Body Composition

AI-based automated body composition analysis from whole-body MRI and (PET-) CT: an intermodal comparison from same-day examinations.

Holzschuh J, Rotkopf L, et al. • Cancer imaging : the official publication of the International Cancer Imaging Society • 2026

PubMed 42928255 DOI 10.1186/s40644-026-01134-5
TL;DR

Automated body composition measures from (PET-)CT and wb-MRI are highly concordant but show consistent intermodal biases that should be considered when pooling multimodal data or applying modality-agnostic thresholds.

Key Findings

Results

AI-derived volumetric body composition metrics from same-day low-dose CT and whole-body MRI showed high agreement for skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue.

  • 42 multiple myeloma patients with same-day [18F]FDG-PET/CT and wb-MRI were analyzed, drawn from the prospective GMMG-HD7 and GMMG-HD8/DSMM-XIX trials.
  • Segmentation used an automated deep learning pipeline based on TotalSegmentator and its MRI variant.
  • The abstract reports "high agreement (ICC > 0.95) across all SM, SAT and VAT" for volumetric measures.
  • Agreement was assessed using Pearson r, ICC, and Bland-Altman analyses.
Results

Volumetric skeletal muscle was systematically and significantly lower when measured from MRI than from CT.

  • Bland-Altman bias was -2582 ml (p < 0.0001) for MRI relative to CT.
Results

Volumetric abdominal visceral adipose tissue was systematically and significantly lower on MRI than on CT.

  • Bland-Altman bias was -211 ml (p < 0.0001).
Results

Volumetric subcutaneous adipose tissue showed no statistically significant systematic difference between MRI and CT.

  • Bias was -263 ml (p = 0.1629).
Results

Single-slice analysis showed high CT–MRI agreement for SAT and VAT but lower agreement for skeletal muscle.

  • ICC was > 0.97 for SAT and VAT.
  • ICC for skeletal muscle single-slice was 0.87.
Results

Single-slice skeletal muscle area was systematically and significantly lower on MRI than on CT.

  • Bland-Altman bias was -18.4 cm2 (p < 0.0001).
Conclusions

The authors conclude that intermodal biases should be considered when pooling multimodal data or applying modality-agnostic thresholds.

  • Despite high concordance, the biases were consistent between modalities.
  • This implies that CT- and MRI-derived body composition values may not be directly interchangeable.

What This Means

This study compared how well computer software can measure body composition (muscle and fat amounts) from two different scans taken on the same day: a low-dose CT scan (part of a PET/CT) and a whole-body MRI. It included 42 people with multiple myeloma, a blood cancer. The software, based on a deep learning tool called TotalSegmentator, outlined skeletal muscle, fat under the skin, and fat in the abdomen on both scan types. The two scan types agreed very closely overall (ICC above 0.95, meaning near-consistent ranking of patients). However, the MRI-based measurements gave systematically lower muscle volume (about 2.6 liters lower on average) and lower abdominal visceral fat (about 211 ml lower) than CT. Subcutaneous fat did not differ significantly. When looking at a single image slice rather than the whole body, fat measures still agreed well, but agreement for muscle was lower (ICC 0.87), with MRI giving a smaller muscle area on average. This research suggests that CT and MRI can both be used for automated body composition analysis, but their numbers are not directly interchangeable. Combining data from the two scan types, or applying a single cutoff value regardless of scan type, may introduce errors unless these consistent differences are taken into account. The study was small and limited to multiple myeloma patients, so results may differ in other groups.

Check Your Own Numbers

Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.

Upload Your Labs

Have a question about this study?

Presence or severity of Hashimoto's thyroiditis does not influence basal calcitonin levels: observations from CROHT biobank.
Journal of endocrinological investigation • 2022

HT patients of all disease stages preserve Ct production as healthy individuals and there is no need for Ct measurements...
Approach to the Patient: Safety of Growth Hormone Replacement in Children and Adolescents.
The Journal of clinical endocrinology and metabolism • 2022

rhGH therapy has been used for more than 3 decades and data are largely reassuring, yet we still have much to learn abou...
Extended Absorption of Liothyronine from Poly-Zinc-Liothyronine: Results from a Phase 1, Double-Blind, Randomized, and Controlled Study in Humans.
Thyroid : official journal of the American Thyroid Association • 2022

PZL possesses the necessary properties to achieve a much improved T3 pharmacokinetic profile, with ~30% lower Cmax that...
Lipoprotein subtypes after testosterone therapy in transmasculine adolescents.
Journal of clinical lipidology • 2022

Testosterone appears to be a major contributor to differences in lipoprotein profiles, a surrogate for cardiovascular di...
Long-term testosterone undecanoate treatment in the elderly testosterone deficient male: An observational cohort study.
Andrology • 2022

Long-acting testosterone undecanoate treatment in elderly testosterone-deficient patients required more frequent dose ad...
Society for Endocrinology guidelines for testosterone replacement therapy in male hypogonadism.
Clinical endocrinology • 2022

The Society for Endocrinology commissioned a multidisciplinary guideline to provide all care providers with a comprehens...

Citation

Holzschuh J, Rotkopf L, Neelsen C, Bauer F, Zhang K, Menis E, et al.. (2026). AI-based automated body composition analysis from whole-body MRI and (PET-) CT: an intermodal comparison from same-day examinations.. Cancer imaging : the official publication of the International Cancer Imaging Society. https://doi.org/10.1186/s40644-026-01134-5