A Markerless 2D Video, Facial Feature Recognition-Based, Artificial Intelligence Model to Assist With Screening for...

Area of Science:

• Medical technology
• Artificial intelligence
• Neurology

Background:

• Parkinson disease (PD) often presents with a masked face, a subjective clinical sign with low diagnostic consistency.
• Current diagnostic methods lack accurate technology for objective facial feature assessment in PD.
• Developing accessible monitoring tools is crucial for early PD detection and management.

Purpose of the Study:

• To develop a markerless 2D video, AI-based model for facial feature recognition in PD patients.
• To assess the model's efficacy in aiding neurologists with early PD diagnosis.
• To investigate the potential of AI in improving the accessibility of PD monitoring.

Main Methods:

• Collected 140 facial expression videos from 70 PD patients and 70 controls.
• Developed an AI model using geometric and texture facial features for masked face recognition.
• Trained and tested the AI model using Random Forest, SVM, and k-NN algorithms, comparing its performance to 5 neurologists.

Main Results:

• The AI model demonstrated effective facial feature recognition for PD diagnosis.
• Accuracy reached 83% with geometric features and 86% with texture features (Random Forest).
• Combined features achieved an F1 score of 88% using Random Forest; facial features were not correlated with PD symptoms.

Conclusions:

• AI-powered facial feature recognition offers a valuable tool for assisting PD diagnosis.
• The developed AI model shows potential for remote monitoring of PD patients.
• This technology is particularly relevant for accessible healthcare, especially during pandemics like COVID-19.