Modeling Mouse-based Pointing and Steering Tasks for People with Parkinson's Disease

Abstract

Mouse-based pointing and steering tasks are two primary interaction actions on computer devices. In this research, we created models for mouse-based pointing and steering tasks for people with Parkinson's Disease (PD); second, we detected PD via pointing and steering actions. The data collected from 24 participants (12 PD patients and 12 age-matched non-PD people) revealed that those with PD showed significant differences in their interaction patterns, characterized by slower movement times (MT), higher error rates, and lower information throughput compared to non-PD people. Leveraging this insight, we proposed a CNN-Transformer-based neural network model adept at PD detection, which demonstrated high accuracy in a leave-one-user-out validation. Combining pointing and steering datasets, it reached 0.96 for both AUC and F1-score. When only 10 pointing and steering actions of a user were observed, it reached an AUC of 0.99 and an F1-score of 0.96. Overall, our research contributes mouse-based pointing and steering models for PD users and provides CNN-Transformer models and computer interfaces for convenient PD detection.