Background

The WHO has issued dietary guidelines for healthy weight management, but obesity continues to constitute a pandemic with increasing incidence rates. Various studies have shown how different diets can impact brain function, such as associations between BMI and increased connectivity in the brain’s extended reward network. The aim of this study was to apply machine learning analyses to functional resting state connectivity based on different diets (American, Mediterranean, Vegetarian) to predict obesity.

Methods

185 subjects were recruited. Binomial tests and bootstrapping methods were applied to test for differences in BMI based on diet group. Principal component, neural network, and random forest analyses were applied to fMRI brain data and important features were chosen for prediction of obesity using diet groups.

Results

Diet and BMI Differences: The individuals on the Mediterranean Diet had the lowest BMI (28.15) compared to those on Vegetarian (28.25) and Western (30.02) diets. Dietary choices resulted in difference of BMI means (p=.009). Brain Signatures: 1440 brain features were selected from the original 13533 using domain knowledge; PCA revealed 20 principal components that explained 56.1% variance. Classification of brain data based on diet achieved 73.2% prediction using multinomial logistic regression, 69.6% prediction using neural network, and 73.2% prediction using random forest. Loadings of the important features were associated with connectivity of the putamen to somatosensory regions on PC1 and the caudate nucleus to anterior insula and emotional regulation regions.

Conclusions

Our results indicate that those individuals on the Western diet had higher BMIs and showed distinct brain signature profiles involving reward regions with connectivity to salience, emotional regulation, and somatosensory regions. These results highlight the relationship between dopamine rich brain regions and a diet rich in meat and fat in contributing to obesity.