Computer-aided facial recognition of Chinese individuals with Angelman syndrome in Hong Kong

Abstract

Background and Objective: Angelman syndrome (AS) is a genetic disorder caused by defective expression of the UBE3A gene; characterised by severe intellectual disability, seizures, absent/minimal speech with a unique behavioural profile, including exceptionally happy demeanor, easily provoked laughter, mouthing of objects and hypermotoric behaviours. Due to the subtle age-dependent facial dysmorphism, the diagnosis is commonly made only after the age of 3. We seek to determine whether a computer-aid ed facial recognition program, Face2Gene, can assist clinicians with early AS recognition in Chinese and prioritise patients for molecular testing. Face2Gene's performance has been previously evaluated in Caucasian AS patients. (http://www.ashg.org/2013meeting/abstracts/fulltext/f130121627.htm) Methods: Chinese subjects with molecularly confirmed AS were recruited with informed consent. Current and past facial images were obtained and analysed by Face2Gene. The system automatically localises hundreds of facial fiducial points and takes measurements. The final classification is based on these measurements, as well as an estimated probability of subjects having AS based on the entire facial image (http://www.fdna.com/face2gene/). Cross validation with normal Chinese and Asian with other syndromes was used to estimate and improve the recognition capability. At each of 10 rounds, the data set was split in half randomly, with the recognition methods independently optimised on the training set and evaluated on the testing set. The performance in recognising Chinese subjects with AS were evaluated before and after system training. Results and Conclusion: We have collected facial images of >30 Chinese subjects with AS. The mean AUC for Chinese AS vs normal Chinese and vs Asian other syndromes were 0.966 and 0.996 respectively which showed good recognition capability. Before system training, Face2Gene recognised AS within the first 10 syndrome matches in 7/15 subjects (47%), 4 within the first 3 ranks (27%). After training, Face2Gene recognized AS within the first 10 syndrome matches in 12 subjects (80%), 9 within the first 3 ranks (60%). Further analyses of system performance on different age, sex and molecular diagnoses are in progress. Performance of Face2Gene in detecting AS nearly improved by one fold after training, highlighting the importance of ethnic-specific data in improving the overall recognition capability of the system before its worldwide application. Acknowledgements: We are most grateful to the Hong Kong Angelman Syndrome Foundation (HKASF) and the families who participated in this study.