COVID-19 Detection Using ECG Signals and Bag-of-Words Classifier

COVID-19 has been associated with several ECG abnormalities, hence this type of data has been investigated as a possible reliable indicator of the presence of the infection. The present paper aims to assess the performances of bag-of-words classifiers when processing digitized versions of paper multi-channel electrocardiograms. Four distinct feature extraction procedures and four different encoding strategies are evaluated in terms of the Area Under Curve (AUC), accuracy, sensitivity, and F1-score. Extensive experiments on an augmented version of a well-known public ECG dataset indicate that sparse encoding yields the best performances in line with more sophisticated classification approaches. The bag-of-words models are evaluated on various combinations of feature type, encoding strategy, and codebook dimension, showing significant robustness against all those parameters.