Medical diagnostic systems using ensembles of neural SOFM classifiers

A design for medical diagnostic systems composed of ensembles of neural self organizing feature map (SOFM) classifiers is presented. Each SOFM classifier was fed with a different feature set extracted from the raw data and their results were combined using (i) majority voting and (ii) a confidence measure derived from the SOFM which weighted the contribution of each feature set to the final classification result. The following two diagnostic systems were developed: (i) a decision support system for the assessment of electromyographic (EMG) signals, and (ii) a system for the characterisation of carotid plaques from ultrasound images. The results in this work shows that combining the classification results of multiple classifiers using as input multiple feature sets, in conjunction with the use of a confidence measure can improve the overall classification performance of the system