Noninvasive detection of lower extremity artery disease using multi-site photoplethysmographic signals and machine learning

Lower extremity stenosis (LES) is a prevalent cardiovascular condition and a strong indicator of systemic arteriosclerosis. Existing diagnostic methods are invasive or operator-dependent and often unavailable at the primary care level. Here, we propose a non-invasive screening method based on morphological features extracted from multi-site PPG signals that were calculated from multi-site photoplethysmography measurements during a proof-of-concept clinical study at University Hospital Tübingen. Machine learning methods were used to classify 42 patients with 2 measurements each at four locations (i.e., a total of 84 measurements and 336 signals) into either the control or stenosis group. An overall classification accuracy of 82.6% (sensitivity: 0.750, specificity: 0.865) is achieved. This result is clinically relevant and shows that the selected features are effective for detecting stenosis and could be used as a screening method at the primary physician level.