Soojin Lee1,2, Ahmed Gouda1,2, Saurabh Garg1,2,, Nasrin Akbari1,2, Saqib Basar1,2, Madhurima Datta1,2, Duc Nguyen1,2, and Sam Hashemi1,2
1 Vigilance Health Imaging Network Inc, Vancouver, Canada
2 Prenuvo Inc, Vancouver, Canada
Purpose:
While BMI is commonly used to determine obesity, it does not account for body composition and relies solely on body weight and height. This study investigates the impact of body composition on brain health using quantitative metrics from AI models applied to whole-body MR imaging.
Materials and Methods:
We used 3D nnU-Net segmentation models to analyze 1.5T whole-body MRI scans from 2,839 participants across the US and Canada. Total skeletal muscle mass percentage (SMP) and fat mass percentage (FMP) were derived by converting segmentation volumes to mass and normalizing by weight. Volume- and thickness-based 97 brain metrics were normalized by intracranial volume.
We matched 300 pairs of participants for age, sex, height, BMI, type 2 diabetes, and hypertension, but differing in SMP by over 5%. Participants with higher SMP were classified into the HighSMP group, while their matched counterparts were placed in the LowSMP group. The two groups were compared using two-sample t-tests with false discovery rate correction. We used multiple linear regression to examine brain metrics in relation to SMP and FMP, controlling for the aforementioned participants’ characteristics as well as pack year and alcohol intake.
Results:
The HighSMP and LowSMP groups differ significantly not only in body composition characteristics but also in 48.5% of brain metrics. The LowSMP group had reduced total brain (p < .001) and hippocampus (p < .001) volumes and increased inferior lateral ventricle volume (p < .001). Cortical thickness, particularly in the temporal lobe, was also lower in the LowSMP group (p < .001). These differences were consistent in both normal and overweight participants. Regression analysis showed brain volumes and thickness positively associated with SMP and negatively with FMP, with the effect size of SMP being around four times greater.
Conclusion:
Even with the same BMI, body composition significantly impacts brain volumes and cortical thickness. Higher SMP is linked to greater brain volume and cortical thickness, regardless of BMI categories. The influence of skeletal muscle mass on brain metrics surpasses that of fat mass.
Clinical Relevance:
AI-based automated segmentation holds significant potential for identifying cross-organ correlations. Our study demonstrates that body composition offers crucial insights, surpassing the traditionally used BMI measure in assessing brain health.
Fig. 1. (A) Characteristics of participants in the HighSMP and LowSMP groups. (B) Top: The LowSMP group exhibits atrophy, indicated by reduced total brain (p < 0.001) and hippocampus (p < 0.001) volumes and increased inferior lateral ventricle (p < 0.001) size across all BMI categories. Bottom: t-statistic maps comparing cortical thickness between the HighSMP and LowSMP groups. Positive t-statistics indicate greater cortical thickness in the HighSMP group. Regions of interest (ROIs) with p < 0.001 are highlighted in color. The names of ROIs that are significant across weight categories (healthy weight, overweight) are provided.
