Abstract:Cardiovascular disease (CVD) has become the leading cause of death worldwide. As a widely used method for diagnosing CVD, currently electrocardiogram (ECG) monitoring tends to be implemented in wearable devices. This paper presents the prototype an ECG delineation and arrhythmia classification (EDAC) system suitable for wearable ECG biosensors. The proposed EDAC system is intended to be implemented after the electrodes and the analog front-end circuit, and its aim is signal processing at a low hardware overhead. The system consists of a Delta-modulator-based analog-to-feature converter (AFC), a corresponding ECG detection/delineation/feature extraction algorithm (DDF), an automatic gain controller (AGC) block, and a patient-dependent linear kernel support vector machine (SVM) classifier. The AFC converts the input analog signal into digital data of the slope and slope variation of the input signal, which is then used for detecting QRS complexes, localizing the fiducial points, and extracting the feature vectors for each heartbeat in the DDF block. At the same time, the AGC sends out a gain control signal based on the detected QRS complex to adjust the gain of the front-end amplifier. Finally, the SVM block performs arrhythmia classification. The EDAC system performance is evaluated using the MIT-BIH arrhythmia database. The system achieves 0.88% (0.93%), 99.1% (99.1%), 87.0% (92.8%), 99.6% (99.5%), and 89.3% (92.9%) in F1 score, accuracy, sensitivity, specificity, and positive predictive values of the supraventricular ectopic beats (ventricular ectopic beats) versus normal heartbeats classification while maintaining a low power dissipation ( $1.66~\mu \text{W}$ at 1 kHz operating frequency in the front-end AFC block). The proposed system is attractive to future wearable long-term ECG monitoring biosensors.

A Near-Sensor ECG Delineation and Arrhythmia Classification System