What does this research mean for the field?

An open-source deep-learning pipeline enables scalable, pooled MRI body-composition phenotyping across diverse cohorts, capturing fat and muscle distribution variations that improve cardiometabolic disease discrimination beyond conventional anthropometry. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

Determine whether a unified, open-source deep-learning tool can accurately quantify regional body composition and cardiometabolic disease risk across pooled multi-cohort MRI scans with varying acquisition protocols.

August 30, 2026Communications MedicineOpen Access

Pooled two-cohort MRI body composition phenotyping with open-source deep learning

View Full Paper

Ask AI

Bookmark

Authors

CMChristian Mertens HHHartmut Häntze SZSebastian Ziegelmayer

Discussion

Member takes

Overview

Cohort study reveals distinct fat-distribution phenotypes for cardiometabolic disease in 45,851 adults, indicating that deep-learning MRI quantification adds value beyond body mass index.

Key Points

Determine whether a unified, open-source deep-learning tool can accurately quantify regional body composition and cardiometabolic disease risk across pooled multi-cohort MRI scans with varying acquisition protocols.
Applied MRSegmentator, an open-source nnU-Net pipeline, to quantify visceral, subcutaneous, and gluteofemoral adipose tissue, trunk muscle, and liver fat in 45,851 adults from the German National Cohort (n = 26,877, 3 T Siemens) and UK Biobank (n = 18,974, 1.5 T Siemens).
Trained a single pooled model on curated reference datasets without site-specific adaptations and evaluated inter-reader agreement using a separate 50-scan reader study.
Modeled associations between BMI-adjusted body composition compartments and cardiometabolic conditions using generalized linear mixed-effects models to assess incremental risk discrimination.
The pipeline achieved a mean Dice similarity coefficient of 0.91 in 5-fold cross-validation, with reader–reader Dice of 0.937 and algorithm–reader Dice of 0.908 across the 50-scan evaluation set.
Visceral adipose tissue displayed the strongest positive risk associations with cardiometabolic diseases, whereas gluteofemoral adipose tissue showed an inverse association, most prominently for type 2 diabetes (OR 0.69, 95% CI 0.66 to 0.72).
Type 2 diabetes was specifically characterized by elevated visceral fat, reduced gluteofemoral fat, and increased liver fat, providing modest discrimination improvements over standard anthropometry.

Cite This Study

Mertens et al. (2026) studied this question.

synapsesocial.com/papers/6a93f1626c1a8fb52e79e4ff https://doi.org/10.1038/s43856-026-01888-w

View Full Paper

Ask AI

Bookmark

August 30, 2026Communications MedicineOpen Access

Pooled two-cohort MRI body composition phenotyping with open-source deep learning

View Full Paper

Ask AI

Bookmark

Authors

CMChristian Mertens HHHartmut Häntze SZSebastian Ziegelmayer

Discussion

Member takes

Overview

Cohort study reveals distinct fat-distribution phenotypes for cardiometabolic disease in 45,851 adults, indicating that deep-learning MRI quantification adds value beyond body mass index.

Key Points

Determine whether a unified, open-source deep-learning tool can accurately quantify regional body composition and cardiometabolic disease risk across pooled multi-cohort MRI scans with varying acquisition protocols.
Applied MRSegmentator, an open-source nnU-Net pipeline, to quantify visceral, subcutaneous, and gluteofemoral adipose tissue, trunk muscle, and liver fat in 45,851 adults from the German National Cohort (n = 26,877, 3 T Siemens) and UK Biobank (n = 18,974, 1.5 T Siemens).
Trained a single pooled model on curated reference datasets without site-specific adaptations and evaluated inter-reader agreement using a separate 50-scan reader study.
Modeled associations between BMI-adjusted body composition compartments and cardiometabolic conditions using generalized linear mixed-effects models to assess incremental risk discrimination.
The pipeline achieved a mean Dice similarity coefficient of 0.91 in 5-fold cross-validation, with reader–reader Dice of 0.937 and algorithm–reader Dice of 0.908 across the 50-scan evaluation set.
Visceral adipose tissue displayed the strongest positive risk associations with cardiometabolic diseases, whereas gluteofemoral adipose tissue showed an inverse association, most prominently for type 2 diabetes (OR 0.69, 95% CI 0.66 to 0.72).
Type 2 diabetes was specifically characterized by elevated visceral fat, reduced gluteofemoral fat, and increased liver fat, providing modest discrimination improvements over standard anthropometry.

Cite This Study

Mertens et al. (2026) studied this question.

synapsesocial.com/papers/6a93f1626c1a8fb52e79e4ff https://doi.org/10.1038/s43856-026-01888-w

View Full Paper

Ask AI

Bookmark