Related Experiment Video
Updated: Mar 13, 2026
Midface Hypoplasia and Cranial Base Morphology in Syndromic Craniosynostosis: A Comparative Analysis Study Using a Predictive Regression Model
Published on: November 4, 2025
Evaluation of a Facial Dysmorphology Analysis Algorithm (Face2Gene) in Identifying Treacher Collins Syndrome Amongst
Jie Han Timothy Sng1, Jonas Jun Jie Hue2, Josep Rubio-Palau3,4
1Discipline of Oral and Maxillofacial Surgery, Faculty of Dentistry, National University Centre for Oral Health, National University of Singapore, Singapore, Singapore.
Background:
Treacher Collins Syndrome (TCS) is an uncommon congenital disease of the craniofacial complex. While there are 'classic' facial manifestations of TCS, they present with a wide range of variability. Face2Gene (F2G) is a deep-learning algorithm that can provide differential diagnoses of syndromes via analysis of 2-dimensional facial images. This may aid early recognition of TCS, thus improving early multidisciplinary management. Our primary aim is to evaluate and compare the performance of F2G in TCS patients of various races. The secondary aim will be to correlate the pathognomonic facial features of these TCS patients with the diagnostic stratification by F2G.
Methods:
Publicly available images of TCS patients of White (n = 81), Chinese (n = 23), and Indian (n = 23) race were sourced for and screened prior to inclusion for documentation of facial dysmorphological features and F2G evaluation. The diagnostic sensitivity of F2G and computed gestalt score of TCS for each image were then derived. Statistical analysis for correlation between clinical features and F2G-derived gestalt scores for the TCS patients was performed.
Results:
A total of 127 images of TCS patients were analysed by F2G. A high diagnostic accuracy of 93.7% was obtained. However, F2G was less confident when diagnosing Asians with TCS than White individuals (p < 0.05). Malar hypoplasia was less prevalent in Indian TCS patients (p < 0.001). However, multidimensionality reduction showed no significant differences across the racial groups. Certain facial features were associated with higher model certainty in TCS diagnosis.
Conclusion:
F2G is a useful diagnostic adjunct for TCS. Nonetheless, further training on non-White datasets may help improve model certainty in diagnosis.
