A Hybrid Computer-Aided Diagnosis System for Central Obesity Screening in a Large Sample of Iranian Children and Adolescents

Central obesity is the basis of metabolic syndrome, which may lead to type 2 diabetes and cardiovascular disease. Its screening is critical in childhood to prevent such problems in adulthood. We presented a computer-aided diagnosis system to classify children and adolescents into obese and non-obese groups based on input features obtained from the subject's nutritional behavior, physical activity, genetics, socioeconomic status, and family history of diseases (the CASPIAN IV study). A total of 13,386 subjects (49% female) with a central obesity prevalence of 19% participated. The categorical features were converted to interval features using the Logit function, and the XGBoost classifier with grid search was then used. Other linear and nonlinear classifiers were also used for comparison. Some selected features were family history of hypertension, weight at birth, number of close friends, breakfast, and screen time categories. The proposed screening system showed a high association between predicted and observed class labels (Matthews correlation coefficient =0.76), excellent balanced diagnosis accuracy (AU-ROC =0.90), and excellent class labeling agreement rate (Kappa = 0.75) using 4-fold cross-validation. It is thus a promising screening tool. Moreover, it significantly outperformed the other tested classifiers (adj. P-value<0.05). Although, as a cross-sectional study, no causality can be inferred.