Accurate identification of Parkinson's disease by distinctive features and ensemble decision trees

Parkinson's disease (PD) is a progressive neurological disorder that primarily leads to a series of motor impairments. Therefore, human gait patterns and information obtained from various sensors are employed to extract distinctive features for recognizing the difference between healthy controls and PD patients. However, improper analysis of these gait symptoms may mislead the diagnosis of PD due to gradually progressive characteristics of gait disorders. Moreover, individual differences of measuring signals are often preferable to the gait intrinsic changes induced by PD. To deal with those issues, the mean, coefficient variance (CV), and asymmetry index (AI) of temporal, VGRF/BW based, and ED-based features are extracted and compared by the violin plot and Mann-Whitney U-Test to find the distinctive features and discernible changes of the PD gait. Moreover, ensemble decision trees is proposed for accurate PD diagnosis. The ensemble decision trees with features from time, VGRF/BW, and ED are tested and evaluated by the prediction accuracy. Results show that based on the mean, CV, and AI of VGRF/BW at both posterior, inside and outside heel, inside and outside arch, inside and outside sole, toe, and the total force of left and right, the proposed ensemble tree method achieves a mean accuracy of 99.52% with a standard deviation of 0.10%. The distinctive features and accurate diagnosis will be helpful for the home-based and continuous monitoring to improve treatment and therapy of PD patients.

Automated summary

Parkinson's disease is a progressive neurological disorder affecting motor functions. Gait analysis and sensor data are used to distinguish between healthy individuals and PD patients, with ensemble decision trees showing high accuracy in predicting PD based on features such as VGRF/BW.