be present. Furthermore, "connected" or "coupled" as used hereinmay include operatively connected or coupled. As used herein, the term "and / or"includes any and all combinations and arrangements of one or more of theassociated listed items.

[0102] Unless otherwise defined, all terms (including technical and scientificterms) used herein have the same meaning as commonly understood by one ofordinary skills in the art to which this disclosure pertains. It will be furtherunderstood that terms, such as those defined in commonly used dictionaries, shouldbe interpreted as having a meaning that is consistent with their meaning in thecontext of the relevant art and will not be interpreted in an idealized or overly formalsense unless expressly so defined herein.

[0103] Although implementations of a system and a method of detectingobstructive sleep apnea using photoplethysmography signals have been describedin language specific to structural features and / or methods, it is to be understood thatthe appended claims are not necessarily limited to the specific features or methodsdescribed. Rather, the specific features and methods are disclosed as examples ofimplementations of a system and a method of detecting obstructive sleep apneausing photoplethysmography signals.

[0104] The invention has been described above with reference to numerousembodiments and specific examples. Many variations will suggest themselves tothose skilled in this art in light of the above-detailed description. All such obviousvariations are within the full intended scope of the appended claims.

Claims

1. A method (400) of detecting obstructive sleep apnea (OSA) using photoplethysmography (PPG) signals, comprising: receiving PPG signals from a wearable device; de-noising the PPG signals by using a Butterworth bandpass filter; segmenting the PPG signals into one or more windows using an adaptive windowing technique, wherein apnea events are centrally positioned within the one or more windows; extracting temporal and spatial features from the PPG signals in the one or more windows using a machine learning (ML) model; classifying the one or more windows into an apnea event and a non-apnea event, based on presence of an apnea event in each window; and providing an output indicating presence of the apnea event within each window.

2. The method (400) as claimed in claim 1, wherein segmenting further comprising: applying a fixed window size of 60 seconds to the one or more windows; positioning an apnea event at center of each window to capture delays in a PPG signal that happen before and after the apnea event; and adjusting the one or more windows to prevent fragmentation of the apnea event.

3. The method (400) as claimed in claim 1, wherein the ML model comprises a Multivariate Long Short-Term Memory - Fully Convolutional Network (MLSTM-FCN) architecture, and wherein: a long short-term memory (LSTM) layer captures the temporal features in the PPG signals; a fully convolutional network (FCN) branch extracts spatial features using a sequence of convolutional layers; one or more squeeze and excitation (SE) blocks adjust feature importance; and a global average pooling (GAP) layer reduces dimensionality while retaining important PPG signal features.

4. The method (400) as claimed in claim 1, wherein classifying further comprising: concatenating temporal features captured by the LSTM layer and the spatial features extracted by the FCN branch; mapping the temporal features and the spatial features to an apnea classification using a fully connected layer, wherein the fully connected layer comprises a sigmoid activation function; utilizing the sigmoid activation function to provide the output classifying the one or more windows into an apnea event and a non-apnea event, based on presence of an apnea event in each window.

5. The method as claimed in claim 1, wherein a 4th-order Butterworth bandpass filter is utilized to de-noise within a frequency range of 0.5-8 Hz.

6. A system (100) to detect obstructive sleep apnea (OSA) using photoplethysmography (PPG) signals, comprises: a processor (104); and a memory (106) coupled with the processor (104), wherein the memory (106) stores a machine learning (ML) model (160) and program instructions configured to: receive PPG signals from a wearable device (110); de-noise the PPG signals by using a Butterworth bandpass filter; segment the PPG signals into one or more windows, wherein apnea events are centrally positioned within the one or more windows; extract temporal and spatial features from the PPG signals in the one or more windows using the machine learning (ML) model (160); classify the one or more windows into an apnea event and a nonapnea event, based on presence of an apnea event in each window; and provide an output indicating presence of the apnea event within each window.

7. The system (100) as claimed in claim 6, wherein to segment the PPG signals into one or more windows, the memory (106) further stores program instructions configured to: apply a fixed window size of 60 seconds to the one or more windows; position an apnea event at center of each window to capture delays in a PPG signal that happen before and after the apnea event; and adjust windows to prevent fragmentation of the apnea event.

8. The system (100) as claimed in claim 6, wherein the ML model (160) comprises a Multivariate Long Short-Term Memory - Fully Convolutional Network (MLSTM-FCN) architecture, and wherein: a long short-term memory (LSTM) layer captures the temporal features in the PPG signals; a fully convolutional network (FCN) branch extracts spatial features using a sequence of convolutional layers; one or more squeeze and excitation (SE) blocks adjust feature importance; and a global average pooling (GAP) layer reduces dimensionality while retaining important PPG signal features.

9. The system (100) as claimed in claim 6, wherein to classify the one or more windows into an apnea event and a non-apnea event, the memory (106) further stores program instructions configured to: concatenate temporal features captured by the LSTM layer and the spatial features extracted by the FCN branch; map the temporal features and the spatial features to an apnea classification using a fully connected layer, wherein the fully connected layer comprises a sigmoid activation function; utilize the sigmoid activation function to provide the output classifying the one or more windows into an apnea event and a non-apnea event, based on presence of an apnea event in each window.

10. The system (100) as claimed in claim 6, wherein a 4th-order Butterworth bandpass filter is utilized to de-noise within a frequency range of 0.5-8 Hz.