Obesity prediction from structural MRI using conformal deep learning with uncertainty quantification

Obesity arises from a neurobehavioral disorder in which the brain’s regulation of hunger and food intake is impaired, leading to an imbalance between energy consumption and expenditure. One key phenotype associated with obesity is body mass index (BMI). BMI is influenced by multiple causal pathways driven by behavioral, metabolic, and genetic factors. Traditional obesity prediction studies often rely on magnetic resonance imaging (MRI) voxel-based morphometry to correlate BMI with obesity-related clinical measurements and brain structure, predominantly gray matter volume (GMV). However, the altered brain regions are variable and widespread between studies, with some literature presenting contradictory results between BMI and GMV within certain brain regions. Therefore, it is crucial to quantify uncertainty in the analysis to provide user-defined confidence, interpretability in the prediction. To address these limitations, we propose a computational model to predict obesity directly from individual T1-weighted structural MRI (sMRI) data. The proposed conformal deep learning (CDL) model achieves a 5-fold cross validated average precision of 77.65% and an F1-score of 75.42%, effectively classifying a patient as obese or healthy with uncertainty score from sMRI. Furthermore, our model provides probabilistic uncertainty quantification paired with gradient-based localization maps that discover key brain regions, such as frontal lobe, caudate, and thalamus from the individual sMRI of patient. Comparisons with standard probabilistic subcortical atlas and existing literature demonstrate that our CDL model offers complementary evidence linking obesity with subcortical regions. The findings highlight specific brain regions linked to obese individuals, with potential towards developing brain region-focused behavioral therapy designs to control impulsivity for individuals who are overweight or obese.