Recognition of Parkinson's disease and Parkinson's dementia based on gait analysis and machine learning

Abstract

Parkinson's disease (PD) is a common neurodegenerative disease, with a high probability of Parkinson's disease dementia (PDD) in patients with intermediate and advanced PD. Gait disorders and cognitive disorders are common symptoms of PD patients and PDD patients. It is of great clinical significance to identify healthy elderly (HC), PD patients and PDD patients with gait characteristics under cognitive tasks. This study found that stride length, toe-off angle and heel-strike angle are important gait markers for identifying HC and PD as well as HC and PDD. Gait characteristics of multiple 7 task gait consumption can preliminarily identify PD and PDD. The gait features under multiple 7 task were used as input variables of machine learning, and the classification model was modeled by training random forest (RF) and support vector machine (SVM), and the accuracy of machine learning classification was evaluated by using the five-fold cross-validation method. The results found that the classification accuracy of all machine learning can reach more than 80%, and RF has a better classification effect. To further improve the recognition accuracy, this paper introduces recursive feature elimination (RFE) for important feature selection. By screening important features, it is found that the accuracy and AUC value of machine learning are improved to a certain extent. The highest classification accuracy of HC and PD is 91.25%, and the AUC value is 0.9127. The classification accuracy of HC and PDD was up to 97.5%, and the AUC value was 0.95. These findings have important application value for clinical diagnosis of PD and PDD. It also paves the way for a better understanding of the utility of machine learning techniques to support clinical decision-making.