Low-Cost IMU-Based System for Automated Parkinson’s Subtype and Stage Classification to Support Precision Rehabilitation

Abstract

Parkinson’s disease (PD) is one of the most common progressive neurodegenerative disorder, for which early detection and precise rehabilitation planning are essential to alleviate its impact on quality of life and reduce societal burden. Accurate, automated PD subtype classification and staging play a key role in designing effective rehabilitation strategies while minimizing reliance on intensive expert assessments. Unlike existing automated methods that typically depend on high–cost medical imaging (e.g., MRI) or extensive sensor networks, we introduce a low–cost motion measurement system employing only two inertial measurement units (IMUs) placed on the lower legs. We propose a Symbiotic Graph Attention Network (SGAT)–based algorithm that fuses node features and whole-body features for automated PD subtype and stage detection. By establishing a symbiotic mechanism between the subtype and staging tasks and using adaptive fusion weights, our method achieves outstanding performance—subtype accuracy of 0.91 and staging accuracy of 0.85—validated on data from 46 participants. Notably, the entire detection and recognition process requires merely a simple walking task and incurs minimal time cost. The system’s affordability, ease of use, and scalability underscore its substantial potential for large-scale clinical deployment.