Qualitative morphological analysis of muscle biopsies using neural networks

Abstract

Qualitative data from human muscle biopsies have been extracted and analyzed by artificial neural network (ANN) models trained with the Kohonen's self-organizing feature maps algorithm to provide an automated medical diagnosis. Data from 6 distinct groups of neuromuscular disorders were examined. Training and evaluation were carried out on 80 and 25 cases respectively. The diagnostic performance of models investigated varied form 87 to 95%, and 88 to 92% for the training and evaluation. Furthermore, the diagnostic usefulness of the self-organizing feature map models was tested on 11 muscle biopsies with no specific diagnostic findings that gave encouraging results.