Objective Assessment of Cognitive Impairment in Parkinson’s Disease Using Evolutionary Algorithm

Parkinson’s disease (PD) is a common and disabling condition without cure. An early and accurate diagnosis is important for monitoring the disease and managing symptoms. Over time, the majority of patients with PD develop cognitive impairment, which is diagnosed using global tests of cognitive function or more detailed neuropsychological assessment. This paper presents an approach to detect PD and to discriminate different degrees of PD cognitive impairment in an objective way, considering a simple and non-invasive “reach and grasp” task performed with the patient wearing sensor-enabled data gloves recording movements in real-time. The PD patients comprised three subgroups: 22 PD patients with normal cognition (PD-NC), 23 PD patients with mild cognitive impairment (PD-MCI) and 10 PD patients with dementia (PDD). In addition, 30 age-matched healthy subjects (Controls) were also measured. From the experimental data, 25 kinematic features were extracted with the aim of generating a classifier that is able to discriminate not only between Controls and PD patients, but also between the PD cognitive subgroups. The technique used to find the best classifier was an Evolutionary Algorithm - Cartesian Genetic Programming (CGP), and this is compared with Support Vector Machine (SVM) and Artificial Neural Network (ANN). In all cases, the CGP classifiers were comparable with SVM and ANN, and in some cases performed better. The results are promising and show both the potential of the computed features and of CGP in aiding PD diagnosis.