Facial Weakness Analysis and Quantification of Static Images

Area of Science:

• Medical Imaging
• Neurology
• Computer Vision

Background:

• Facial weakness, a symptom of neurological injury, impacts conditions like stroke and Bell's palsy.
• Digital imaging offers potential for early detection and monitoring of facial weakness.
• Current methods often rely on facial landmarks, which can be prone to localization inaccuracies.

Purpose of the Study:

• To evaluate feature extraction methods for facial weakness detection from static images.
• To compare landmark-based and intensity-based features, including Histogram of Oriented Gradients (HoG).
• To assess the accuracy of different methods on a neurologist-certified dataset.

Main Methods:

• Experimental evaluation of feature extraction techniques.
• Utilized a dataset of 186 normal, 125 left facial weakness, and 126 right facial weakness images.
• Compared landmark-based features with intensity-based features, specifically HoG.

Main Results:

• Approaches incorporating Histogram of Oriented Gradients (HoG) features demonstrated superior accuracy.
• HoG-based methods proved more effective for facial weakness detection from single static images.
• The study highlights the potential of HoG for improved diagnostic capabilities.

Conclusions:

• Histogram of Oriented Gradients (HoG) features are highly effective for detecting facial weakness in static images.
• Digital imaging combined with HoG offers a promising, accurate tool for neurological condition assessment.
• This approach can facilitate faster patient triage and recovery monitoring.