Abstract :
[en] INTRODUCTION AND METHODS
Recent advances in morphometric analysis have explored the possibility to perform an objective evaluation of the facial gestalt from 2D or 3D facial images and find a reliable syndrome match. This requires matching the face of a patient with similar patients in a database of individuals with known syndromes (Hammond, Hutton et al. 2005; Ferry, Steinberg et al. 2014). Such a user-friendly tool holds great promise of reaching a rapid diagnosis for common genetic syndromes. Especially in low resource countries, where access to laboratory testing is limited, the potential of such a tool is great.
Studying dysmorphism in Central Africa is challenging, because the facial morphology in normal individuals presents obvious differences between African and other populations (Talbert, Kau et al. 2014). In addition, the craniofacial presentation of some syndromes in a patient of African origin may differ from a Caucasian with the same syndrome as showed for the del22q11, Fragile X and fetal alcohol syndromes (McDonald-McGinn et al. 2005; Schwartz, Phelan et al. 1988; Moore, Ward et al. 2007).
We aimed to assess the performance of the existing computed phenotyping tool Face2Gene, at the current stage of its development, to recognize Down syndrome in Congolese versus Caucasian patients.
The study is part of an etiological diagnostic study in 127 patients with intellectual disability, recruited in 6 specialized clinics and schools in Kinshasa in the DR Congo.
We uploaded to Face2Gene the facial photographs of 17 DS patients from Congo and 20 DS patients from Flanders in Belgium. Patients from the 2 groups were sex and age matched. Our research protocol was approved by the Ethical Committee of the University of Kinshasa, Kinshasa, the DR Congo.
RESULTS
Face2Gene reported DS match within the first 10 matches in 16 (80 %) Belgian patients but only in 7 (35.29 %) Congolese patients. In Congolese patients, Down syndrome was the first suggested match in 2, ranked within the first 5 matches in 5 and within the first 10 in 6 of them. Conversely, Down syndrome was the first suggested match in 8, ranked within the first 5 matches in 13 and within the first 10 in 17 of them in the 20 Flemish cases. The mean rank in the Congolese patients was 8.29, which is significantly lower than the mean 4.65 recorded from Europeans (p = 0.004446 ± 0.000674). Altogether, Face2Gene showed an Accuracy of 0.35 in Congolese against 0.8 in Belgians and a Precision of 1 in both groups.
DISCUSSION
Our data indicate that the system has a high precision in both groups. However, the accuracy in the Caucasian cohort was much higher compared to the African cohort. This is interesting, since it confirms that there are differences in facial appearance of Caucasian versus African Down syndrome patients. The most likely explanation why Face2gene is underperforming in Congolese Down syndrome is that the tool is trained mostly with Caucasian cases. We therefore anticipate that the performance will improve when the system is trained with more Down syndrome cases from Central Africa. A collaborative effort to test this hypothesis is ongoing. Likewise, we expect that the efficiency of Face2Gene might improve if an increasing number of cases with a known diagnosis are uploaded.
