Knowledge based refining of K-NN fuzzy classifier: a case study in ventricular arrhythmia diagnosis

A process for the detection of lethal ventricular arrhythmias is presented. This consists of the refinement, based on knowledge, of the confidence factors (membership functions) in the classification provided by a K-NN fuzzy statistical classifier. This classifier differentiates, from a set of spectral parameters obtained from segments of the electrocardiographic signal, between ventricular flutter-fibrillation, ventricular arrhythmias with complexes of aberrant morphology, and artifacts imitating both categories. In this context, one looks for the evidence that complements the confidence factors, from the initial statistical classification, in the behavior of the pressure in the pulmonary artery