Unsupervised Learning based Feature Points Detection in ECG

ECG will change its shape according to the patient condition, and the ECG of different patients are also differ from each other. Therefore it is the requirement that we must choose and proposed system that can cater these abilities to adopt themselves accordingly to make the system robust. To manage altering shape of ECG a self-organized, unsupervised learning based, robust Neural Network is needed. This study proposes a new method of Feature Points (FP's) detection in ECG using hybrid of DP-Matching and ART2 neural networks in Multichannel-ART (MART) neural networks. To manage the altering shape of ECG's a self-organized and robust Neural Network is needed. In this study we use two channels Multichannel ART (MART). Channel one uses the robust technique of ART2 for the detection of FP's of QRS wave (Q and S points) using template-matching method of triangle patterns. The second Channel uses rectangle output from DP-Matching technique for the exact location of FPs. The method updates the channel one templates through learning. The method is evaluated using MIT/BIH arrhythmia database. The standard deviations (SD's) between detected FP's and FP's identified by the referee are well within the limit of the SD's recommended by the CSE committee.