Body surface potential map classifications: trade-off in group representations

For the overall quality, Q, of statistical body surface potential map (BSPM) characterisation the product of the Karhunen-Loeve (K-L) domain representation accuracy, A, and the reliability of the group conditional probability function representation R, were introduced. Reliability R, was computed according to the Kolmogorov-Smirnov distance definition taken for the optimal projections in the sense of the Sebestyen transformation and for the projection providing optimal separation of two sample sets taken from the same continuous probability distributions. Results show that at a finite set of the learning samples, Q has a peaking character for both R definitions, i.e. above and below of the optimal K-L domain dimensionality, M, the overall quality of statistical group representation deteriorates. According to the criteria formulated, current BSPM databases are small for estimating the clinical utility of maps.