Accelerating Facial Anomaly Appraisal: A Knowledge Distillation Approach.

Area of Science:

• Computer Vision
• Machine Learning
• Medical Imaging

Background:

• Facial anomaly assessment is crucial for surgical planning and outcome evaluation.
• Current methods can be subjective and time-consuming.
• Objective, rapid, and reliable tools are needed for clinical practice.

Purpose of the Study:

• To develop a machine learning framework for detecting, locating, and evaluating facial anomalies.
• To create a universal and objective method for assessing facial abnormalities.
• To achieve high sensitivity for both minor and significant deformities.

Main Methods:

• Utilized an efficient knowledge distillation model to generate an anomaly heatmap.
• Transformed the heatmap into a severity score for facial deformities.
• Developed a system trained without anomalous data, focusing on general facial normality.

Main Results:

• Achieved state-of-the-art performance in anomaly detection and evaluation.
• Demonstrated significantly faster processing times (100 ms per image).
• Showcased 88% correlation between AI-generated scores and human judgment.

Conclusions:

• The novel framework provides a fast, objective, and reliable method for facial anomaly assessment.
• Its efficiency and accuracy support integration into mobile health applications.
• The system aids in pre-surgical planning, insurance approval, and post-surgical result evaluation.