Diagnosis of Cardiac Abnormalities Applying Scattering Transform and Fourier-Bessel Expansion on ECG Signals

This work intends to devise an efficient feature extraction scheme for identifying common cardiac abnormalities using the Fourier-Bessel (FB) expansion of RR-intervals and time-frequency based features of Electrocardiogram (ECG) signals. The Bessel basis, when used for representing the RR-intervals, meaningfully enhances the pathologically induced low-frequency changes in terms of FB coefficients. To ensure the characterization of diverse pathological variability present in the ECG signals, time-frequency domain features are also extracted using scattering transform. The multi-label classification of the ECG signals, for five different lead combinations as mentioned in the Phys-ioNet/CinC Challenge 2021, is performed using Gated recurrent unit into specified twenty-six categories. We have participated in this Challenge as team "Medics". Our code failed to run on the validation set during the official phase of the Challenge, hence our entry was not officially ranked in the Challenge. The experimental outcomes, for five-fold cross validation using 2021 PhysioNet/CinC Challenge dataset, demonstrates the mean Challenge scoring metric on the twelve-lead, six-lead, four-lead, three-lead, and two-lead combinations as 0.40, 0.43, 0.43, 0.44, and 0.45 respectively. According to the results, the proposed method justifies the use of the FB and scattering transform together for the detection and identification of common cardiac problems using ECG signals.