Human obesity has become a global epidemic. Body mass index (BMI) is clinically useful data for the diagnosis of overall adiposity. The purpose of this study was to identify normal and overweight patients based on facial characteristics extracted from subject image data, irrespective of the measurement of weight and height. In this paper, we propose a prediction method for normal and overweight from morphological facial characteristics that are associated with overweight and normal BMI statuses. A total of 1244 subjects participated in this study. The subjects were divided into 6 groups based on age- and gender-specific differences. The area under the receiver operating characteristics curve (AUC) and kappa of the prediction model ranged from 0.760 to 0.931, and from 0.401 to 0.586, respectively, for all groups, except for the group comprising females aged ≥61 years. Statistical analysis revealed many features that were significantly different between overweight and normal in the 6 groups. Furthermore, compact and useful feature sets were identified for BMI prediction using facial features in gender- and age-specific groups. We identified a relationship between facial morphology and BMI status, and the possibility of predicting the BMI status of individuals. Our results will facilitate the development of improved applications for age- and gender-specific groups in the fields of adiposity, facial recognition, and
medicine. |
