HealthGuard
Two purposes of healthguard
Monitoring - Continuously collects health metrics from inbuilt smartwatch sensors to provide users with alerts about potential health risks
Diagnosing - Using a smartwatch to accurately detect heart failure in nonclinical environments
In a clinical setting, a 12-lead ECG is currently needed to diagnose heart failure (ie Myocardial Infarction)
HealthGuard
Technical and Product OverviewAugust 04, 2023
Abstract
HealthGuard is a pioneering health monitoring application designed to harness the power of wearable technology to predict and provide early warnings for strokes and heart attacks. Utilizing EdgeMind's edge machine learning (ML) technology, HealthGuard is an application operating on Android-based smartwatches. This whitepaper provides a detailed overview of HealthGuard's technical implementation, the integration of EdgeMind, and the application's functionalities, making it a valuable resource for developers, health-tech enthusiasts, and healthcare professionals alike.
I. Product Overview: HealthGuard
HealthGuard, available on the Play Store for Android-based smartwatches running Wear OS, leverages the watch's integrated sensors to gather vital health data. It employs sophisticated ML algorithms directly on the device to analyze these data and predict potential heart attack or stroke risks, offering an invaluable early warning system for users.
II. HealthGuard: Key Features
Real-Time Health Monitoring: Continuously collects health metrics from inbuilt smartwatch sensors.
On-Device Inference: Utilizes EdgeMind's ML technology to conduct data inference directly on the watch, guaranteeing low latency and increased privacy.
Early Warning System: Provides users with timely alerts about potential health risks, enabling early interventions.
Data Privacy: Processes and analyzes all data on the device, ensuring user privacy and compliance with data protection regulations.
III. HealthGuard Application Development
The HealthGuard application was developed using a combination of Android's Kotlin language for its native capabilities and the powerful, cross-platform Flutter framework for the user interface. The application was designed and built with a particular focus on leveraging the inbuilt sensors of Android smartwatches effectively. It interacts with these sensors via Android's Sensor API to collect necessary health data.
The application incorporates EdgeMind's platform-specific SDK, allowing seamless integration of EdgeMind's Runtime Container, which houses the ML model. The HealthGuard app interacts with this Runtime Container through a well-defined API provided by EdgeMind's SDK, enabling the application to execute ML model inference and receive the prediction results.
IV. EdgeMind Integration and ML Workflow
HealthGuard's key strength is the integration of EdgeMind, enabling efficient deployment of ML models onto the smartwatch. The process follows EdgeMind's standard workflow:
Model Development: ML models to predict stroke and heart attack risks are created in a resource-rich environment using Python.
Model Training: These models are trained on relevant health datasets to minimize prediction errors.
Model Validation: After training, models are validated using separate datasets, ensuring accuracy before deployment.
Model Conversion: Validated models are converted into a format compatible with EdgeMind's Runtime Container, optimizing for execution in low-resource environments.
Containerization: The converted model is encapsulated within EdgeMind's Runtime Container, along with necessary data preprocessing and postprocessing scripts.
Deployment: The Runtime Container is deployed to the smartwatch using EdgeMind's platform-specific SDK.
Execution of Model: The ML model executes on the device, providing real-time analysis of sensor data.
Monitoring and Updating: The performance of the deployed model is continuously monitored, with updates or retraining triggered as needed.
Continuous Operation: If the model performs satisfactorily and no new data prompts an update, the model continues operating on the device, delivering real-time health monitoring.
V. Security and Privacy
HealthGuard prioritizes user privacy and security. It employs HTTPS for secure data communication, encrypts data at rest, and adheres to international data security standards and health data regulations, including HIPAA and GDPR.
VI. Future Developments
HealthGuard anticipates significant enhancements, including the integration of Electronic Health Records (EHR), expanded predictive models for other health conditions, and compatibility with additional smartwatch models and other wearable devices.
VII. Conclusion
HealthGuard, backed by EdgeMind's edge ML technology, represents a significant advancement in health-tech wearable solutions. Its focus on real-time, on-device data processing, and privacy makes it a promising tool for wider adoption, offering the potential to save lives through early detection and prevention.
Report content on this page
Product
The document introduces HealthGuard, an innovative health monitoring application for Android-based smartwatches. Developed in collaboration with EdgeMind's edge machine learning technology, HealthGuard aims to predict and provide early warnings for strokes and heart attacks. The paper covers various aspects of HealthGuard:
Product Overview: HealthGuard operates on Android smartwatches and uses integrated sensors to gather health data. It employs on-device machine learning algorithms to analyze the data and predict potential heart attack or stroke risks, offering timely alerts.
Key Features: HealthGuard offers real-time health monitoring through smartwatch sensors. It uses EdgeMind's machine learning technology for on-device analysis, ensuring low latency and data privacy. It functions as an early warning system, alerting users to potential health risks.
Application Development: The app is developed using Android's Kotlin language and the Flutter framework for the user interface. It effectively utilizes the smartwatch's sensors through Android's Sensor API and integrates EdgeMind's SDK for machine learning model execution.
EdgeMind Integration and ML Workflow: HealthGuard's strength lies in its integration of EdgeMind. The process involves model development, training, validation, conversion for the Runtime Container, deployment to the smartwatch, execution, monitoring, and updating.
Security and Privacy: HealthGuard ensures data security with HTTPS communication, data encryption, and compliance with data protection regulations like HIPAA and GDPR.
Future Developments: HealthGuard plans to integrate Electronic Health Records, expand predictive models for various health conditions, and support more smartwatch models and wearable devices.
Smartwatch tech
photoplethysmography (PPG) technology
ECG - electrocardiogram
Some smartwatches have special algorithms that can use PPG light technology to detect atrial fibrillation
Biometric data collection
Heart rate detection: This can include a heart rate calculation (if your heart rate is between 30 and 210 beats per minute), heart rate variability (changes in the time between each heartbeat), and alerts for an abnormally high or low heart rate.
Atrial fibrillation detection: This is the only arrhythmia the FDA has cleared some smartwatches to detect. The watches that can detect atrial fibrillation do this using PPG and a single-lead ECG.
Blood pressure detection: An inflatable cuff built into the watch can measure this. Or the same light-sensor technology (PPG) used to detect heart rate (more on this below) can measure it.
Blood oxygen level: This can give information about how well your heart is circulating blood to the lungs and the rest of the body.
Diseases/issues
Atril fribulation
Stemi & Nstemi heart attack
Stroke in the brain - loss of blood (ischaemic or hemorrhagic)
hypertension
Myocardial infarction (MI) (heart attack) Apple watch
The standard diagnostic step for MI is the 12-lead ECG, which requires specific equipment and expertise. The Apple Watch records a single-lead ECG using two electrodes. The Apple Watch is not intended to replace the standard 12-lead ECG but could be a new screening tool for MI. Despite other mobile ECG devices being available, the Apple Watch's popularity gives it an edge. About 20% of US residents own a wearable device. The goal is to simplify the ECG recording procedure and leverage wearable devices to promote heart health.
Watches
Apple: heart rate detection, atrial fibrillation detection, blood oxygen detection
Fitbit: heart rate detection, atrial fibrillation detection, blood oxygen detection
Samsung: heart rate detection, atrial fibrillation detection, blood oxygen detection
Google: heart rate detection, atrial fibrillation detection
Omron: heart rate detection, blood pressure detection
Garmin: heart rate detection, blood oxygen detection
Withings: heart rate detection
Limitations
Artifact: This is “noise” in the signals the watches can detect and report on. It can lead to a false reading. In turn, this can cause unnecessary anxiety.
Lower accuracy for heart rate detection in people with darker skin: A recent review found that the light detection methods used by smartwatches may not work as well in people who have more pigment in their skin.
12-lead electrocardiogram vs one sensor on the writst
Blood pressure readings
Smartwatch blood pressure detection is a work in progress. The Omron Heartguide is the only smartwatch that’s FDA cleared for blood pressure monitoring. It uses an inflatable cuff within the watchband, similar to the blood pressure cuffs in medical offices.
Other smartwatch companies are developing the photoplethysmography technology already used for heart rate measurements. This would be even easier and more comfortable than the inflatable cuff models currently available. Companies like Valencell and Fitbit are still validating this technology, so it’s not quite ready.
When it comes to monitoring blood pressure, the most accurate measurements are made with a more traditional automatic arm cuff, either at home or in a provider’s office.
Wearables ECG study
A study from University College London analyzed 83,000 healthy individuals aged 50 to 70 who took a 15-second ECG similar to those conducted using smartwatches.
Approximately 1 in 25 participants with an extra heartbeat were found to have twice the risk of developing heart failure or atrial fibrillation in the next decade.
Dr. Michele Orini from UCL suggests that ECGs from consumer-grade wearables could aid in detecting and preventing future heart diseases.
Last year, Mayo Clinic researchers used AI analysis of Apple Watch ECG readings, uploaded via an app, to identify people with a weak heart pump, or left ventricular dysfunction – a condition affecting some 9% of people over 60. In 2020, Harvard researchers found smartwatch-generated ECGs were 93-95% accurate in identifying and distinguishing between different types of heart attacks.
Current smart watch technology
The three current Apple Watch model — the Apple Watch Series 7, Apple Watch SE, and Apple Watch Series 3 — send high and low heart rate and Irregular heart rhythm notifications. Fitbit, Withings, Samsung, and other smart watches also have FDA clearance to identify and send alerts about potential AFib.
Good quotes
Users - According to Pew Research Center, 1 in 5 Americans uses a smartwatch or wearable fitness tracker.
Heart attack - About every 40 seconds, someone in the United States has a heart attack.
Although we associate chest pain with the onset of a heart attack, up to a third of people don’t have this symptom
“The Apple Watch is not intended to replace the standard ECG. Rather, it is meant to be used as a self-check screening tool for people who have chest pains or other heart attack symptoms at home or in other environments, with the goal of decreasing the time to treatment and improving the outcome,” Dr. Perin.
Smart watch as a 9-lead ECG
9-lead ECG by having the wearer perform multiple maneuvers to obtain a recording
“These larger studies are needed to ensure that the Apple Watch can provide accurate results similar to those obtained with a standard 12-lead ECG in helping to identify or rule out a heart attack,” according to Emerson C. Perin, MD, PhD.
Future studies should aim to develop a procedure that involves only 3 to 4 recordings, which would minimize the time required for the ECG and the chance of making mistakes.
Study - Using the Apple Watch to Record Multiple-Lead Electrocardiograms in Detecting Myocardial Infarction (heart attack)
ECG vs EKG
Why do we need an early warning version of a 12-lead ECG About every 40 seconds, someone in the United States (US) has a myocardial infarction (MI).1 The outcomes after MI depend on the time that elapses before treatment begins. More than half of individuals who have an MI die in an emergency room or before reaching a hospital within an hour of symptom onset.2 The benefit of early treatment of MI is clear: survival rates increase by up to 50% if reperfusion is achieved within one hour of symptom onset and by 23% if achieved within 3 hours.3–5 Therefore, immediate treatment is essential. Without it, patients have higher rates of mortality and severe complications caused by increased infarct size, including cardiogenic shock, arrhythmias, and heart failure.
To receive immediate diagnosis and treatment, patients need to seek medical attention as soon as symptoms develop. Unfortunately, in the US, the median time from symptom onset to hospital arrival ranges from 1.5 to 6 hours
To check out
Samsung 3 in 1 sensor
Detect MI https://www.ncbi.nlm.nih.gov/books/NBK459269/
Resources
https://www.goodrx.com/health-topic/heart/smartwatch-heart-health
Last updated
