CT-Scan Free Neural Network-Based Reconstruction of Heart Surface Potentials From ECG Recordings

The inverse problem in electrocardiography concerns mapping electrical activity measured on the surface of the body back onto the heart in a non-invasive way. With the use of CT-scans and mathematical/geometric models of the human body, it is possible to translate body surface recording into epicardial potentials which provide advanced diagnostic information of the heart activity that a standard ECG or BSPM is unable to, especially for specific heart conditions such as arrhythmia. An encoder-decoder structure is proposed as an approach which encodes body surface potentials into latent representations before using them as input to be decoded into epicardial potentials without the use of geometric information obtained from a CT-scan. Using data from an ECG-Imaging experiment performed on dogs [1], a proof of concept is created by predicting the general wave-forms of 98 heart surface electrodes based on 168 body electrodes. The neural network manages to reconstruct the heart surface potentials with a mean square error of 0.332mV +/- 0.442 on the training set and 0.763mV +/- 0.336 on the testing set.