Share：
Share this content in WeChat
X
[Chinese] [PDF] 18 1
• Review •
Advances in deep learning-based body composition analysis with magnetic resonance imaging
LIANG Yongzhao  WANG Yiou  ZHANG Yu  ZHANG Xiaodong 

Cite this article as: LIANG Y Z, WANG Y O, ZHANG Y, et al. Advances in deep learning-based body composition analysis with magnetic resonance imaging[J]. Chin J Magn Reson Imaging, 2026, 17(9): 216-223, 234. DOI:10.12015/issn.1674-8034.2026.09.028.


[Abstract] Deep learning is transforming magnetic resonance imaging (MRI)-based body composition analysis from manual delineation to automated segmentation and high-throughput quantification. MRI can characterize subcutaneous and visceral adipose tissue, skeletal muscle volume, and muscular fat infiltration, providing imaging biomarkers for metabolic risk, sarcopenia, prognosis, and treatment monitoring. This review surveys studies published from January 2016 to July 2026 and synthesizes the principles and evidence for convolutional neural networks, U-Net/nnU-Net, transformers, generative approaches, and vision foundation models. Particular attention is paid to adipose-tissue and skeletal-muscle segmentation, downstream quantitative error, scan-rescan repeatability, cross-center generalization, and clinical validity. Although many models achieve high Dice scores and markedly reduce analysis time in selected datasets, segmentation accuracy alone does not establish clinical utility. Single-center training, pathological fatty infiltration, label noise, inconsistent evaluation, and limited external validation remain major barriers. Future studies should establish standardized multicenter, multisequence, and multidisease datasets; report volume or cross-sectional-area error, fat-fraction bias, repeatability, failure rate, and uncertainty in addition to Dice; and integrate automated quality control with human review before routine clinical deployment.
[Keywords] metabolic diseases；magnetic resonance imaging；deep learning；body composition analysis；risk assessment

LIANG Yongzhao1   WANG Yiou2   ZHANG Yu3   ZHANG Xiaodong2*  

1 The First Clinical Medical School of Southern Medical University, Foshan 528300, China

2 Department of Medical Imaging, the Third Affiliated Hospital of Southern Medical University (Guangdong Provincial Orthopedic Academy), Guangzhou 510630, China

3 School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China

Corresponding author: ZHANG X D, E-mail: ddautumn@126.com

Conflicts of interest   None.

Received  2026-03-11
Accepted  2026-08-03
DOI: 10.12015/issn.1674-8034.2026.09.028
Cite this article as: LIANG Y Z, WANG Y O, ZHANG Y, et al. Advances in deep learning-based body composition analysis with magnetic resonance imaging[J]. Chin J Magn Reson Imaging, 2026, 17(9): 216-223, 234. DOI:10.12015/issn.1674-8034.2026.09.028.

[1]
RADOVIC A H, HAJRO S, DURMISEVIC A J, et al. Age-stratified characterization of body composition and biochemical profiles in cardiometabolic risk assessment[J]. Med Arch, 2025, 79(5): 353-357. DOI: 10.5455/medarh.2025.79.353-357.
[2]
BRANTLOV S, JØDAL L, HVAS C L, et al. Metrology in bioelectrical impedance analysis (BIA): from measurement science to clinical and research applications[J/OL]. Sensors (Basel), 2026, 26(13): 4017 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/42451262/. DOI: 10.3390/s26134017.
[3]
TINSLEY G M, NICKERSON B S. Exclusion of trunk region reduces biological error but increases technical error of DXA lean soft tissue estimates from nonfasted assessments[J]. Int J Sport Nutr Exerc Metab, 2019, 29(3): 309-314. DOI: 10.1123/ijsnem.2018-0132.
[4]
BASTY N, THANAJ M, WHITCHER B, et al. Comparing DXA and MRI body composition measurements in cross-sectional and longitudinal cohorts[J/OL]. Commun Med, 2026, 6: 227 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/41786921/. DOI: 10.1038/s43856-026-01440-w.
[5]
CHEN C, MAT ISA N A, LIU X. A review of convolutional neural network based methods for medical image classification[J/OL]. Comput Biol Med, 2025, 185: 109507 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/39631108/. DOI: 10.1016/j.compbiomed.2024.109507.
[6]
ESTEVA A, ROBICQUET A, RAMSUNDAR B, et al. A guide to deep learning in healthcare[J]. Nat Med, 2019, 25(1): 24-29. DOI: 10.1038/s41591-018-0316-z.
[7]
GAO Y X, JIANG Y, PENG Y H, et al. Medical image segmentation: a comprehensive review of deep learning-based methods[J/OL]. Tomography, 2025, 11(5): 52 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40423254/. DOI: 10.3390/tomography11050052.
[8]
WEI J Y, CHEN H L, YAO L J, et al. BioCompNet: a deep learning workflow enabling automated body composition analysis toward precision management of cardiometabolic disorders[J/OL]. Cyborg Bionic Syst, 2025, 6: 0381 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40842914/. DOI: 10.34133/cbsystems.0381.
[9]
JUNG M, RAGHU V K, REISERT M, et al. Deep learning-based body composition analysis from whole-body magnetic resonance imaging to predict all-cause mortality in a large western population[J/OL]. EBioMedicine, 2024, 110: 105467 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/39622188/. DOI: 10.1016/j.ebiom.2024.105467.
[10]
JUNG M, RIEDER H, REISERT M, et al. Association between myosteatosis and impaired glucose metabolism: a deep learning whole-body magnetic resonance imaging population phenotyping approach[J]. J Cachexia Sarcopenia Muscle, 2024, 15(5): 1750-1760. DOI: 10.1002/jcsm.13527.
[11]
LINDER N, DENECKE T, BUSSE H. Body composition analysis by radiological imaging-methods, applications, and prospects[J]. Rofo, 2024, 196(10): 1046-1054. DOI: 10.1055/a-2263-1501.
[12]
WINDER C, CLARK M, FROOD R, et al. Automated extraction of body composition metrics from abdominal CT or MR imaging: a scoping review[J/OL]. Eur J Radiol, 2024, 181: 111764 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/39368243/. DOI: 10.1016/j.ejrad.2024.111764.
[13]
TEJANI A S, KLONTZAS M E, GATTI A A, et al. Checklist for artificial intelligence in medical imaging (CLAIM): 2024 update[J/OL]. Radiol Artif Intell, 2024, 6(4): e240300 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/38809149/. DOI: 10.1148/ryai.240300.
[14]
COLLINS G S, MOONS K G M, DHIMAN P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods[J/OL]. BMJ, 2024, 385: e078378 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/38626948/. DOI: 10.1136/bmj-2023-078378.
[15]
MOONS K G M, DAMEN J A A, KAUL T, et al. PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods[J/OL]. BMJ, 2025, 388: e082505 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40127903/. DOI: 10.1136/bmj-2024-082505.
[16]
HUANG S Q, ZHANG D M, WANG Q, et al. Advances in the application of fat quantification techniques based on CT and MRI in general population cohorts[J]. Chin J Magn Reson Imaging, 2026, 17(2): 228-234. DOI: 10.12015/issn.1674-8034.2026.02.034.
[17]
WOHLFARTH V, KWAY Y M, SOOD A, et al. Deep learning-based framework for comprehensive quantification of thigh and calf muscles and adipose tissues from MRI[J]. Magn Reson Mater Phys Biol Med, 2026, 39(2): 305-318. DOI: 10.1007/s10334-025-01308-5.
[18]
CRUZ-JENTOFT A J, BAHAT G, BAUER J, et al. Sarcopenia: revised European consensus on definition and diagnosis[J]. Age Ageing, 2019, 48(1): 16-31. DOI: 10.1093/ageing/afy169.
[19]
YAO R H, GAO B L, HU J H, et al. Characteristics of mDIXON-Quant and proton magnetic resonance spectroscopy imaging of thigh skeletal muscles in patients with stroke-related sarcopenia: a prospective case-control study[J/OL]. J Yeungnam Med Sci, 2025, 42: 49 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40874399/. DOI: 10.12701/jyms.2025.42.49.
[20]
MENG X Y, LI Z Y, LI H. Magnetic resonance imaging in evaluating fat content quantitatively: a review[J]. J China Clin Med Imaging, 2018, 29(5): 362-365.
[21]
ASSI A, RAMADAN B, KARAM M, et al. Muscle fat infiltration: a narrative review of the magnetic resonance (MR)-based evaluation methods and their clinical applications[J]. Quant Imaging Med Surg, 2024, 14(12): 9563-9588. DOI: 10.21037/qims-24-177.
[22]
KAFALI S G, SHIH S F, LI X Z, et al. Automated abdominal adipose tissue segmentation and volume quantification on longitudinal MRI using 3D convolutional neural networks with multi-contrast inputs[J]. MAGMA, 2024, 37(3): 491-506. DOI: 10.1007/s10334-023-01146-3.
[23]
DING J, CAO P, CHANG H C, et al. Deep learning-based thigh muscle segmentation for reproducible fat fraction quantification using fat-water decomposition MRI[J/OL]. Insights Imaging, 2020, 11(1): 128 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/33252711/. DOI: 10.1186/s13244-020-00946-8.
[24]
YU A C, MOHAJER B, ENG J. External validation of deep learning algorithms for radiologic diagnosis: a systematic review[J/OL]. Radiol Artif Intell, 2022, 4(3): e210064 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/35652114/. DOI: 10.1148/ryai.210064.
[25]
YANG Q Q, SHOMAL-ZADEH F, GHOLIPOUR A. Harmonization in magnetic resonance imaging: a survey of acquisition, image-level, and feature-level methods[J/OL]. Med Image Anal, 2026, 111: 104066 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/41936201/. DOI: 10.1016/j.media.2026.104066.
[26]
HAUBOLD J, POLLOK O B, HOLTKAMP M, et al. Moving beyond CT body composition analysis: using style transfer for bringing CT-based fully-automated body composition analysis to T2-weighted MRI sequences[J]. Invest Radiol, 2025, 60(8): 552-559. DOI: 10.1097/RLI.0000000000001162.
[27]
ISENSEE F, JAEGER P F, KOHL S A A, et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation[J]. Nat Meth, 2021, 18(2): 203-211. DOI: 10.1038/s41592-020-01008-z.
[28]
GRAF R, PLATZEK P, RIEDEL E O, et al. VIBESegmentator: full body MRI segmentation for the NAKO and UK Biobank[J]. Eur Radiol, 2026, 36(4): 2548-2562. DOI: 10.1007/s00330-025-12035-9.
[29]
CASALI N, BRUSAFERRI A, BASELLI G, et al. Exploring foundation models for multi-class muscle segmentation in MR images of neuromuscular disorders: a comparative analysis of accuracy and uncertainty[J/OL]. Comput Meth Programs Biomed, 2025, 272: 109035 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40907134/. DOI: 10.1016/j.cmpb.2025.109035.
[30]
LANGNER T, GUSTAFSSON F K, AVELIN B, et al. Uncertainty-aware body composition analysis with deep regression ensembles on UK Biobank MRI[J/OL]. Comput Med Imaging Graph, 2021, 93: 101994 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/34624770/. DOI: 10.1016/j.compmedimag.2021.101994.
[31]
WHITCHER B, THANAJ M, CULE M, et al. Precision MRI phenotyping enables detection of small changes in body composition for longitudinal cohorts[J/OL]. Sci Rep, 2022, 12(1): 3748 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/35260612/. DOI: 10.1038/s41598-022-07556-y.
[32]
SCHNEIDER D, EGGEBRECHT T, LINDER A, et al. Abdominal fat quantification using convolutional networks[J]. Eur Radiol, 2023, 33(12): 8957-8964. DOI: 10.1007/s00330-023-09865-w.
[33]
NOWAK M, NÚÑEZ L, HILL C E, et al. Automated volumetric MRI quantification of body fat and skeletal muscle in the UK Biobank: technical validation against expert manual segmentations and comparison with single-slice techniques[J]. Abdom Radiol (NY), 2026, 51(4): 2180-2191. DOI: 10.1007/s00261-025-05170-w.
[34]
LIN D Y, WANG Z Y, LI H, et al. Automated measurement of pancreatic fat deposition on Dixon MRI using nnU-net[J]. J Magn Reson Imaging, 2023, 57(1): 296-307. DOI: 10.1002/jmri.28275.
[35]
LANGNER T, MARTÍNEZ MORA A, STRAND R, et al. MIMIR: deep regression for automated analysis of UK biobank MRI scans[J/OL]. Radiol Artif Intell, 2022, 4(3): e210178 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/35652115/. DOI: 10.1148/ryai.210178.
[36]
KWAY Y M, THIRUMURUGAN K, MICHAEL N, et al. A fully convolutional neural network for comprehensive compartmentalization of abdominal adipose tissue compartments in MRI[J/OL]. Comput Biol Med, 2023, 167: 107608 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/37897959/. DOI: 10.1016/j.compbiomed.2023.107608.
[37]
CHEN S Z, AN D, FENG C L, et al. Segmentation of pericardial adipose tissue in CMR images: a benchmark dataset MRPEAT and a triple-stage network 3SUnet[J]. IEEE Trans Med Imaging, 2023, 42(8): 2386-2399. DOI: 10.1109/TMI.2023.3251368.
[38]
GAJ S, ECK B L, XIE D X, et al. Deep learning-based automatic pipeline for quantitative assessment of thigh muscle morphology and fatty infiltration[J]. Magn Reson Med, 2023, 89(6): 2441-2455. DOI: 10.1002/mrm.29599.
[39]
HOSTIN M A, OGIER A C, MICHEL C P, et al. The impact of fatty infiltration on MRI segmentation of lower limb muscles in neuromuscular diseases: a comparative study of deep learning approaches[J]. J Magn Reson Imaging, 2023, 58(6): 1826-1835. DOI: 10.1002/jmri.28708.
[40]
MARTIN S, ANDRÉ R, TRABELSI A, et al. Importance of neural network complexity for the automatic segmentation of individual thigh muscles in MRI images from patients with neuromuscular diseases[J]. Magn Reson Mater Phys Biol Med, 2025, 38(2): 175-189. DOI: 10.1007/s10334-024-01221-3.
[41]
AGOSTI A, SHAQIRI E, PAOLETTI M, et al. Deep learning for automatic segmentation of thigh and leg muscles[J]. Magn Reson Mater Phys Biol Med, 2022, 35(3): 467-483. DOI: 10.1007/s10334-021-00967-4.
[42]
NIEMEYER F, ZANKER A, JONAS R, et al. An externally validated deep learning model for the accurate segmentation of the lumbar paravertebral muscles[J]. Eur Spine J, 2022, 31(8): 2156-2164. DOI: 10.1007/s00586-022-07320-w.
[43]
BALERDI G, HENCKEL J, LAURA A D, et al. Automated deep-learning quantification of intramuscular fat in lumbar spine muscles on Dixon MRI: validation and normative reference values from 173 healthy adults[J/OL]. BMC Med Imaging, 2026, 26(1): 220 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/41917883/. DOI: 10.1186/s12880-026-02329-9.
[44]
KIM H S, PARK H, KANG J, et al. Neck-to-knee Dixon MRI thigh volume as a superior mass biomarker for Sarcopenia: evidence from the UK biobank[J/OL]. NPJ Digit Med, 2026, 9(1): 239 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/41644627/. DOI: 10.1038/s41746-026-02379-x.
[45]
ESPARZA J P, AHMED M M, SHRESTHA U, et al. A dual-contrast deep learning model for automated assessment of sarcopenia using MRI[J/OL]. Abdom Radiol, 2026 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/42362936/. DOI: 10.1007/s00261-026-05655-2.
[46]
SLUTSKY-GANESH A B, et al. Optimizing MRI annotation workflows for high-accuracy deep learning thigh muscle segmentation in athletes[J/OL]. Radiol Adv, 2026, 3: umag005 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/41694457/. DOI: 10.1093/radadv/umag005.
[47]
MEDINA G, BUCKLESS C G, THOMASSON E, et al. Deep learning method for segmentation of rotator cuff muscles on MR images[J]. Skeletal Radiol, 2021, 50(4): 683-692. DOI: 10.1007/s00256-020-03599-2.
[48]
ZHU J Y, BOLSTERLEE B, CHOW B V Y, et al. Deep learning methods for automatic segmentation of lower leg muscles and bones from MRI scans of children with and without cerebral palsy[J/OL]. NMR Biomed, 2021, 34(12): e4609 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/34545647/. DOI: 10.1002/nbm.4609.
[49]
WEBER K A, ABBOTT R, BOJILOV V, et al. Multi-muscle deep learning segmentation to Automate the quantification of muscle fat infiltration in cervical spine conditions[J/OL]. Sci Rep, 2021, 11(1): 16567 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/34400672/. DOI: 10.1038/s41598-021-95972-x.
[50]
JUNG M, REISERT M, RIEDER H, et al. Body composition in the general population: whole-body MRI-derived reference curves from over 66 000 individuals[J/OL]. Radiology, 2026, 319(2): e251939 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/42084507/. DOI: 10.1148/radiol.251939.
[51]
TAKAHASHI S, SAKAGUCHI Y, KOUNO N, et al. Comparison of vision transformers and convolutional neural networks in medical image analysis: a systematic review[J/OL]. J Med Syst, 2024, 48(1): 84 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/39264388/. DOI: 10.1007/s10916-024-02105-8.
[52]
THOMAS B, ALI M A, ALI F M H, et al. Deep learning models for deriving optimised measures of fat and muscle mass from MRI[J/OL]. Sci Rep, 2025, 15: 25960 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40676073/. DOI: 10.1038/s41598-025-07867-w.
[53]
MAHDI M A, AL-SHALABI M, ALNFRAWY E T, et al. 3D latent diffusion model for MR-only radiotherapy: accurate and consistent synthetic CT generation[J/OL]. Diagnostics (Basel), 2025, 15(23): 3010 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/41374391/. DOI: 10.3390/diagnostics15233010.
[54]
MÜLLER D, SOTO-REY I, KRAMER F. Towards a guideline for evaluation metrics in medical image segmentation[J/OL]. BMC Res Notes, 2022, 15(1): 210 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/35725483/. DOI: 10.1186/s13104-022-06096-y.
[55]
ROBINSON R, VALINDRIA V V, BAI W J, et al. Automated quality control in image segmentation: application to the UK Biobank cardiovascular magnetic resonance imaging study[J/OL]. J Cardiovasc Magn Reson, 2019, 21(1): 18 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/30866968/. DOI: 10.1186/s12968-019-0523-x.
[56]
LEKADIR K, FRANGI A F, PORRAS A R, et al. FUTURE-AI: international consensus guideline for trustworthy and deployable artificial intelligence in healthcare[J/OL]. BMJ, 2025, 388: e081554 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/39909534/. DOI: 10.1136/bmj-2024-081554.
[57]
KASSANI S H, PATEL K, COMMEAN P K, et al. Vision transformer-based segmentation of abdominal subcutaneous and visceral fat on MRI[J/OL]. J Imaging Inform Med, 2026 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/42067725/. DOI: 10.1007/s10278-026-01970-8.
[58]
SALHI A, ITALIA K, VIEDMA I, et al. Trustworthy deep learning for the automated quantification of the fatty infiltration of the rotator cuff muscles using magnetic resonance imaging[J]. JSES Int, 2025, 9(6): 1999-2007. DOI: 10.1016/j.jseint.2025.06.020.
[59]
BIZZOZERO S, BASSANI T, SCONFIENZA L M, et al. Gender difference in cross-sectional area and fat infiltration of thigh muscles in the elderly population on MRI: an AI-based analysis[J/OL]. Eur Radiol Exp, 2025, 9(1): 64 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40624409/. DOI: 10.1186/s41747-025-00606-w.
[60]
LEE H J, KUO C Y, TSAO Y C, et al. Brain age gap associations with body composition and metabolic indices in an Asian cohort: an MRI-based study[J/OL]. Arch Gerontol Geriatr, 2025, 133: 105830 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/40127523/. DOI: 10.1016/j.archger.2025.105830.
[61]
RANKIN A G, MCSWEENEY D M, ZHONG J, et al. Assessing body composition using auto-segmentations of muscle and subcutaneous adipose tissue in prostate cancer patients receiving magnetic resonance-guided radiotherapy[J/OL]. Phys Imaging Radiat Oncol, 2025, 36: 100882 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/41476866/. DOI: 10.1016/j.phro.2025.100882.
[62]
GAO X, DUAN J H, CUI B B, et al. Impact of sleeve gastrectomy on ectopic fat deposition and skeletal muscle composition across glucose metabolic states[J]. Obes Surg, 2025, 35(7): 2689-2697. DOI: 10.1007/s11695-025-07940-3.
[63]
RAMEDANI S, KELESOGLU E, STUTZIG N, et al. Quantification of training-induced alterations in body composition via automated machine learning analysis of MRI images in the thigh region: a pilot study in young females[J/OL]. Physiol Rep, 2025, 13(3): e70187 [2026-03-10]. https://pubmed.ncbi.nlm.nih.gov/39878619/. DOI: 10.14814/phy2.70187.

PREV Research progress of magnetic resonance habitat imaging in common gynecological malignant tumors
NEXT Clinical application progress of imaging-based body composition analysis
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn