Simplifying Diagnosis of Fetal Alcohol Syndrome Using Machine Learning Methods

Abstract

Introduction: The fetal alcohol spectrum disorder (FASD) is a complex and heterogeneous disorder, caused by gestational exposure to alcohol. Patients with fetal alcohol syndrome (FAS—most severe form of FASD) show abnormal facial features. The aim of our study was to use 3D- metric facial data of patients with FAS and identify machine learning methods, which could improve and objectify the diagnostic process. Material and Methods: Facial 3D scans of 30 children with FAS and 30 controls were analyzed. Skeletal, facial, dental and orthodontic parameters as collected in previous studies were used to evaluate their value for machine learning based diagnosis. Three machine learning methods, decision trees, support vector machine and k-nearest neighbors were tested with respect to their accuracy and clinical practicability. Results: All three of the above machine learning methods showed a high accuracy of 89.5%. The three predictors with the highest scores were: Midfacial length, palpebral fissure length of the right eye and nose breadth at sulcus nasi. Conclusions: With the parameters right palpebral fissure length, midfacial length and nose breadth at sulcus nasi, machine learning was an efficient method for the objective and reliable detection of patients with FAS within our patient group. Of the three tested methods, decision trees would be the most helpful and easiest to apply method for everyday clinical and private practice.