DiffuseGaitNet: Improving Parkinson's Disease Gait Severity Assessment With a Diffusion Model Framework.

Area of Science:

• Artificial Intelligence
• Biomedical Engineering
• Neurology

Background:

• Assessing Parkinson's disease (PD) gait severity via MDS-UPDRS is subjective and resource-intensive.
• Current methods lack objectivity and efficiency in gait severity evaluation.
• Need for automated, reliable tools for PD gait assessment.

Purpose of the Study:

• To develop an automated system for predicting Parkinson's disease gait severity.
• To leverage guided diffusion models and transformer architectures for enhanced prediction.
• To generate synthetic gait data for improved model training and accuracy.

Main Methods:

• Utilized a Guided Diffusion Model with an encoder-only transformer.
• Generated synthetic PD gait video frames conditioned on expert-defined clinical features.
• Developed a novel classification algorithm trained on real and synthetic data.
• Evaluated performance on two human motion datasets for severity prediction and action classification.

Main Results:

• The proposed diffusion model accurately predicts PD gait severity.
• Synthetic data generation improved prediction performance compared to using real data alone.
• The classification algorithm demonstrated high accuracy in assessing PD severity.
• The approach showed potential for general applications with healthy subjects.

Conclusions:

• The AI-driven approach offers an objective and efficient alternative to manual PD gait assessment.
• Guided diffusion models show promise in generating realistic synthetic data for medical applications.
• The developed system can aid in clinical evaluations and research for Parkinson's disease.