An SVM-Based Atrial Fibrillation Detection Processor IC with a 95.8-% Sensitivity for Low-Power Event-Driven Cardiac Monitoring Applications

This work presents an energy-efficient cardiac signal processor IC for continuous atrial fibrillation (AF) detection in implantable and wearable cardiac monitoring applications. The processor utilizes the Pan and Tompkins algorithm for R-peak detection and temporal feature extraction and employs the fast Fourier transform (FFT) to capture frequency-domain features. A support vector machine (SVM) classifier module achieves AF-detection sensitivity (SE) and specificity (SP) of 95.76% and 90.98%., respectively., based on the MIT-AFDB database. The proposed cardiac monitoring system is realized in an event-driven architecture., i.e.., only when AF is detected., the ASIC transfers the corresponding episodes for secondary inspection. The ASIC operates at a clock frequency of 256 Hz and is implemented in a 180-nm CMOS process technology, which occupies an area of 5.63 mm2 including on-chip memory blocks.