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


Buyukcavus M. H., Aydogan Akgun F., SOLAK S., UÇAR M. H. B., FINDIK Y., BAYKUL T.

JOURNAL OF OROFACIAL ORTHOPEDICS-FORTSCHRITTE DER KIEFERORTHOPADIE, 2023 (SCI-Expanded) identifier

Özet

Introduction This study aimed to investigate whether the facial soft tissue changes of individuals who had undergone surgically assisted rapid maxillary expansion (SARME) would be detected by three different well-known facial biometric recognition applications.Methods To calculate similarity scores, the pre- and postsurgical photographs of 22 patients who had undergone SARME treatment were examined using three prominent cloud computing-based facial recognition application programming interfaces (APIs): AWS Rekognition (Amazon Web Services, Seattle, WA, USA), Microsoft Azure Cognitive (Microsoft, Redmond, WA, USA), and Face++ (Megvii, Beijing, China). The pre- and post-SARME photographs of the patients (relaxed, smiling, profile, and semiprofile) were used to calculate similarity scores using the APIs. Friedman's two-way analysis of variance and the Wilcoxon signed-rank test were used to compare the similarity scores obtained from the photographs of the different aspects of the face before and after surgery using the different programs. The relationship between measurements on lateral and posteroanterior cephalograms and the similarity scores was evaluated using the Spearman rank correlation.Results The similarity scores were found to be lower with the Face++ program. When looking at the photo types, it was observed that the similarity scores were higher in the smiling photos. A statistically significant difference in the similarity scores (P < 0.05) was found between the relaxed and smiling photographs using the different programs. The correlation between the cephalometric and posteroanterior measurements and the similarity scores was not significant (P > 0.05).Conclusion SARME treatment caused a significant change in the similarity scores calculated with the help of three different facial recognition programs. The highest similarity scores were found in the smiling photographs, whereas the lowest scores were found in the profile photographs.