pith. sign in

arxiv: 2605.30829 · v1 · pith:QHAI5MJ6new · submitted 2026-05-29 · 💻 cs.CV

LegSegNet: A Public Deep Learning System for Lower Extremity CT Tissue Segmentation and Quantification

Yuwen Chen , Yaqian Chen , Roy Colglazier , Haoyu Dong , Hanxue Gu , Maciej A. Mazurowski , Kevin W. Southerland This is my paper

Pith reviewed 2026-06-28 23:06 UTC · model grok-4.3

classification 💻 cs.CV
keywords lower extremity CTtissue segmentationdeep learningbody compositionsarcopenia assessmentmedical image analysisCT quantificationpublic model
0
0 comments X

The pith

LegSegNet is the first public end-to-end system that segments four key tissues in lower extremity CT scans and computes body composition metrics.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Lower extremity CT scans contain data useful for body composition analysis, sarcopenia assessment, and musculoskeletal monitoring, but existing public tools lack support for comprehensive tissue segmentation especially of inter/intramuscular adipose tissue and stop at mask output rather than full quantification. The paper builds and releases LegSegNet to take a CT scan as input, produce segmentations of bone, skeletal muscle, subcutaneous adipose tissue, and inter/intramuscular adipose tissue, then calculate quantitative measurements. The model was trained on 1,302 radiologist-reviewed annotated slices and tested on 900 held-out slices, where it outperforms a range of CNN, transformer, and foundation-model baselines with an average Dice score of 89.31 and shows generalization on an external public dataset. The authors make code and weights available so that researchers can use the complete workflow. A sympathetic reader would care because the release removes a barrier to scaling automated analysis of routine clinical scans.

Core claim

LegSegNet is a deep learning system that segments bone, skeletal muscle, subcutaneous adipose tissue, and inter/intramuscular adipose tissue from lower extremity CT scans and then derives quantitative tissue measurements for downstream clinical analysis. Trained on 1,302 manually annotated slices and evaluated on 900 held-out test slices with radiologist review, it records the highest average Dice score of 89.31 among tested 2D segmentation methods and generalizes to an external CT dataset. The authors state it is the first publicly released end-to-end system for this task.

What carries the argument

The LegSegNet segmentation model followed by a quantification step that converts predicted masks into tissue-specific measurements such as volume or area.

If this is right

  • Routine CT scans can be processed automatically to yield body composition numbers without manual contouring.
  • Large-scale studies of sarcopenia and musculoskeletal conditions become feasible using existing imaging archives.
  • Future computer vision methods in medical imaging gain a concrete public benchmark and starting point.
  • Clinical monitoring of tissue changes can rely on consistent, repeatable quantification outputs.
  • Other research groups can reproduce or extend the workflow using the released code and weights.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same segmentation-plus-quantification pattern could be retrained for CT scans of other body regions once suitable annotations exist.
  • Embedding the system in radiology reporting software might shorten the time needed for body composition reports in practice.
  • Performance on scans from scanners or patient populations outside the training distribution would need separate verification.
  • Longitudinal use on repeated scans from the same patient could track tissue changes over time if calibration remains stable.

Load-bearing premise

The 1,302 annotated slices and 900 test slices, along with their radiologist-reviewed labels, are representative of real-world lower extremity CT variability and accurately capture tissue boundaries.

What would settle it

An independent collection of lower extremity CT scans with fresh radiologist annotations on which LegSegNet produces average Dice scores well below 89.31 or quantification errors outside acceptable clinical limits.

Figures

Figures reproduced from arXiv: 2605.30829 by Hanxue Gu, Haoyu Dong, Kevin W. Southerland, Maciej A. Mazurowski, Roy Colglazier, Yaqian Chen, Yuwen Chen.

Figure 1
Figure 1. Figure 1: Dataset overview. Left: Example coronal CT view of the lower extremities with the annotated region indicated. Right: Demographics of the training and test sets, including sex distribu￾tion, CT scanner manufacturer, and age distribution. tal muscle (SM) includes the visible muscle of the thigh and calf. Inter/intramuscular adipose tissue (IAT) refers to fat located within the muscle, either between muscle g… view at source ↗
Figure 2
Figure 2. Figure 2: Overall pipeline of LegSegNet. The system takes lower extremity CT scans as input, performs automated tissue segmentation, and generates visualizations and quantitative body composition measurements for downstream analysis [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative segmentation results. Example CT slices from the test set with ground truth annotations and predictions from LegSegNet. Colors indicate different tissue types: bone (yellow), skeletal muscle (blue), subcutaneous adipose tissue (red), and in￾ter/intramuscular adipose tissue (orange) [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: Overview of LegSegNet system. LegSegNet takes lower extremity CT scans as input and outputs tissue masks, visu￾alizations, and body composition measurements for downstream analysis. 4.5. LegSegNet System As shown in [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 4
Figure 4. Figure 4: 3D visualization of predicted segmentation. Volumetric reconstruction of LegSegNet predictions. SAT VAT IAT Bone [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Correlation between predicted and ground-truth tissue volumes. Pearson correlation between tissue volumes computed from LegSegNet and ground truth volumes for SAT, SM, IAT, and Bone on the held-out test set. mentation within the selected region. The predicted masks are then used to generate visual outputs including slice-level overlays, as well as quantitative body composition measure￾ments such as tissue … view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative SAROS segmentation results. Example SAROS CT slices with ground truth annotations (SAT and SM) and predictions from LegSegNet. Class-wise performance. Bone segmentation is the most reliable target, likely because cortical bone has strong CT contrast and well-defined boundaries. In contrast, IAT re￾mains the most challenging class because it is relatively sparse and can have similar intensity to… view at source ↗
read the original abstract

Lower extremity computed tomography (CT) contains clinically relevant information for body composition analysis, sarcopenia assessment, and musculoskeletal disease monitoring, but extracting these measurements at scale requires accurate tissue segmentation and an automated quantification workflow. Existing public segmentation tools are not designed for comprehensive lower extremity CT analysis, particularly for clinically important inter/intramuscular adipose tissue, and most public methods only provide mask prediction rather than an end-to-end quantification system. To address this problem, we present LegSegNet, a deep learning system for lower extremity CT tissue segmentation and body composition quantification. Given an input CT scan, LegSegNet segments bone, skeletal muscle, subcutaneous adipose tissue, and inter/intramuscular adipose tissue. It then computes quantitative tissue measurements for downstream analysis. We developed the segmentation model using 1,302 manually annotated CT slices and evaluated it on 900 held-out test slices, with all annotations reviewed by radiologists. We benchmark LegSegNet against a broad set of 2D segmentation methods, including CNN-based models, transformer-based models, and finetuned foundation models, and further evaluate its generalization on an external public CT dataset. LegSegNet achieves the best overall segmentation performance, with an average Dice score of 89.31 on the held-out test set. To our knowledge, LegSegNet is the first publicly available end-to-end system for lower extremity CT tissue segmentation and quantification, providing a practical evaluation tool for future computer vision research in medical image analysis. The code and model weights are available at: https://github.com/mazurowski-lab/LegSegNet

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript presents LegSegNet, a deep learning system for segmenting bone, skeletal muscle, subcutaneous adipose tissue, and inter/intramuscular adipose tissue in lower extremity CT scans, followed by automated quantification of tissue volumes. The model is developed on 1,302 manually annotated slices (radiologist-reviewed) and evaluated on 900 held-out test slices, achieving an average Dice score of 89.31 while outperforming a broad set of 2D CNN, transformer, and finetuned foundation models; it also reports generalization on an external public dataset and releases code plus model weights, positioning itself as the first public end-to-end system for this task.

Significance. If the evaluation details are strengthened, the work would offer a practical, reproducible public tool for body composition and sarcopenia analysis from lower-extremity CT, filling a noted gap for comprehensive segmentation that includes inter/intramuscular adipose tissue. The public code release, broad benchmarking against multiple model families, and held-out evaluation are explicit strengths that support reproducibility.

major comments (3)
  1. [Abstract] Abstract: The headline claim of 'best overall segmentation performance' with average Dice 89.31 is presented without per-class Dice scores, error bars, or any statistical tests comparing against the benchmarked methods; this information is required to substantiate superiority and to allow readers to assess whether gains are uniform across the four tissue classes.
  2. [Abstract] Abstract: No information is supplied on the number of distinct patients or scanners in the 1,302 + 900 slices, whether the train/test split was performed at the patient level (versus slice level), or any inter-rater agreement statistics for the radiologist-reviewed annotations; these details are load-bearing for the assumption that the test set is representative and that reported Dice reflects model capability rather than annotation variability or selection bias.
  3. [Abstract] Abstract: The generalization experiment on an external public CT dataset is mentioned without quantitative results, adaptation details, or evaluation protocol; this omission weakens the robustness claim that is invoked to support the overall contribution.
minor comments (1)
  1. [Abstract] Abstract: The four tissue classes should be listed explicitly when stating the average Dice score so readers immediately understand which structures contribute to the reported metric.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. We agree that the abstract would be strengthened by incorporating additional details on per-class performance, dataset characteristics, and generalization results. We have revised the abstract accordingly, drawing from the detailed information already present in the Methods and Results sections of the manuscript. Point-by-point responses follow.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline claim of 'best overall segmentation performance' with average Dice 89.31 is presented without per-class Dice scores, error bars, or any statistical tests comparing against the benchmarked methods; this information is required to substantiate superiority and to allow readers to assess whether gains are uniform across the four tissue classes.

    Authors: We agree that the abstract would benefit from more granular information to support the superiority claim. The main text already contains per-class Dice scores for LegSegNet and all benchmarked methods, along with error bars (standard deviations across slices) and statistical comparisons. In the revised manuscript we have updated the abstract to note that LegSegNet achieves the highest scores across all four tissue classes with statistically significant improvements, explicitly directing readers to the full per-class results and tests in the Results section. revision: yes

  2. Referee: [Abstract] Abstract: No information is supplied on the number of distinct patients or scanners in the 1,302 + 900 slices, whether the train/test split was performed at the patient level (versus slice level), or any inter-rater agreement statistics for the radiologist-reviewed annotations; these details are load-bearing for the assumption that the test set is representative and that reported Dice reflects model capability rather than annotation variability or selection bias.

    Authors: These details are essential for evaluating the robustness of the reported results. The Methods section of the manuscript already specifies the number of patients and scanners, confirms that the split was performed at the patient level, and reports inter-rater agreement statistics on the annotations. We have revised the abstract to include concise statements summarizing these aspects so that readers can immediately assess the evaluation design without needing to consult the main text. revision: yes

  3. Referee: [Abstract] Abstract: The generalization experiment on an external public CT dataset is mentioned without quantitative results, adaptation details, or evaluation protocol; this omission weakens the robustness claim that is invoked to support the overall contribution.

    Authors: We concur that quantitative support is needed in the abstract to back the generalization claim. The Results section already provides the quantitative Dice scores on the external dataset, describes the adaptation approach, and outlines the evaluation protocol. We have updated the abstract to incorporate the key quantitative outcome and a brief description of the protocol and adaptation method used. revision: yes

Circularity Check

0 steps flagged

No circularity: standard held-out evaluation on independent test slices.

full rationale

The paper reports a conventional supervised segmentation workflow: model developed on 1,302 annotated slices, evaluated on 900 explicitly held-out test slices, with Dice 89.31 reported on the test set. No equations, fitted parameters, self-citations, or ansatzes are invoked that would reduce the reported metric to a definition or input by construction. The central performance claim rests on external test data and is therefore independent of the training inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim of superior performance and first-public status rests on the representativeness of the 1,302/900 annotated slice split and on the assumption that radiologist-reviewed labels constitute reliable ground truth; the model itself contains many standard deep-learning hyperparameters whose values are not enumerated in the abstract.

free parameters (1)
  • neural-network architecture and training hyperparameters
    The segmentation model contains numerous architecture choices and optimization settings that were selected or tuned to reach the reported Dice score.
axioms (1)
  • domain assumption Radiologist-reviewed manual annotations accurately delineate true tissue boundaries in the CT slices
    All reported Dice scores and the claim of best performance depend on the fidelity of these labels.

pith-pipeline@v0.9.1-grok · 5845 in / 1426 out tokens · 36057 ms · 2026-06-28T23:06:18.872531+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    Moco-transfer: Investigating out-of-distribution contrastive learning for limited-data domains.arXiv preprint arXiv:2311.09401, 2023

    Yuwen Chen, Helen Zhou, and Zachary C Lipton. Moco-transfer: Investigating out-of-distribution contrastive learning for limited-data domains.arXiv preprint arXiv:2311.09401, 2023. 1

    work page arXiv 2023

  2. [2]

    Cao, Adrian Camarena, Christopher Mantyh, Roy Colglazier, and Maciej A

    Yaqian Chen, Hanxue Gu, Yuwen Chen, Jicheng Yang, Haoyu Dong, Joseph Y . Cao, Adrian Camarena, Christopher Mantyh, Roy Colglazier, and Maciej A. Mazurowski. Au- tomated muscle and fat segmentation in computed tomogra- phy for comprehensive body composition analysis.Machine Learning for Biomedical Imaging, 3:581–618, 2025. 1, 2

    2025

  3. [3]

    Mazurowski

    Yuwen Chen, Nicholas Konz, Hanxue Gu, Haoyu Dong, Yaqian Chen, Lin Li, Jisoo Lee, and Maciej A. Mazurowski. Contourdiff: Unpaired medical image translation with struc- tural consistency.Machine Learning for Biomedical Imag- ing, 3:711–727, 2025. 1

    2025

  4. [4]

    Segmentanymuscle: A universal muscle segmentation model across different locations in mri

    Roy Colglazier, Jisoo Lee, Haoyu Dong, Hanxue Gu, Yaqian Chen, Joseph Cao, Zafer Yildiz, Zhonghao Liu, Nicholas Konz, Jichen Yang, et al. Segmentanymuscle: A universal muscle segmentation model across different locations in mri. arXiv preprint arXiv:2506.22467, 2025. 1

    work page arXiv 2025

  5. [5]

    Sar- copenia: revised european consensus on definition and diag- nosis.Age and ageing, 48(1):16–31, 2019

    Alfonso J Cruz-Jentoft, G ¨ulistan Bahat, J ¨urgen Bauer, Yves Boirie, Olivier Bruy`ere, Tommy Cederholm, Cyrus Cooper, Francesco Landi, Yves Rolland, Avan Aihie Sayer, et al. Sar- copenia: revised european consensus on definition and diag- nosis.Age and ageing, 48(1):16–31, 2019. 1, 2

    2019

  6. [6]

    Mri- core: a foundation model for magnetic resonance imaging

    Haoyu Dong, Yuwen Chen, Hanxue Gu, Nicholas Konz, Yaqian Chen, Qihang Li, and Maciej A Mazurowski. Mri- core: a foundation model for magnetic resonance imaging. arXiv preprint arXiv:2506.12186, 2025. 1

    work page arXiv 2025

  7. [7]

    Quantitative analysis of skeletal muscle by com- puted tomography imaging—state of the art.Journal of or- thopaedic translation, 15:91–103, 2018

    Klaus Engelke, Oleg Museyko, Ling Wang, and Jean-Denis Laredo. Quantitative analysis of skeletal muscle by com- puted tomography imaging—state of the art.Journal of or- thopaedic translation, 15:91–103, 2018. 1, 2

    2018

  8. [8]

    3d slicer as an image comput- ing platform for the quantitative imaging network.Magnetic resonance imaging, 30(9):1323–1341, 2012

    Andriy Fedorov, Reinhard Beichel, Jayashree Kalpathy- Cramer, Julien Finet, Jean-Christophe Fillion-Robin, Sonia Pujol, Christian Bauer, Dominique Jennings, Fiona Fen- nessy, Milan Sonka, et al. 3d slicer as an image comput- ing platform for the quantitative imaging network.Magnetic resonance imaging, 30(9):1323–1341, 2012. 3

    2012

  9. [9]

    Segmentanybone: A universal model that segments any bone at any location on mri.Medi- cal Image Analysis, 101:103469, 2025

    Hanxue Gu, Roy Colglazier, Haoyu Dong, Jikai Zhang, Yaqian Chen, Zafer Yildiz, Yuwen Chen, Lin Li, Jichen Yang, Jay Willhite, et al. Segmentanybone: A universal model that segments any bone at any location on mri.Medi- cal Image Analysis, 101:103469, 2025. 1, 5

    2025

  10. [10]

    Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images

    Ali Hatamizadeh, Vishwesh Nath, Yucheng Tang, Dong Yang, Holger R Roth, and Daguang Xu. Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images. InInternational MICCAI brainlesion workshop, pages 272–284. Springer, 2021. 4

    2021

  11. [11]

    Mahdi Imani, Jared Buratto, Thang Dao, Erik Meijering, Sara V ogrin, Timothy CY Kwok, Eric S Orwoll, Peggy M Cawthon, and Gustavo Duque. Deep learning technique for automatic segmentation of proximal hip musculoskeletal tis- sues from ct scan images: a mros study.Journal of Cachexia, Sarcopenia and Muscle, 16(2):e13728, 2025. 2

    2025

  12. [12]

    nnu-net: a self-configuring method for deep learning-based biomedical image segmen- tation.Nature methods, 18(2):203–211, 2021

    Fabian Isensee, Paul F Jaeger, Simon AA Kohl, Jens Pe- tersen, and Klaus H Maier-Hein. nnu-net: a self-configuring method for deep learning-based biomedical image segmen- tation.Nature methods, 18(2):203–211, 2021. 2, 3, 5

    2021

  13. [13]

    nnu-net revisited: A call for rigorous validation in 3d medical image segmentation

    Fabian Isensee, Tassilo Wald, Constantin Ulrich, Michael Baumgartner, Saikat Roy, Klaus Maier-Hein, and Paul F Jaeger. nnu-net revisited: A call for rigorous validation in 3d medical image segmentation. InInternational Confer- ence on Medical Image Computing and Computer-Assisted Intervention, pages 488–498. Springer, 2024. 2, 3

    2024

  14. [14]

    Deep learning-based automatic muscle segmentation of the thigh using lower ex- tremity ct images.Diagnostics, 15(22):2823, 2025

    Young Jae Kim, Ji-Eun Kim, Yeonho Park, Jae Won Chai, Kwang Gi Kim, and Ja-Young Choi. Deep learning-based automatic muscle segmentation of the thigh using lower ex- tremity ct images.Diagnostics, 15(22):2823, 2025. 2, 3

    2025

  15. [15]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. InProceedings of the IEEE/CVF international confer- ence on computer vision, pages 4015–4026, 2023. 2, 4

    2023

  16. [16]

    Saros: A dataset for whole-body region and organ segmentation in ct imaging.Scientific Data, 11(1):483, 2024

    Sven Koitka, Giulia Baldini, Lennard Kroll, Natalie van Lan- deghem, Olivia B Pollok, Johannes Haubold, Obioma Pelka, Moon Kim, Jens Kleesiek, Felix Nensa, et al. Saros: A dataset for whole-body region and organ segmentation in ct imaging.Scientific Data, 11(1):483, 2024. 3

    2024

  17. [17]

    Breastseg- net: Multi-label segmentation of breast mri

    Qihang Li, Jichen Yang, Yaqian Chen, Yuwen Chen, Hanxue Gu, Lars J Grimm, and Maciej A Mazurowski. Breastseg- net: Multi-label segmentation of breast mri. InDeep Breast Workshop on AI and Imaging for Diagnostic and Treatment Challenges in Breast Care, pages 196–205. Springer, 2025. 3

    2025

  18. [18]

    Segment anything in medical images.Nature communications, 15(1):654, 2024

    Jun Ma, Yuting He, Feifei Li, Lin Han, Chenyu You, and Bo Wang. Segment anything in medical images.Nature communications, 15(1):654, 2024. 2, 4

    2024

  19. [19]

    3d mri brain tumor segmentation using autoencoder regularization

    Andriy Myronenko. 3d mri brain tumor segmentation using autoencoder regularization. InInternational MICCAI brain- lesion workshop, pages 311–320. Springer, 2018. 2, 4

    2018

  20. [20]

    Automated ct seg- mentation for lower extremity tissues in lymphedema eval- uation using deep learning.European Radiology, 35(11): 6842–6852, 2025

    Seongwon Na, Se Jin Choi, Yousun Ko, Bushra Urooj, Jimi Huh, Seungwoo Cha, Chul Jung, Hwayeong Cheon, Jae Yong Jeon, and Kyung Won Kim. Automated ct seg- mentation for lower extremity tissues in lymphedema eval- uation using deep learning.European Radiology, 35(11): 6842–6852, 2025. 1, 2, 3

    2025

  21. [21]

    Attention U-Net: Learning Where to Look for the Pancreas

    Ozan Oktay, Jo Schlemper, Loic Le Folgoc, Matthew Lee, Mattias Heinrich, Kazunari Misawa, Kensaku Mori, Steven McDonagh, Nils Y Hammerla, Bernhard Kainz, et al. Atten- tion u-net: Learning where to look for the pancreas.arXiv preprint arXiv:1804.03999, 2018. 2, 4

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  22. [22]

    Lean tissue imaging: a new era for nutritional assessment and interven- tion.Journal of Parenteral and Enteral Nutrition, 38(8):940– 953, 2014

    Carla MM Prado and Steven B Heymsfield. Lean tissue imaging: a new era for nutritional assessment and interven- tion.Journal of Parenteral and Enteral Nutrition, 38(8):940– 953, 2014. 1, 2

    2014

  23. [23]

    Automated segmenta- tion of five different body tissues on computed tomography using deep learning.Medical physics, 50(1):178–191, 2023

    Lucy Pu, Naciye S Gezer, Syed F Ashraf, Iclal Ocak, Daniel E Dresser, and Rajeev Dhupar. Automated segmenta- tion of five different body tissues on computed tomography using deep learning.Medical physics, 50(1):178–191, 2023. 2, 3

    2023

  24. [24]

    U- net: Convolutional networks for biomedical image segmen- tation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. InInternational Conference on Medical image com- puting and computer-assisted intervention, pages 234–241. Springer, 2015. 2, 4

    2015

  25. [25]

    Validation of musculoskeletal segmentation model with uncertainty estimation for bone and muscle as- sessment in hip-to-knee clinical ct images.Scientific reports, 15(1):125, 2025

    Mazen Soufi, Yoshito Otake, Makoto Iwasa, Keisuke Ue- mura, Tomoki Hakotani, Masahiro Hashimoto, Yoshitake Yamada, Minoru Yamada, Yoichi Yokoyama, Masahiro Jin- zaki, et al. Validation of musculoskeletal segmentation model with uncertainty estimation for bone and muscle as- sessment in hip-to-knee clinical ct images.Scientific reports, 15(1):125, 2025. 1, 2, 3

    2025

  26. [26]

    Rethinking model scaling for convolutional neural networks

    Mingxing Tan, Q Efficientnet Le, et al. Rethinking model scaling for convolutional neural networks. InProceedings of the International conference on machine learning, Long Beach, CA, USA, 2019. 2, 4

    2019

  27. [27]

    Antti Tolonen, Tomppa Pakarinen, Antti Sassi, Jere Kytt ¨a, William Cancino, Irina Rinta-Kiikka, Said Pertuz, and Otso Arponen. Methodology, clinical applications, and future di- rections of body composition analysis using computed to- mography (ct) images: a review.European journal of radi- ology, 145:109943, 2021. 1, 2

    2021

  28. [28]

    To- talsegmentator: robust segmentation of 104 anatomic struc- tures in ct images.Radiology: Artificial Intelligence, 5(5): e230024, 2023

    Jakob Wasserthal, Hanns-Christian Breit, Manfred T Meyer, Maurice Pradella, Daniel Hinck, Alexander W Sauter, Tobias Heye, Daniel T Boll, Joshy Cyriac, Shan Yang, et al. To- talsegmentator: robust segmentation of 104 anatomic struc- tures in ct images.Radiology: Artificial Intelligence, 5(5): e230024, 2023. 1, 2, 3

    2023

  29. [29]

    Quantification of muscle, bones, and fat on single slice thigh ct

    Qi Yang, Xin Yu, Ho Hin Lee, Yucheng Tang, Shunxing Bao, Kristofer S Gravenstein, Ann Zenobia Moore, Sokratis Makrogiannis, Luigi Ferrucci, and Bennett A Landman. Quantification of muscle, bones, and fat on single slice thigh ct. InMedical Imaging 2022: Image Processing, pages 422–

    2022

  30. [30]

    Intermuscular adipose tissue rivals visceral adipose tissue in independent associations with cardiovascu- lar risk.International journal of obesity, 31(9):1400–1405,

    JE Yim, S Heshka, J Albu, S Heymsfield, P Kuznia, T Harris, and D Gallagher. Intermuscular adipose tissue rivals visceral adipose tissue in independent associations with cardiovascu- lar risk.International journal of obesity, 31(9):1400–1405,

  31. [31]

    Deep learning– based fully automated body composition analysis of thigh ct: comparison with dxa measurement.European Radiology, 32 (11):7601–7611, 2022

    Hye Jin Yoo, Young Jae Kim, Hyunsook Hong, Sung Hwan Hong, Hee Dong Chae, and Ja-Young Choi. Deep learning– based fully automated body composition analysis of thigh ct: comparison with dxa measurement.European Radiology, 32 (11):7601–7611, 2022. 2

    2022

  32. [32]

    Intermuscular adipose tissue in obesity and related disorders: cellular ori- gins, biological characteristics and regulatory mechanisms

    Ting Zhang, Jun Li, Xi Li, and Yanjun Liu. Intermuscular adipose tissue in obesity and related disorders: cellular ori- gins, biological characteristics and regulatory mechanisms. Frontiers in Endocrinology, 14:1280853, 2023. 3

    2023

  33. [33]

    How to select slices for annotation to train best-performing deep learning segmentation models for cross-sectional med- ical images?arXiv preprint arXiv:2412.08081, 2024

    Yixin Zhang, Kevin Kramer, and Maciej A Mazurowski. How to select slices for annotation to train best-performing deep learning segmentation models for cross-sectional med- ical images?arXiv preprint arXiv:2412.08081, 2024. 1, 3

    work page arXiv 2024