Ecg Arrhythmia Classification Using Spatial Representations and Convolutional Neural Networks

We report a comparative analysis of the classification performances of three distinct types of spatial representations of ECG arrythmia signals when combined with various Convolutional Neural Networks architectures. The time series-toimage conversion methods are using classical spectrograms, a modified version of the Continuous Wavelet Transform (the STransform), and the Gramian Angular Field. Experiments conducted on the Shaoxing-Chapman ECG database revealed that the analyzed solutions compare favorably against previously reported performances.