FaceKit measures 120 facial features to provide an objective basis for the diagnosis of rare genetic disorders

Using just a simple front-facing photograph, the tool compares a patient’s facial features with population reference data and also generates synthetic faces whilst managing privacy risks.

Many rare genetic diseases are accompanied by recognizable facial features, but their description is still largely based on the clinician's eye and on qualitative terms, which are difficult to reproduce from one examiner to another. FaceKit, developed by a team of twelve researchers, offers a numerical approach.

From a front photo, the tool identifies anatomical landmarks and derives 120 morphological characteristics, expressed in deviations from reference distributions built on the FairFace dataset, which covers various origins. Tested on 50 cohorts of rare diseases from the GestaltMatcher database, it finds the known facial features of these conditions.

FaceKit also knows how to generate images of synthetic faces to enrich training data, a crying need when each disease has only a few patients. The authors checked that these images do not leak information that could identify real patients.