Rules found by multimodal learning in one group of patients help to determine optimal treatment to other group of Parkinson’s patients

Abstract

We have already demonstrated that measurements of eye movements in Parkinson’s disease (PD) are diagnostic. We have performed experimental measurements of fast reflexive saccades (RS) in PDs in order to predict effects of different therapies. We have also found rules by means of data mining and machine learning (ML) in order to classify how different doses of medication have determined motor symptoms (UPDRS III) improvements. These rules from one group of 23 patients only on medications were supplied to another group of 18 patients under medications and DBS (deep brain stimulation) therapies in order to predict motor symptoms changes. Such parameters as patient’s age, neurological and saccade’s parameters gave a global accuracy in the motor symptoms predictions of 76% based on the cross-validation. Our approach demonstrated that rough set rules are universal between groups of patients with different therapies that may help to predict optimal treatments for individual PDs.