Identification and Developmental Analysis of the Facial Characteristics Associated with Sickle Cell Disease using Machine Learning

Abstract Background Sickle cell disease (SCD) is a common inherited genetic disorder and contributor to global childhood mortality and morbidity. In the Democratic Republic of the Congo, nearly 40,000 newborns, approximately 2% of all newborns, are estimated to be affected each year. Despite progress in the treatment and care of the disorder, its detection and management in lower-resource settings remain challenging. Methods We collected 308 front facing photos of patients and their age-and sex-matched controls aged from 5 months to 19 years in the Democratic Republic of the Congo. Facial features were extracted and categorized into geometric and texture-based descriptors. A support vector machine ranked features according to their relevance for distinguishing SCD patients from controls. Results The facial analysis algorithm identified eight geometric and six texture discriminative features that were significantly different between the cohorts. An explainable machine learning model identified sickle cell disease with 79.5% accuracy using a combination of six geometric features: distance between medial and lateral canthi, angle at nasal ala, distance from nasion to philtrum, distance from medial canthi to the columella, distance from columella to the lower lip, and distance between nasal alae. SCD related features were identified to become increasingly discriminative with age. Conclusion These findings demonstrate the potential machine learning based methodologies to be leveraged to inform point-of-care tools in the screening and management of sickle cell disease. The discriminative facial features identified here may provide further opportunities into Artificial-Intelligence based diagnostics and personalized care strategies of sickle cell disease.