Early Diagnosis of Parkinson’s Disease and Severity Assessment based on Gait using 1D-CNN

Abstract

Gait irregularities are among the crucial signs that doctors should take into account when making a diagnosis. However, gait analysis is difficult and can depend on the knowledge of experts and the clinician’s subjectivity. To assess gait data, this research suggests a smart cutting-edge system, for diagnosis of Parkinson’s disease (PD) based on a deep learning approach. The proposed method analyzes 1-D inputs from sensors (which are connected to foot) that measure the virtual ground reaction force (VGRF). The first section of the network is composed of eighteen parallel 1D-CNNs that correlate to the system’s inputs. In the second section, the eighteen number of 1D-CNN outputs are concatenated into one unique deep array. In the third section, various classifiers such as support vector machine, multi-layer perceptron and random forest are used for final classification. The proposed methodology is used to predict between the two classes, i.e., control (CO) and PD subjects, as well as to predict the severity of Parkinson’s gait according to the Unified Parkinson’s Disease Rating Scale (UPDRS). Our test shows that the suggested method is highly effective in detecting PD from gait data. Experiments were conducted on the Physionet dataset, and the results specify that the suggested model outperforms alternative methods in terms of classification outcomes. This model can assist in the severity diagnosis of PD by using gait data.