A computerized method to assess Parkinson's disease severity from gait variability based on gender

Parkinson's disease (PD) is related to dopaminergic neuronal loss and it is progressive. Although there is no available cure for the disease yet, symptom-based treatments are available. PD can be clinically misdiagnosed in early stages because motor features become evident long after the onset of neuronal loss. Therefore, different remote monitoring tests were studied by the scholars for early detection. It has shown that people with PD exhibit gait variability with respect to healthy subjects. In this study, gait signals of PD patients were analyzed to detect severity of PD. Gait signals were converted to fuzzy recurrence plots and deep features were extracted. Machine learning techniques were applied to perform several classification experiments. Binary classifications to discriminate PD patients and multiclass classifications to predict the disease severity based on gender were conducted. Experimental results were assessed with different performance metrics. In PD severity prediction, gender based classification tests produced better performances than the test involving all cases. Proposed system produced state of the art results. The system estimated the disease severity with 1.00 and 0.99 accuracies for females and males respectively.