Machine learning-based classification of children affected by malnutrition using multimodal sMRI and DTI brain images

Child malnutrition is a prevalent pathological condition in developing countries, such as Mexico, with scarce studies on its anatomical cerebral effects. This study analyzed 26 infants subjects that included brain volumes from two neuroimaging modalities: DTI and sMRI. Twelve of the subjects involved children with normal nutrition in their early years of life, and 14 represented severe malnutrition cases. Segmentation was obtained by FreeSurfer from 94 and 103 regions for sMRI and DTI modalities, respectively. Each region was analyzed using several morphological and diffusion features; afterwards, seven classification models were trained to differentiate between the two groups. The results indicate notable differences that allow accurate classification, with an accuracy up to 1.0 and an AUROC of 1.0 for classification with SVM, LR, KNN, and MLP algorithms.