Parkinson's Disease Detection via Bilateral Gait Camera Sensor Fusion Using CMSA-Net and Implementation on Portable...

Area of Science:

• Biomedical Engineering
• Neurology
• Computer Vision

Background:

• Parkinson's disease (PD) incidence is rising, necessitating advanced detection methods.
• Gait video analysis offers a promising, non-invasive biomarker for PD detection.
• Existing methods require improvement for early and accurate PD diagnosis.

Purpose of the Study:

• To develop an accurate and efficient method for Parkinson's disease detection using gait video analysis.
• To create a novel deep learning model for gait feature extraction and PD classification.
• To design a practical, portable device for real-world PD screening.

Main Methods:

• Developed a single-step segmentation method using Savitzky-Golay (SG) filtering and sliding window peak selection.
• Introduced a Cross-Attention Fusion with Mamba-2 and Self-Attention Network (CMSA-Net) for gait analysis.
• Utilized a Maximum Mean Discrepancy (MMD) loss function to enhance feature fusion.
• Collected and evaluated the method on a dual-view gait video dataset (304 healthy controls, 84 PD patients).

Main Results:

• The CMSA-Net model achieved 89.10% accuracy and 81.11% F1-score in PD detection.
• Outperformed existing methods in detecting Parkinson's disease from gait video.
• Demonstrated superior performance on a hospital-collaborated dual-view gait dataset.

Conclusions:

• The developed CMSA-Net method shows high efficacy for Parkinson's disease detection via gait analysis.
• The proposed approach enables the creation of a user-friendly, portable PD detection device.
• The device's adaptable operating modes ensure practical applicability in diverse clinical and home settings.