Slope-Based Event-Driven Feature Extraction For Cardiac Arrhythmia Classification

To detect cardiovascular diseases (CVD), electrocardiogram (ECG) of a patient must be recorded and analyzed for a long period. For an effective diagnosis, the ECG recording system must automatically adapt to new patients. This paper presents a low-complexity artificial neural network that exclusively uses the consecutive slopes of ECG signal as inputs. These features are extracted using a level-crossing ADC and a simple TDC-based event-driven processing chain. The proposed clockless system can detect arrhythmias in ECG with 98.4%. accuracy and reduce the ANN hardware complexity by more than half compared to recent literature. It is perfectly adapted to integrated wearable monitoring systems and shows good adaptability to new patients.