Static-Dynamic Temporal Networks for Parkinson's Disease Detection and Severity Prediction

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• Parkinson's disease (PD) significantly impacts patient mobility, with movement disorders being a key characteristic.
• Gait analysis offers a non-invasive method to identify subtle changes indicative of PD.
• Current gait analysis methods require improvement for accurate PD diagnosis and severity assessment.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for enhanced gait analysis in Parkinson's disease patients.
• To improve the accuracy of diagnosing Parkinson's disease using temporal gait patterns.
• To assess the model's capability in predicting the severity of Parkinson's disease.

Main Methods:

• Proposed Static-Dynamic temporal networks integrating two pathways: a Static temporal pathway using 1D-Convnet for sensor data and a Dynamic temporal pathway using 2D-Convnet for foot surface motion.
• The Dynamic pathway treats foot surface as an image and force point transfer as optical flow.
• Independent processing of sensor time-series data and foot sole motion information.

Main Results:

• The Static-Dynamic temporal networks demonstrated superior performance in gait detection for PD patients compared to existing methods.
• Achieved a high accuracy of 96.7% for Parkinson's disease diagnosis.
• Reached an accuracy of 92.3% for predicting the severity of Parkinson's disease.

Conclusions:

• Static-Dynamic temporal networks represent a significant advancement in gait analysis for Parkinson's disease.
• The model's high accuracy in diagnosis and severity prediction highlights its clinical potential.
• This approach offers a promising avenue for objective and early detection of Parkinson's disease.