Â鶹´«Ã½Ó³»­

This project is developing a lightweight, single-lead wearable ECG patch that can record the heart’s electrical activity and perform preliminary heart-failure risk classification directly on the device. 

The system combines comfortable textile-based dry electrodes, low-power ECG electronics and an interpretable machine-learning model. Short ECG recordings are checked for signal quality, analysed on the patch and summarised for the user without relying on cloud computing. 

The long-term aim is to support convenient, personalised monitoring for people living with heart failure at home.

The prototype uses a single-lead ECG sensor, an ultra-low-power analogue front end and an ESP32 processor to acquire ECG at 256 samples per second. The signal is filtered on the device and divided into five-second windows. A quality-checking stage rejects segments affected by motion, poor contact or electrical interference. Two lightweight and interpretable waveform features – Hjorth mobility and peak-to-average power ratio – are then passed to a logistic-regression model embedded in the firmware. A rolling decision from recent segments can provide a stable result in about 30 seconds, while keeping processing local to improve speed and privacy. 

In a separate subject-independent analysis using public ECG datasets, the approach achieved up to 98.3% accuracy on a small held-out cohort. Prototype functionality has also been demonstrated with three healthy volunteers using wet-gel and textile dry electrodes. These results are preliminary and require validation in larger, independently recorded cohorts, including people with heart failure, before clinical use.