Facial recognition by cloud-based APIs following surgically assisted rapid maxillary expansion

Area of Science:

• Orthodontics and Dentofacial Orthopedics
• Computer Science and Artificial Intelligence

Background:

• Surgically Assisted Rapid Maxillary Expansion (SARME) is a procedure to widen the palate.
• Facial soft tissue changes following SARME are not fully understood.
• Facial biometric recognition applications are widely used for identification.

Purpose of the Study:

• To evaluate if facial soft tissue alterations after SARME are detectable by facial biometric recognition applications.
• To compare the performance of three different facial recognition APIs in detecting these changes.

Main Methods:

• Pre- and post-SARME photographs of 22 patients were analyzed using AWS Rekognition, Microsoft Azure Cognitive, and Face++ APIs.
• Similarity scores were calculated for relaxed, smiling, profile, and semiprofile views.
• Statistical analysis included Friedman's test, Wilcoxon signed-rank test, and Spearman correlation.

Main Results:

• SARME significantly altered facial similarity scores across the three tested facial recognition programs.
• Face++ yielded lower similarity scores compared to AWS Rekognition and Microsoft Azure.
• Smiling photographs resulted in higher similarity scores than other views, with statistically significant differences.

Conclusions:

• Facial recognition APIs can detect soft tissue changes post-SARME.
• Smiling facial views are more sensitive in detecting post-SARME changes.
• Profile views showed the lowest similarity scores, suggesting significant alteration in this facial aspect.