Abstract
This study proposes a new method for detecting characteristic points (CPs), the Q and S points, in elec-trocardiogram (ECG) using a multichannel ART-based neural network (MART). The method integrates the previous two methods: the slope detection techniques and neural networks. The slope detection techniques are able to locate CPs exactly. However, it is not robust to noise. On the other hand. the method of neural networks locates approximate locations of the Cps and self-organizes in response to newly input patterns. This self-organizing ability makes the method robust. The MART integrates these two methods to implement a reliable CPs detection. For the CPs detection. ECG is divided into cardiac cycles by preprocessor, and each cardiac cycles is input to the channel one of the MART. A rectangle is made from each cardiac cycle and input to channel two of the MART. Patterns of the two channels are transmitted to the F3 layer of the MART, and then the winner node of the F3 layer recalls template patterns to the channels in the Fl layer. When the pattern recognition carried out by the MART. the template locates CPs in the ECG. The method were evaluated using MIT/BIH arrhythmia database. The standards deviation between detected CPs and CPs estimated by referee are within the limit of the SDs recommended by the CSE committee.
