This is published under the terms of CC BY-NC.

Downloaded: 539 times

Abstract

Parkinson’s disease (PD) is a chronic and long-term disease that seriously affects patients’ quality of life. In underdeveloped areas, early detection of PD is primarily based on medical observation and patient self-description. Early diagnosis of PD can effectively reduce the disease’s progression. Recent studies have suggested that the motor symptoms of PD can be reflected in plantar pressure. However, traditional machine learning models require manual feature selection, which can be time-consuming. Furthermore, although deep learning has seen rapid development, many clinical characteristics have not been taken into consideration. To address these limitations, a dual self-attention Transformer model is proposed to explore the spatial correlation of plantar space and the temporal correlation of the gait cycle. Considering the presence of symptoms such as foot tremors in PD patients, a masking mechanism is designed to focus locally on the unilateral foot during the support phase. An experimental paradigm is designed to evaluate the model’s generalization capability across different subjects. The experimental results demonstrate that the proposed model achieves superior classification performance for the early detection of PD based on plantar pressure data.