Diagnostic wandering and delayed management are major issues in rare diseases. Here, we report a new Next-Generation Phenotyping (NGP) model for diagnosing Coffin Siris syndrome (CSS) on clinical photographs among controls and distinguish the different genotypes. This retrospective and prospective study, conducted from 1998 to 2023, included frontal and lateral pictures of confirmed CSS. After automatic placement of landmarks, geometric features extraction using procrustes superimposition, and textural features using a gray-level co-occurrence matrix (GLCM), we incorporated age, gender, and ethnicity and used XGboost (eXtreme Gradient Boosting) for classification. An independent validation set of confirmed CSS cases from centers in Bangalore (India) and Tbilissi (Georgia) was used. We then tested for differences between genotype groups. Finally, we introduced a new approach for generating synthetic faces of children with CSS. The training set included over 196 photographs from our center, corresponding to 58 patients (29 controls, 29 CSS). We distinguished CSS from controls in the independent validation group with an accuracy of 90.0% (73.5%-97.9%, p = 0.001). We found no facial shape difference between the different genotypes.
Keywords: ARID1B; ARID2; Coffin Siris syndrome; dysmorphology; next generation phenotyping.
© 2024 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd.