Personalized quantification of facial normality: a machine learning approach

A key challenge in facial reconstructive surgery is to define what constitutes a normal face for an individual patient and to quantify deviations and surgical changes relative to that personalized norm. The authors introduce a computerized model that generates a realistic, normalized version of any given facial image and objectively measures the perceptual distance between the raw and normalized image pair. Leveraging a StyleGAN-based generator and perceptual similarity (LPIPS), the model performs multi-objective optimization in latent space to remove anomalous features while preserving individual facial identity. The extracted distances between raw and normalized images serve as features to predict human ratings of facial normality. The system closely matches human scoring, offering a potential paradigm shift for objective surgical planning, patient education, and outcome measurement.

Osman Boyaci, Erchin Serpedin, Mitchell A. Stotland