Comparison of the Accuracy in Provisional Diagnosis of 22q11.2 Deletion and Williams Syndromes by Facial Photos in Thai Population Between De-Identified Facial Program and Clinicians.

Abstract

Introduction: There are more than 6000 genetic syndromes, therefore the recognition of facial patterns may present a challenge for clinicians. The 22q11.2 deletion syndrome (22q11.2 DS) and Williams syndrome (WS) are two different genetic syndromes but share some common phenotypic traits and subtle facial dysmorphisms. Therefore, any tool that would help clinicians recognize genetic syndromes would likely result in a more accurate diagnosis.

Methods: The syndrome identification accuracy was compared between 2 different facial analysis algorithms (DeepGestalt and GestaltMatcher) of the Face2Gene (F2G) tool and a group of 9 clinicians with different levels of expertise before and after using F2G for a cohort of 64 Thai participants' frontal facial photos divided into 3 groups of 22q11.2 DS, WS and unaffected controls.

Results: The higher accuracy from the DeepGestalt algorithm than from clinicians was demonstrated, especially when comparing between the two syndromes. The accuracy was highest when clinicians use the tool combined with their own decision-making process. The tool's second algorithm, GestaltMatcher revealed clear separation among these three groups of photos.

Discussion: The result of F2G outperforming clinicians was not surprising. However, the highest increase in accuracy was with nondysmorphology clinicians using F2G.

Conclusion: Face2Gene would be a useful tool to help clinicians in facial recognition of genetic syndromes, before ordering specific tests to confirm the definite diagnosis.