Improved Deep Learning for Parkinson’s Diagnosis Based on Wearable Sensors

• DOI: 10.3390/electronics13234638
• Publication Date: 2024-11-25T13:38:24Z
• Authors: Jintao Yu; Ke Meng; Tingwei Liang; He Liu; Xiaowen Wang
• Keywords: Feature (linguistics)

Parkinson’s disease is a neurodegenerative that seriously affects the quality of life patients. In this study, we propose new diagnosis method using deep learning techniques. The takes multi-channel sensor signals as inputs, and full convolutional LSTM blocks model perceive same time-series inputs from two different views, connect extracted spatial features with temporal features. order to improve detection performance, channel attention mechanism was incorporated into model, data augmentation approach used eliminate effect unbalanced datasets on training. pd vs. hc dd classification tasks were performed, which improved accuracy by 4.25% 8.03%, respectively, compared previous best results. Both improvements higher than methods machine combined feature extraction. To utilize available resources more effectively, study conducted triple task for first time, model’s ability identify task, rate reached 78.23%. experimental results fully demonstrated effectiveness proposed diagnosis.