Monitoring the Unseen: A 30,000-Foot View
For individuals who have recently survived a stroke, the risk of a secondary, potentially fatal stroke looms largest when they are most vulnerable: while they are asleep. This project is a wearable, real-time stroke detection system disguised as a comfortable sleep eye mask. By utilizing electrooculography (EOG) to track eye movement and an onboard microcontroller to perform real-time signal processing, the device acts as an automated sentinel. It is designed to identify the exact onset of a stroke during sleep and immediately alert caregivers and emergency personnel, transforming a delayed medical crisis into an actionable, treatable event.
Racing Against the Four-Hour Window: Framing the “Why”
In the United States alone, roughly 29,000 people suffer a recurrent stroke while asleep every year. The golden rule of stroke treatment is time; patients typically have a strict four-hour window from the onset of symptoms to receive tPA (tissue plasminogen activator), a critical clot-busting drug.
The current standard of care for stroke detection relies heavily on the F.A.S.T. method (observing Facial drooping, Arm weakness, and Speech difficulty). However, this methodology entirely breaks down when a patient is asleep. Because the time of onset is unknown for “wake-up strokes,” patients frequently miss the therapeutic window, leading to severe, irreversible brain damage or death. This project was built to solve a critical blind spot in modern post-stroke care: we needed a way to detect a stroke without waking the patient, accurately establishing the “time zero” of the event.
A Non-Invasive, Real-Time Sentinel: Delivering the Solution
To solve the silent danger of sleep strokes, the device shifts the diagnostic focus from physical limb/face symptoms to involuntary ocular changes. The core features include:
- Continuous Ocular Tracking: Embedded EOG electrodes constantly measure the corneo-retinal dipole potential, tracking eye position and movement even while the user’s eyelids are closed.
- Real-Time Anomaly Detection: An onboard algorithm continuously analyzes the differential voltage between the left and right eyes, specifically searching for “skew deviation”—a primary biological indicator of stroke onset.
- Automated Emergency Alerting: Upon detecting sustained deviation, the system triggers a localized alarm and pushes a notification to emergency contacts within a 10-second threshold.
- Unobtrusive Form Factor: All hardware is integrated into a lightweight, user-friendly sleep mask that safely monitors the patient (transmitting less than 10 µA) without interrupting their natural sleep cycle.
Bridging Hardware and Biological Signals: Architecture & Design Choices
The Stack
- Hardware: 5-Electrode EOG array, Operational Amplifiers, Low-Pass Filters, Microcontroller Unit (MCU).
- Algorithm & Processing: Signal bootstrapping, dynamic Confidence Interval (CI) generation, voltage differential analysis.
- Biological Framework: Clinical HINTS examination principles (Head-Impulse, Nystagmus, Test-of-Skew).
The Decision Matrix
- Electrooculography (EOG) over EEG or VIPS: While other monitoring solutions like EEGs or Volumetric Impedance Phase-Shift Spectroscopy (VIPS) exist, they are bulky, uncomfortable, and not viable for sleeping patients. We chose EOG because the human eye acts as a natural dipole (the cornea is positively charged, the retina is negatively charged). EOG allows us to measure eye movement highly accurately with a minimal footprint, ensuring the user can actually sleep comfortably.
- Dynamic Bootstrapping: Rather than hardcoding generic voltage thresholds that might fail across different facial structures or sleep patterns, the algorithm dynamically constructs a 95% confidence interval using real-time bootstrapping. This ensures the baseline is tailored to the specific user’s resting state, heavily reducing the likelihood of false positives.
The Algorithmic Hurdle: Filtering Normal Sleep from Stroke Indicators
The most significant engineering challenge in this project was not capturing the biological signal, but interpreting it. During normal sleep—especially REM sleep—the eyes move rapidly. How does a system distinguish between the harmless rapid eye movement of a dreaming patient and the onset of a stroke?
The solution required bridging clinical neurology with system logic. We adapted the “Test-of-Skew” from the clinical HINTS exam. During normal sleep, while the eyes may dart around horizontally, they remain vertically aligned. If a stroke occurs, it causes a disruption in the vestibular system, leading to a “skew deviation” where the eyes decouple and shift vertically in opposite directions.
To capture this programmatically, the MCU continuously processes the live EOG data from both the left and right eyes. The algorithm computes the difference between the two signals and compares it against the dynamically generated 95% confidence interval. To account for temporary artifacts or noise, a stroke is only flagged if 10 consecutive differential values fall outside the interval. In final prototype testing, this elegant, logic-driven approach successfully filtered out normal sleep noise, achieving a remarkable 93% sensitivity and 100% specificity for detecting skew deviation. (Note: Granular code implementation and circuit schematics for this logic will be covered in a future technical deep-dive).
Engineering for the Human Element: From Concept to Conclusion
Designing a wearable medical device at the intersection of hardware, software, and human biology yielded several high-level takeaways:
- Empathy is an Engineering Constraint: The most accurate sensor in the world is useless if the patient refuses to wear it. Integrating medical-grade electrodes into an off-the-shelf, comfortable sleep mask proved that user comfort and compliance must dictate the hardware footprint, not the other way around.
- Cross-Disciplinary Synthesis: The breakthrough for the detection algorithm didn’t come from computer science alone; it came from researching emergency room neurology (the HINTS exam). Translating a physical bedside exam into an automated mathematical model highlights the immense value of cross-disciplinary product thinking.
Scaling the Solution: The Road Ahead
While the finalized prototype successfully proves the concept with high clinical accuracy, maturing this device into a Class II medical product presents an exciting roadmap:
- Advanced Material Integration: Transitioning from traditional gel electrodes to dry sensors or conductive graphene woven directly into the mask fabric to improve long-term durability and patient comfort.
- Wireless Telemetry & App Ecosystem: Developing a robust mobile companion app using Bluetooth/Wi-Fi to instantly push telemetry and alerts to a network of caregivers and dispatch emergency medical services automatically.
- Machine Learning Pipelines: Upgrading the threshold-based CI algorithm by training a lightweight, on-device neural network on a wider dataset of polysomnographic recordings, further refining the device’s ability to classify anomalous eye movements.
