Heart Disease Detection Architecture for Lead I Off-the-Person ECG Monitoring Devices

With the rise of smart-watches and other wearables, off-the-person electrocardiography is gaining momentum as high-quality Lead 1 ECG signals can now be acquired from a persons hands or arms. Although several heart disease detection algorithms have been described in recent years, they are not designed considering Lead 1-only setups. This work bridges this gap with an architecture for a robust Lead 1 real-time heart disease detection system and an FPGA-based implementation. The proposed system is based on a signal processing pipeline composed of: ECG signal denoising; heartbeat detection and segmentation; extraction of dynamic morphological features; and heartbeat classification (standard and different abnormal heartbeats). Resorting to the only database from MITs Physiobank with Lead 1 annotated recordings, InCarTDb, the proposed pipeline resulted in a 4-class model with a classification accuracy of up to 96.5%. Moreover, when implemented in a Zynq-7 ZC702 Evaluation Board, the proposed architecture requires less than 30% of the FPGA resources and a total power consumption of 192 mW at a clock frequency of 35 MHz.