Unveiling Hidden Patterns: Harnessing the Power of Short PPG-Traces for Atrial Fibrillation Detection

Unveiling Hidden Patterns: Harnessing the Power of Short PPG-Traces for Atrial Fibrillation Detection

This groundbreaking study investigates the detection of atrial fibrillation (AF) using short photoplethysmography (PPG) traces. Drawing inspiration from both cardiologists’ methods of listening to cardiac activity and the characterization of musical recordings (timbre, pitch, and scalograms), we extract musical features from the cardiac PPG recordings. By employing this innovative approach and validating the results using the MIMIC-PERform dataset, a hand-crafted approach achieves up to 89% accuracy, while a convolutional neural network with an attention module (CBAM) achieves up to 95% accuracy. These results are comparable to solutions with FDA clearance approval, but with the added advantage of using shortened traces. The findings demonstrate the feasibility of using short PPG signal recordings for AF detection and highlight the universal nature of this solution compared to more expensive alternatives that rely on ECG signals.