A self-organizing neural net clustering Parkinson patients and control persons using motor data

Abstract

Parkinson's disease (PD) is a neurodegenerative disorder characterized by akinesia (absence or poverty of voluntary movements), rigidity (increased muscular tonus) and tremor (involuntary oscillation). Control of the course of disease and therapeutic measures impose great demands on standardisation of clinical evaluation, including standards of therapeutic control and means of disease monitoring as well as for proof of efficacy of new therapeutic substances. We present the results of the application of a self-organizing feature map (SOM) to training data, obtained from a large study, where more than 450 patients have been observed under therapy for over 2 years. The training data are obtained from an instrumental test battery, whose items like steadiness, aiming or tapping describe the motor impairment of the patients for given tasks. Different SOM nets with a size of 50/spl times/50 neurons were presented to learn 501 data sets including 49 control persons, each set consisting of 28 components. The result of the learning process, which has been achieved after 10,000 learning steps is the clear separation of Parkinsonian patients from the control persons by the neural net.<>