On the design of a class of CNN's for ECG classification

The paper discusses the possibility of using the dynamics of a class of Cellular Neural Networks (CNN's) for electrocardiogram (ECG) signals classification. The main idea is that of segmentation and transformation of the temporal signal into a 1D spatial one which is further processed by means of a bank of linear spatial filters using a parallel architecture of CNN type. A major advantage of the proposed solution is the independence of the filters spatial frequency characteristics on the number of samples of the ECG pattern, which allows dealing very easily with the heart rate variability. The principle of the proposed architecture is briefly discussed and the design of a bank of spatial filters for ECG classification is presented. Transistor level simulation and considerations regarding the architecture reconfiguration are given as well.