Dynamic cheek surface modeling for enhanced hypomimia detection in Parkinson's disease

Area of Science:

• Neurology
• Biomedical Engineering
• Computer Science

Background:

• Hypomimia, or facial masking, is a common Parkinson's disease (PD) symptom affecting up to 90% of patients.
• Reduced facial expressivity in PD significantly impacts patient quality of life.
• Current diagnostic tools for hypomimia lack robustness and automated analysis capabilities.

Purpose of the Study:

• To introduce a novel method for detecting hypomimia using cheek surface variability.
• To compare the proposed method against machine learning and deep learning models.
• To enhance the accuracy of PD diagnosis and monitoring through improved hypomimia detection.

Main Methods:

• Developed a novel approach analyzing cheek surface variability across temporal granularities in video frames.
• Implemented and compared traditional machine learning (geometric/surface features) and deep learning (3D CNN) models.
• Utilized a pre-processing pipeline to correct for head rotation and posture biases, ensuring frontal face orientation.

Main Results:

• The cheek surface variability approach achieved an 80.6% unweighted averaged recall.
• This method demonstrated superior performance compared to both machine learning and deep learning models.
• Effectively distinguished de novo, drug-naive PD patients from healthy controls.

Conclusions:

• Cheek surface variability is a valuable biomarker for hypomimia detection in PD.
• Integrating temporal dynamics into hypomimia modeling significantly enhances classification accuracy.
• The proposed method offers a promising tool for precise monitoring and diagnosis of Parkinson's disease.