Abstract
Body surface potential maps (BSPMs) derived from multi-channel ECG recordings enable the detection and diagnosis of electrophysiological phenomena beyond the standard 12-lead ECG. In this work, we developed two AI-based methods for the automatic detection of location of the electrodes used for BSPM: a rapid method using a specialized 3D Depth Sensing (DS) camera and a slower method that can use any 2D camera. Both methods were validated on a phantom model and in 7 healthy volunteers. With the phantom model, both 3D DS camera and 2D camera method achieved an average localization error less than 2 mm when compared to CT-scan or an Electromagnetic Tracking System (ETS). With healthy volunteers, the 3D camera yielded average 3D Euclidean distances ranging from 2.61 ± 1.2 mm to 5.78 ± 3.09 mm depending on the patient, similar to that seen with 2D camera (ranging from 2.45 ± 1.32 mm to 5.88 ± 2.73 mm). These results demonstrate high accuracy and provide practical alternatives to traditional imaging techniques, potentially enhancing the interest of BSPMs in a clinical setting.