Nature inspired concepts in the electrocardiogram interpretation process

In this paper we compare and evaluate the use of the following methods: Ant Colony inspired Clustering, Ant Colony inspired method for Decision Tree generation, Radial Basis Function Neural Networks with different learning algorithms and compare them to classical approaches, such as hierarchical clustering and k-means. We have evaluated the methods on the annotated MIT-BIH database. In the case of Ant Colony inspired clustering we have also studied the Dynamic Time Warping (DTW) measure. The DTW measure improved Se about 0.7% and Sp about 0.9% when compared to classical feature extraction for a #106 signal. The best-performing has been the agglomerative hierarchical clustering (Se=94.3, Sp=74.1), however it is practically unusable as it is memory and computational demanding. Acceptable results (complexity vs. error) have been obtained by the Ant-Colony inspired method for Decision tree generation (Se=93.1, Sp=72.8).