Parkinson's severity diagnosis explainable model based on 3D multi-head attention residual network
Jiehui Huang1, Lishan Lin2, Fengcheng Yu1
1Artificial Intelligence Medical Research Center, School of Intelligent Systems Engineering, Shenzhen Campus of Sun Yat-sen University, Shenzhen, Guangdong 518107, China.
Computers in Biology and Medicine
|January 12, 2024
Summary
This study introduces a novel deep learning model for Parkinson's disease (PD) severity assessment using video analysis. The explainable MARNet achieves state-of-the-art performance, offering a non-invasive tool for PD telemedicine.
Area of Science:
- Neurology
- Computer Science
- Medical Imaging
Background:
- Parkinson's disease (PD) severity assessment is crucial for effective treatment.
- Current methods for PD severity evaluation have limitations, including reliance on prior knowledge or invasive procedures.
Purpose of the Study:
- To develop a generalized and explainable deep learning model for non-invasive Parkinson's disease severity evaluation using video data.
Main Methods:
- Proposed an explainable 3D multi-head attention residual convolution network (MARNet).
- Utilized 3D attention-based convolution layers for video feature extraction.
- Employed LSTM and residual backbone networks to capture video contextual information.
- Incorporated a feature compression module to condense learned features.
Main Results:
- The MARNet model achieved state-of-the-art performance in Parkinson's disease severity diagnosis.
- Interpretable experiments demonstrated the model's generalizability and explainability.
Conclusions:
- The proposed MARNet is a lightweight and effective deep learning model for video-based PD severity evaluation.
- This model serves as a potential end-to-end baseline for future research and large-scale application in PD telemedicine.
