TeleHealthMonitor: Edge AI for remote patient monitoring (CamvsCovid)
During the COVID-19 pandemic surge (early 2020), Indonesian public hospitals faced patient overflow that required monitoring high-risk home-isolated patients remotely. Standard telemonitoring systems assumed broadband connectivity and expensive hardware (pulse oximeters, ECG patches) -- neither assumption held for the majority of Indonesia's population, where rural broadband penetration was below 30% and healthcare hardware supply chains were disrupted.
Camera-based vitals estimation (respiratory rate via chest motion analysis) is computationally expensive if inference runs server-side -- the video stream alone would overwhelm rural 2G/3G connections. On-device inference on consumer smartphones (low-end Android, 2017-2019 chipsets) requires aggressive model compression and runtime optimization. GDPR-equivalent patient privacy requirements prohibit raw video transmission to cloud servers -- on-device inference is not just a bandwidth solution, it is a compliance requirement.
Designed an edge-first architecture: respiratory rate estimation via optical flow analysis on chest region (bounding box from lightweight pose estimation) running entirely on-device. Only the derived vital sign estimate (a single float, not video frames) is transmitted to the monitoring dashboard. Symptom self-reporting via structured mobile form provides complementary clinical signal.
Model quantized and optimized for ARM CPU inference on target chipset range. Evaluated inference latency and accuracy tradeoff across quantization levels (FP32, FP16, INT8) on physical device testing.
GDPR-compliant by design: zero raw biometric data leaves the device. Data transmission limited to aggregated vital sign estimates and structured symptom reports.
Top 3 globally at Cambridge's international COVID-19 innovation challenge. Validated the edge-first AI healthcare architecture for emerging markets -- a design pattern that has since been adopted across multiple global health AI deployment contexts where bandwidth and hardware constraints are binding. Demonstrated that GDPR-compliant vitals monitoring is achievable on commodity 2019-era smartphones at 2G data rates.