Training Of ECG Signals In Neural Network Pattern Recognition
Qiuzhen Xue, Yu Hen Hu, Willis J. Tompkins · 2005
We propose a new method for training multi-layer feed-forward neural networks to recognize ECG signal patterns. Specifically, we apply a shifted versions of multi-cycle ECG signals to the ANN model. This approach is compared with a higher-order input connections ANN model. From a principal components analysis of the weight patterns using these two method, we found that their performances are similar. We also demonstrate the matched filtering effect of neural networks in ECG pattern recognition when large number of hidden units are used.