WEARABLE DEVICE FOR REAL-TIME DETECTION OF OBSTRUCTIVE SLEEP APNEA USING DEEP LEARNING

Abstract

The present invention relates to a wearable device for real-time detection of Obstructive Sleep Apnea (OSA) using deep learning. The wearable device comprises multiple sensors like ECG, pulse oximetry for blood oxygen monitoring, respiratory effort detection, an accelerometer for body position, a skin temperature sensor; a microphone for sound analysis; MobileNet V1 combined with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks to analyze ECG and other sensor data; secure data transmission protocols to ensure sensitive ECG and other physiological data are encrypted; and a Raspberry Pi Pico microcontroller for continuous operation on minimal power. The wearable device designed for home use, this device provides a non-invasive, cost-effective alternative to polysomnography, enabling real-time alerts and secure data transmission for enhanced patient privacy. It offers a practical solution for continuous health monitoring, making sleep apnea detection more accessible and comprehensive monitoring.

Specification

Description:Field of the Invention
The present invention relates to the technical field of biomedical devices, more particular to a wearable device for real-time detection of Obstructive Sleep Apnea (OSA) using deep learning.
Background of the Invention
The invention addresses the complexity, cost, and inconvenience of traditional sleep apnea detection methods, specifically Polysomnography (PSG), which requires patients to be wired to multiple sensors and monitored in specialized facilities. This process is not only expensive but also inaccessible for many patients.
The wearable device solves this problem by offering a portable, non-invasive, and cost-effective solution that can be used in home environments. It provides real-time detection of Obstructive Sleep Apnea (OSA) using advanced deep learning models (MobileNet V1, LSTM, and GRU) to analyze ECG signals. The device eliminates the need for complex monitoring equipment, providing a more user-friendly and accurate alternative for early detection and continuous monitoring of sleep apnea episodes. Additionally, it addresses data security by integrating encryption, ensuring patient privacy during the transmission of sensitive medical data.
CN212645929U: RFID label body temperature monitoring system, described to a RFID label body temperature monitoring system belongs to measurement thermometer technical field. The body temperature monitoring system comprises an RFID temperature measurement label, a reader-writer and a background server; the RFID temperature measurement tag, the reader-writer and the background server are sequentially connected; the RFID temperature measurement tag can obtain radio frequency energy from the reader-writer, does not need a power supply, and has a service life as long as 10 years. The reader-writer identifies the temperature measurement module information and transmits the temperature measurement information to the background server. The background server is used for receiving the human body temperature identified by the reader-writer, storing the body temperature, judging the measured temperature and early warning the fever body temperature. Through the utility model discloses heartache can accomplish early discovery, early isolation, early treatment, can the effective control infectious source, will avoid taking ill work, going on class and meeting, has important using value and promotes the meaning.
US20200178842A1: Breath analysis system and methods for asthma, tuberculosis and lung cancer diagnostics and disease management, disclosed a method and system for the detecting of whether a subject has a lung disorder such as asthma, tuberculosis or lung cancer. Monitoring the subject's health and prognosis is also disclosed.
US11663898B2: Remote health monitoring system, A data collection system collects and stores physiological data from an ambulatory patient at a high resolution and/or a high data rate ("more detailed data") and sends a low-resolution and/or downsampled version of the data ("less detailed data") to a remote server via a wireless network. The server automatically analyzes the less detailed data to detect an anomaly, such as an arrhythmia. A two-tiered analysis scheme is used, where the first tier is more sensitive and less specific than the second tier. If the more sensitive analysis detects or suspects the anomaly, the server signals the data collector to send more detailed data that corresponds to a time period associated with the anomaly. The more specific second tier analyses the more detailed data to verify the anomaly. The server may also store the received data and make it available to a user, such as via a graphical or tabular display.
US20190304582A1: Methods and System for Real Time, Cognitive Integration with Clinical Decision Support Systems featuring Interoperable Data Exchange on Cloud-Based and Blockchain Networks, system is deployed as a Software-as-a-Service (SAAS) application; it implements a cloud-based, real-time architecture comprising: (a) an adaptive user interface providing responsive dashboards for real-time data presentation and user interaction, (b) a hub controller for real-time data flow transformation, (c) a data validation engine which incorporates cognitive natural language processing (NLP) to extract structured and unstructured patient record data from the EHR and employs methods that provide an optimally automated input to the CDS, (d) cloud storage of aggregated CDS data and other clinical datasets, and (e) plug-in support for additional cognitive capabilities such as predictive analytics and data mining.
None of the prior art indicated above either alone or in combination with one another disclose what the present invention has disclosed.
Drawings
Fig.1 illustrates the block diagram of the present invention
Fig.2 illustrates the design of the device
Detailed Description of the Invention
The following description includes the preferred best mode of one embodiment of the present invention. It will be clear from this description of the invention that the invention is not limited to these illustrated embodiments but that the invention also includes a variety of modifications and embodiments thereto. Therefore, the present description should be seen as illustrative and not limiting. While the invention is susceptible to various modifications and alternative constructions, it should be understood, that there is no intention to limit the invention to the specific form disclosed, but, on the contrary, the invention is to cover all modifications, alternative constructions, and equivalents falling within the spirit and scope of the invention as defined in the claims.
In any embodiment described herein, the open-ended terms "comprising," "comprises," and the like (which are synonymous with "including," "having" and "characterized by") may be replaced by the respective partially closed phrases "consisting essentially of," consists essentially of," and the like or the respective closed phrases "consisting of," "consists of, the like. As used herein, the singular forms "a", "an", and "the" designate both the singular and the plural, unless expressly stated to designate the singular only.
The wearable device for Obstructive Sleep Apnea (OSA) detection stands out due to its unique combination of advanced sensors, real-time analysis, and user-friendly design. Here are the salient features that make this invention distinctive:
1. Multi-Sensor Integration for Comprehensive Monitoring
• Unlike traditional sleep apnea devices, this wearable integrates multiple sensors-ECG, pulse oximetry, respiratory effort, accelerometer, temperature, and microphone-allowing it to monitor a broader set of physiological parameters that correlate with OSA events.
• This multi-sensor setup provides a holistic view of the patient's sleep health, enhancing the accuracy and reliability of apnea detection.
2. Advanced Deep Learning Models
• The device incorporates MobileNet V1 combined with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks to analyze ECG and other sensor data. This combination is optimized for detecting apnea events with high sensitivity and specificity, providing 90.29% accuracy in distinguishing between apnea and normal conditions.
• This model architecture allows for rapid, low-latency processing, essential for real-time monitoring.
3. Real-Time Detection and Alert System
• Real-time data processing enables immediate detection of apnea events, making it possible to send alerts when episodes occur. This feature is crucial for patients with severe apnea, as timely interventions can reduce health risks.
• By notifying users or healthcare providers as events happen, the device supports proactive management of sleep apnea.
4. Non-Invasive, Home-Based Solution
• Designed for comfortable home use, the device eliminates the need for polysomnography, which traditionally requires patients to undergo complex monitoring in a sleep lab.
• This user-friendly approach encourages compliance, allowing patients to monitor their sleep apnea from the comfort of their home, which also reduces healthcare costs.
5. Data Security with Encryption
• The device features secure data transmission protocols, ensuring that sensitive ECG and other physiological data are encrypted. This focus on data security meets stringent healthcare privacy standards, protecting patients' medical information.
6. Compact, Portable Design with Low-Power Operation
• The wearable is compact and lightweight, optimized for overnight use without disturbing the user's sleep. It is powered by a Raspberry Pi Pico microcontroller, allowing for continuous operation on minimal power.
• This design makes it ideal for long-term monitoring without the need for frequent recharging, enhancing user convenience and device reliability.
7. Scalability and Adaptability for Broader Health Monitoring
• The architecture of the device allows for easy updates and customization, enabling the inclusion of additional health monitoring parameters if needed. This makes the device adaptable for monitoring other conditions, such as heart rate irregularities, respiratory disorders, and general wellness tracking.
• The deep learning algorithms can also be retrained to adapt to other health monitoring applications, making the device highly versatile.
8. Automated Diagnosis and Minimal Need for Human Intervention
• The device autonomously analyzes the collected data, detecting apnea episodes and triggering alerts without manual input. This automation saves time for healthcare professionals, reduces the burden on sleep clinics, and streamlines diagnosis.
9. Insightful Sleep Quality Analysis
• Beyond apnea detection, the device provides insights into overall sleep quality by monitoring body position, movement, and snoring. This allows users to understand the impact of various factors on their sleep, encouraging healthier sleep habits.
10. Cost-Effective Alternative to Polysomnography
• The device offers a more affordable, home-based solution compared to expensive polysomnography tests, democratizing access to sleep apnea monitoring and making it available to a larger population.
• By replacing the need for specialized equipment and medical personnel, it offers a low-cost solution without compromising on accuracy or reliability.
11. Interoperability with Healthcare Systems
• The device can securely upload data to healthcare providers or cloud-based systems, enabling easy integration with existing healthcare infrastructure. This supports remote monitoring and diagnosis, streamlining the communication between patients and providers.
These features collectively make the invention a comprehensive, convenient, and cost-effective solution for sleep apnea monitoring, setting it apart from traditional and existing wearable devices. Its real-time, multi-sensor approach and advanced deep learning integration allow it to deliver high accuracy and ease of use in an accessible, at-home setting.
The process for the real-time detection of Obstructive Sleep Apnea (OSA) using deep learning, comprising the following steps:
1. Data Collection
o The device uses single-lead ECG signals from the PhysioNet Apnea-ECG database, containing recordings with 16-bit resolution and a 100 Hz sampling rate, covering 7-10 hours of sleep per patient.
o Additional sensor data, such as oxygen saturation, respiratory effort, body position, temperature, and sound levels, were collected during prototype testing to simulate real-world conditions for comprehensive apnea detection.
2. Data Preprocessing
o ECG Signal Segmentation: ECG signals were segmented into 60-second intervals to capture meaningful data while reducing computational load.
o Noise Removal: Preprocessing techniques, such as filtering and auto-correlation functions, were applied to remove noisy and distorted segments, ensuring high-quality data.
o Feature Extraction: Relevant features were extracted from each sensor, including heart rate variability from ECG, oxygen saturation levels, respiratory rate, body position, skin temperature, and snoring patterns, to enhance apnea classification accuracy.
3. Model Development
o MobileNet V1: Used as a lightweight Convolutional Neural Network (CNN) model optimized for mobile devices, responsible for capturing spatial features from ECG data.
o MobileNet V1 + LSTM / GRU: Integrated with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks to process sequential dependencies in ECG and sensor data, enhancing classification performance for apnea and non-apnea states.
o Training Process: Models were trained on 35 patient ECG and sensor records using Adam optimization, with a learning rate of 0.001 over 100 epochs. The loss function was categorical cross-entropy, and ReLU activation functions were used for nonlinear feature extraction.
4. Evaluation Metrics
o Performance was assessed using accuracy, sensitivity (recall), specificity, precision, and F1 score to evaluate how well the models distinguish between apnea and normal conditions.
o Each model's performance on apnea classification was compared to traditional detection methods to confirm its superiority in accuracy and reliability.
5. Device Implementation
o After achieving satisfactory model performance, the trained models were deployed on the wearable device powered by a Raspberry Pi Pico microcontroller.
o ECG and other sensor data are captured, processed in real-time by the deep learning models, and transmitted securely via encrypted protocols to a cloud-based platform for further analysis.
6. Security Implementation
o Data encryption mechanisms were incorporated to ensure the secure transmission of physiological data, adhering to healthcare security standards and protecting patient privacy during data transfer.
The performance of the device is as follows:
1. Model Accuracy and Performance
o MobileNet V1 achieved an accuracy of 89.5%.
o MobileNet V1 + LSTM reached 90.0% accuracy.
o MobileNet V1 + GRU obtained the highest accuracy of 90.29%, with sensitivity of 90.01% and specificity of 90.72%.
o The GRU model demonstrated superior performance in detecting apnea events with minimal false positives and negatives, showing a high precision of 94.71% and an F1 score of 92.33%.
2. Wearable Device Functionality
o The device successfully integrated the trained models for real-time signal processing, enabling continuous monitoring throughout the night.
o It accurately captured and classified apnea events, triggering real-time alerts and securely transmitting data to healthcare providers, enabling remote monitoring.
3. Low Latency and Energy Efficiency
o Running on a Raspberry Pi Pico microcontroller, the device demonstrated low latency, processing data in near real-time with efficient power consumption, making it suitable for prolonged, uninterrupted use.
4. User Compliance and Comfort
o The wearable's compact, low-profile design facilitated comfortable overnight use, leading to higher user compliance for regular monitoring.
o Its non-invasive design and simple setup make it accessible to users without medical expertise, promoting consistent use and yielding long-term health insights.
5. Cost-Effectiveness
o The device provided a cost-effective alternative to traditional polysomnography, offering comparable accuracy and convenience while reducing the financial burden on patients and healthcare systems.
6. Broader Health Insights
o Beyond sleep apnea detection, the device provided insights into sleep quality and user habits, supporting health-conscious adjustments and better sleep hygiene.
The advantages of the present device are as follows:
1. Comprehensive Multi-Sensor Monitoring
• Enhanced Accuracy: With the addition of pulse oximetry, respiratory effort, body position, temperature, and sound sensors, the device captures a broader range of physiological indicators, improving detection accuracy.
• Real-time Detection of Apnea Events: Real-time data from multiple sensors allow the device to detect OSA episodes as they occur, enabling timely intervention, which is critical for patients with severe apnea.
2. Non-Invasive and User-Friendly Design
• Home-Based Monitoring: The wearable is designed for comfortable use at home, eliminating the need for complex, invasive procedures like polysomnography, making apnea detection accessible to more users.
• Enhanced Patient Compliance: Lightweight and comfortable, the device is easy to wear regularly, encouraging consistent use and better long-term health monitoring.
3. Technical Innovation in Deep Learning Integration
• Advanced Model Accuracy: The integration of MobileNet V1, LSTM, and GRU models with additional sensor data enhances classification accuracy, achieving sensitivity and specificity that outperform traditional methods.
• Real-time Signal Processing: The device offers low-latency, real-time signal analysis, which is essential for continuous, real-time monitoring during sleep.
4. Portability and Energy Efficiency
• Low-Power Microcontroller: Powered by a Raspberry Pi Pico, the device is optimized for low power consumption, allowing for continuous, overnight use without frequent recharging.
• Compact Design: The compact, portable form factor makes the device convenient for users, enabling hassle-free sleep monitoring.
5. Secure Data Transmission and Privacy
• Data Encryption: The device ensures secure data transmission with built-in encryption, safeguarding sensitive medical information, a crucial factor for healthcare applications.
• Remote Monitoring: Allows secure data transmission to healthcare providers for continuous monitoring, enabling remote diagnosis and management of OSA.
6. Cost-Effective Alternative to Traditional Methods
• Affordable Solution: By replacing polysomnography with a compact, home-based solution, the device significantly reduces costs for both patients and healthcare providers.
• Reduced Healthcare Burden: The automation and portability of the device reduce the need for specialized sleep clinics and staff, making OSA monitoring more widely available and cost-effective.
7. Scalability and Adaptability
• Multi-Condition Monitoring Potential: The device's architecture can be adapted to monitor other health conditions, such as heart rate variability, arrhythmia, or respiratory disorders, broadening its utility beyond sleep apnea detection.
• Customizable and Upgradable: The deep learning algorithms and sensor suite can be updated or customized to accommodate additional physiological parameters or emerging AI advancements.
8. Improved Data for Long-Term Health Tracking
• Continuous Data Collection: By collecting data over extended periods, the device helps create a detailed health profile, valuable for tracking sleep patterns and health trends.
• Sleep Quality Insights: In addition to detecting apnea events, the device provides feedback on sleep quality, encouraging users to make healthier sleep-related choices.
9. Automated Diagnosis and Alert System
• Timely Interventions: The device autonomously analyzes data and generates real-time alerts for detected apnea events, helping users and healthcare providers respond quickly to potentially harmful episodes.
• Reduced Need for Manual Supervision: The automated diagnosis feature reduces reliance on healthcare professionals for initial assessment, streamlining the monitoring process.
10. Faster Detection and Response
• Low Latency Processing: By using optimized deep learning models, the device processes ECG and multi-sensor data with minimal delay, providing prompt alerts compared to slower, conventional sleep studies.
• Adaptable for Mobile and Remote Environments: The device's portability makes it suitable for mobile health monitoring, offering flexibility for users even outside their homes.
11. Enhanced Sleep Analysis with Position and Sound Tracking
• Body Position and Snoring Analysis: With body position and sound sensors, the device identifies sleep positions and snoring patterns linked to apnea, enriching the analysis and providing insights into potential positional therapies.

, Claims:1. A wearable device for real-time detection of Obstructive Sleep Apnea (OSA), comprises of:
• an electrocardiogram (ECG) sensor for capturing ECG signals;
• pulse oximetry for blood oxygen monitoring;
• respiratory effort detection;
• an accelerometer for body position;
• a skin temperature sensor;
• a microphone for sound analysis;
• MobileNet V1 with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks to classify ECG signals into apnea or normal patterns with high accuracy; and
• a Raspberry Pi Pico microcontroller for continuous operation on minimal power.
2. The wearable device for real-time detection of Obstructive Sleep Apnea (OSA) as claimed in the claim 1, wherein device features secure data transmission protocols, ensuring that sensitive ECG and other physiological data are encrypted.
3. The wearable device for real-time detection of Obstructive Sleep Apnea (OSA) as claimed in the claim 1, wherein device mechanism comprises the following steps:
• Step 1: data collection - device uses single-lead ECG signals from the PhysioNet Apnea-ECG database; additionally collect form sensor data, such as oxygen saturation, respiratory effort, body position, temperature, and sound levels;
• Step 2: Data pre-processing-
o ECG signals were segmented into 60-second intervals to capture meaningful data while reducing computational load;
o Preprocessing techniques, such as filtering and auto-correlation functions, were applied to remove noisy and distorted segments, ensuring high-quality data; and
o Relevant features were extracted from each sensor, including heart rate variability from ECG, oxygen saturation levels, respiratory rate, body position, skin temperature, and snoring patterns, to enhance apnea classification accuracy.
• Step 3: Development of model
o MobileNet V1: Used as a lightweight Convolutional Neural Network (CNN) model optimized for mobile devices, responsible for capturing spatial features from ECG data;
o MobileNet V1 + LSTM / GRU: Integrated with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks to process sequential dependencies in ECG and sensor data, enhancing classification performance for apnea and non-apnea states; and
o Training Process: Models were trained on 35 patient ECG and sensor records using Adam optimization, with a learning rate of 0.001 over 100 epochs. The loss function was categorical cross-entropy, and ReLU activation functions were used for nonlinear feature extraction.
• Step 4: Evaluation Metrics- Performance was assessed using accuracy, sensitivity (recall), specificity, precision, and F1 score to evaluate how well the models distinguish between apnea and normal conditions.
• Step 5: Device Implementation- the trained models were deployed on the wearable device powered by a Raspberry Pi Pico microcontroller; wherein ECG and other sensor data are captured, processed in real-time by the deep learning models, and transmitted securely via encrypted protocols to a cloud-based platform for further analysis.
4. The wearable device for real-time detection of Obstructive Sleep Apnea (OSA) as claimed in the claim 1, wherein device achieves accuracy:
• MobileNet V1 achieved an accuracy of 89.5%.
• MobileNet V1 + LSTM reached 90.0% accuracy.
• MobileNet V1 + GRU obtained the highest accuracy of 90.29%, with sensitivity of 90.01% and specificity of 90.72%.
• The GRU model demonstrated superior performance in detecting apnea events with minimal false positives and negatives, showing a high precision of 94.71% and an F1 score of 92.33%.
5. The wearable device for real-time detection of Obstructive Sleep Apnea (OSA) as claimed in the claim 1, wherein the device demonstrated low latency and processing data in near real-time with efficient power consumption.
6. The wearable device for real-time detection of Obstructive Sleep Apnea (OSA) as claimed in the claim 1, wherein device is compact, and low-profile design facilitated comfortable overnight use, leading to higher user compliance for regular monitoring.

Documents