Classification of body surface potential map sequences during ventricular activation using Kohonen networks

The authors present a new method based on Kohonen networks for the analysis and classification of body surface potential map (BSPM) sequences. First, BSPM sequences obtained from a time interval of the cardiac cycle (e.g. QRS, ST) are presented to an untrained Self-Organizing Map (SOM). During the learning process the SOM units organize in such a way that similar BSPMs are represented in particular areas of the SOM. Time traces from the cardiac activation are then created on the trained SOM and forwarded to a Learning Vector Quantization network for final classification. In this paper the method was applied to BSPM sequences obtained during catheter pace mappings with the aim to noninvasively localize sources of ventricular tachycardia