Cardiac Abnormalities Detection Through 12-Lead/Reduced-Lead ECG Spectrograms and 2D-CRNN

Heart abnormalities represent around 32 % of the deaths for illnesses in the world. This work presents an automated pattern recognition method for detecting 25 different cardiac arrhythmias and the normal sinus rhythm type. A two-dimensional convolutional recurrent neural network (2D-CRNN) model was employed by using raw-data images and signal spectrograms. With a database of 88,253 ECG signals of 12-leads, a four-step method was created. 1. Preprocessing. The data were filtered and downsample or filled in so that they were the same length. 2. Representation. Two sets of images were obtained, one with the time series and the other with spectrograms through Wavelet Synchrosqueezing (WS). 3. Feature extraction. A CNN network was chosen to get relevant features of the images; these were flattened in a vector to feed a recurrent neural network. 4. Classification. A fully connected layer was used to classify the signals. A 10-fold cross-validation method (CV) was executed. The grid search technique was used to obtain the optimal threshold to improve the Challenge Score (Ch-score). Results: With this model, our UIDT_UNAM team received an unofficial score of 0.34 for all sets of established leads utilizing the hidden test data.