Preprocessing and analysis of ECG signals - A self-organizing maps approach

The objective of this paper is to consider self-organizing maps (SOMs) as a vehicle for analysis of ECG data and making decisions as to further preprocessing and selecting classification algorithms. In contrast to other commonly used methods of unsupervised learning (such as e.g., Fuzzy C-Means or K-Means), the results formed by SOMs are more user-oriented allowing for an intensive interaction with the user in supporting various tasks of ''what-if'' analysis. In this manner, the map scan serve as a preliminary vehicle supporting a detailed system design. In the study, the standard model of SOM is augmented by several interpretation-oriented features such as region analysis and feature descriptors. The map helps reveal a structure in a set of ECG patterns and visualize a topology of such data. The role of the designer of any subsequent classifier or signal analyzer is associated with an inspection of some already visualized regions of the self-organizing map characterized by a significant level of data homogeneity and based on the discovered topology, make decisions as to the development of classification schemes. The experimental part illustrating the proposed design practices is concerned with the data coming from the MIT-BIH ECG database being commonly utilized in the realm of ECG signal analysis and classifier design.