Deep Learning-Based Assessment of Lip Symmetry for Patients With Repaired Cleft Lip

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Surgical outcome assessment

Background:

• Post-surgical lip symmetry is crucial for evaluating cleft repair success.
• Traditional landmark-based methods are limited for video analysis and lack texture assessment.
• Automated, quantitative lip symmetry evaluation is needed.

Purpose of the Study:

• To develop an artificial intelligence (AI) approach for automated lip symmetry assessment.
• To quantify lip symmetry using contrastive learning on lateral lip morphology.
• To classify the severity of asymmetry without requiring patient images for training.

Main Methods:

• Utilized contrastive learning to measure similarity between lip image representations.
• Introduced simulated lip asymmetry via temporal misalignment and face transformations in control subjects.
• Trained a deep learning model to differentiate left and right lip image representations.
• Evaluated the model on 146 images of patients with repaired cleft lip.

Main Results:

• The model trained with face transformations achieved 75% weighted accuracy and a Pearson correlation of 0.31 with expert evaluations.
• The model trained with temporal misalignment achieved 69% weighted accuracy and a Pearson correlation of 0.27.
• The AI model successfully categorized patient images into five asymmetry levels.

Conclusions:

• An automated AI approach for lip asymmetry assessment in repaired cleft lip patients was developed.
• The method uses transformed control subject images, eliminating the need for manual landmarks.
• This AI tool offers a more efficient and objective method for evaluating surgical outcomes.