Mobile App for the Digitization and Deep-Learning-Based Classification of Electrocardiogram Printed Records

The interpretation of electrocardiograms (ECG) plays an important role in diagnosis and monitoring cardiovascular diseases, where 75% of deaths are in low and middle-income countries, whose access to experienced cardiologists is more limited. Moreover, the implementation of computer-assisted clinical decision making systems is hampered by old ECG equipment, which does not allow for exporting a digital copy of the trace. These environments need new ECG interpretation techniques that enable a rapid and simple diagnosis. This could be provided by mobile phone applications where ECG classification algorithms, based on machine learning (ML) techniques, could be embedded. In this work, firstly, we present an user-friendly Android-based mobile app for the embedding of the algorithms and the ECG record capture. Secondly, an algorithm for ECG digitisation and ML-based classification, considering different orientations and illuminations defaults requiring some processing for the extraction of the signal. Afterwards, the post processed ECG is introduced into a deep learning algorithm, specifically a residual neural network, pre-trained on the China Physiological Challenge database. The proposed methodology was tested on a set of synthetic and 50 real ECGs, achieving an accuracy of 88%. These preliminary results pave the way for improved ECG interpretation in clinical environments such as emergency units.