AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy
Open accessThe DOI matches its Crossref record; 35 of the 35 references it cites were found in a registry or an index; none of the 44 sources its claims cite is checked for support yet.
Abstract
Background: Older patients with lung cancer receiving thoracic radiotherapy have heterogeneous outcomes that are not fully captured by conventional clinical descriptors. We evaluated whether artificial intelligence (AI)-assisted three-dimensional (3D) thoracic body-composition analysis of pretreatment radiotherapy simulation computed tomography (CT) could identify candidate body-composition imaging markers associated with overall survival. Methods: This exploratory retrospective single-institution study included 88 patients aged 70 years or older who received thoracic radiotherapy for lung cancer between 2019 and 2021. Thoracic skeletal muscle and intermuscular fat were segmented using an AI-assisted workflow followed by expert review and Hounsfield unit-constrained refinement. Twelve knowledge-informed candidate groups were evaluated using conventional Cox regression. After marker selection, the selected marker pair was assessed in continuous form using bootstrap optimism correction, repeated five-fold cross-validation, and Uno’s C-index. Penalized and alternative model-family analysis were performed as exploratory sensitivity analysis. Results: During follow-up, 28 deaths occurred. In an 18-case internal technical validation subset, the raw AI contours achieved Dice similarity coefficients of 0.930 for thoracic skeletal muscle and 0.723 for intermuscular fat. For Group 12, intermuscular adiposity index (IAI) and attenuation-weighted thoracic muscle fraction (TMF) were the variables that met the univariable screening criterion; the resulting two-marker model met the discrimination threshold without problematic collinearity. When modeled together, IAI values above the cohort median (hazard ratio [HR], 2.78; 95% confidence interval [CI], 1.25–6.18; P = 0.012) and TMF values above the cohort median (HR, 3.77; 95% CI, 1.52–9.32; P = 0.004) were associated with poorer overall survival. In continuous form, the selected marker pair had an optimism-corrected Harrell’s C-index of 0.751 (95% CI, 0.661–0.849), a mean repeated cross-validation C-index of 0.753 ± 0.112, and Uno’s C-index of 0.737. These performance estimates were conditional on prior marker selection. Conclusion: AI-assisted analysis of routine radiotherapy simulation CT identified IAI and TMF as candidate body-composition imaging markers associated with overall survival in this clinically heterogeneous cohort. The findings are exploratory and require independent evaluation in larger, clinically characterized cohorts before clinical use.
Overview
CoChat Summary
TL;DR
In 88 older patients with lung cancer receiving thoracic radiotherapy, AI-assisted 3D body-composition analysis of pretreatment simulation CT identified intermuscular adiposity index (IAI) and attenuation-weighted thoracic muscle fraction (TMF) as candidate markers associated with overall survival. The selected two-marker model showed post-selection internal C-index estimates of 0.737–0.753, but the findings are exploratory and require external validation.
Key findings
- Among 88 patients (median age 77 years; 86.4% male; median BMI 23.3 kg/m²; median radiation dose 6000 cGy), 28 deaths occurred during follow-up; median follow-up was 18.7 months.
- Raw AI contours achieved Dice similarity coefficients of 0.930 for thoracic skeletal muscle and 0.723 for intermuscular fat, with HD95 values of 3.200 mm and 8.196 mm, respectively.
- In the selected Group 12 model, IAI above the cohort median was associated with poorer overall survival (HR, 2.78; 95% CI, 1.25–6.18; P = 0.012), as was TMF above the cohort median (HR, 3.77; 95% CI, 1.52–9.32; P = 0.004).
- For continuous IAI + TMF after marker selection, the optimism-corrected Harrell’s C-index was 0.751 (95% CI, 0.661–0.849), repeated five-fold cross-validation C-index was 0.753 ± 0.112, and Uno’s C-index was 0.737; the higher model-score group had poorer overall survival (log-rank P = 0.0048).
Methods
- Retrospective single-institution cohort of 88 patients aged ≥70 years who received thoracic radiotherapy for lung cancer between 2019 and 2021; primary outcome was overall survival from radiotherapy start.
- Pretreatment radiotherapy simulation CT was analyzed with an AI-assisted nnU-Net-based segmentation workflow for thoracic skeletal muscle and intermuscular fat, followed by expert review and Hounsfield unit-constrained refinement; 12 knowledge-informed candidate groups were evaluated using conventional Cox regression with median dichotomization, C-index screening threshold 0.70, and collinearity thresholds of |ρ| ≥ 0.70 or VIF > 5.
- Post-selection internal performance used bootstrap optimism correction, repeated five-fold cross-validation, and Uno’s C-index; exploratory sensitivity analyses included elastic-net Cox, random survival forest, gradient boosting survival, and a limited clinical reference model.
Limitations
- Stated in the paper: retrospective single-center design with 88 patients and 28 deaths; mixed histologic subtypes and stages, with stage information unavailable for 26 of 88 patients; incomplete data on performance status, comorbidities, systemic therapy, frailty, nutritional interventions, pulmonary function, and cause-specific mortality; segmentation was technically validated in an 18-case subset from the same cohort and all final contours underwent expert review and HU-constrained refinement; resampling began after IAI and TMF were selected, and the complete 12-group marker-selection procedure was not repeated within each resample; the clinical reference model was limited, calibration and decision-curve analyses were not performed, and the model-derived grouping threshold was cohort-specific.
- Evident from the text: no external validation was performed; no formal paired comparison was made between the clinical + IAI + TMF model and the IAI + TMF-only model; the exploratory penalized and machine-learning analyses were not used to change marker selection; and the study did not directly measure physical function, metabolic status, or cardiovascular risk, leaving the biological interpretation of TMF unclear.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 0 Not valid · 12 Not checked · 88 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The corresponding HD95 values were 3.947, 3.200, and 8.196 mm, respectively.”
§ AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy > Results > AI-assisted segmentation performance
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The corresponding HD95 values were 3.947, 3.200, and 8.196 mm, respectively.”
§ AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy > Results > AI-assisted segmentation performance
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“the selected marker pair had an optimism-corrected Harrell’s C-index of 0.751 (95% CI, 0.661–0.849)”
- Also stated in: § Post-selection performance and descriptive survival analysis · plain
“The optimism-corrected Harrell’s C-index was 0.751 (95% CI, 0.661–0.849)”
§ AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy > Abstract > Results
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on 2 assumptions not shown here.
- Also stated in: § Post-selection performance and descriptive survival analysis · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“the raw AI-generated contours yielded DSC values of 0.913 for the combined thoracic muscle–intermuscular fat volume, 0.930 for thoracic skeletal muscle, and 0.723 for intermuscular fat.”
§ AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy > Results > AI-assisted segmentation performance
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“the raw AI contours achieved Dice similarity coefficients of 0.930 for thoracic skeletal muscle and 0.723 for intermuscular fat.”
- Also stated in: § AI-assisted segmentation performance · plain
“the raw AI-generated contours yielded DSC values of 0.913 for the combined thoracic muscle–intermuscular fat volume, 0.930 for thoracic skeletal muscle, and 0.723 for intermuscular fat.”
§ AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy > Abstract > Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § AI-assisted segmentation performance · plain
37 more claims
Create a free account to see all 100 claims and to verify the 44 references still open.
- Probably not a substantial claim. 42
- It may state two separate claims as one. 29
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy > Discussion
“After IAI and TMF had been selected, the continuous two-marker model yielded C-index estimates ranging from 0.737 to 0.753 across the post-selection analysis.”
§ AI-assisted 3D thoracic body composition analysis identifies candidate imaging markers associated with overall survival in older patients with lung cancer receiving radiotherapy > Results > Post-selection performance and descriptive survival analysis
“| Continuous IAI + TMF | 0.764 | 0.751 (0.661–0.849) | 0.753 ± 0.112 | 0.737 |”
Citations
Cites
nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation
Isensee, F., Jaeger, P. F., Kohl, S. A. A., Petersen, J., & Maier-Hein, K. H. (2021) · Nature Methods · doi:10.1038/s41592-020-01008-z · Cited by 10,477
Siegel, R. L., Kratzer, T. B., Giaquinto, A. N., Sung, H., & Jemal, A. (2025) · CA: A Cancer Journal for Clinicians · doi:10.3322/caac.21871 · Cited by 3,060
Ishwaran, H., Kogalur, U. B., Blackstone, E. H., & Lauer, M. S. (2008) · The Annals of Applied Statistics · doi:10.1214/08-aoas169 · Cited by 2,714
Regularization Paths for Cox's Proportional Hazards Model via Coordinate Descent
Simon, N., Friedman, J., Hastie, T., & Tibshirani, R. (2011) · Journal of Statistical Software · doi:10.18637/jss.v039.i05 · Cited by 2,499
Uno, H., Cai, T., Pencina, M. J., D'Agostino, R. B., & Wei, L. J. (2011) · Statistics in Medicine · doi:10.1002/sim.4154 · Cited by 1,535
scikit-survival: A Library for Time-to-Event Analysis Built on Top of scikit-learn
Pölsterl, S. (2020) · Journal of Machine Learning Research · Cited by 701
Mohile, S. G., Mohamed, M. R., Xu, H., Culakova, E., Loh, K. P., Magnuson, A., Flannery, M. A., Obrecht, S., Gilmore, N., Ramsdale, E., et al. (2021) · The Lancet · doi:10.1016/s0140-6736(21)01789-x · Cited by 602
Force, L. M., Kocarnik, J. M., May, M. L., Bhangdia, K., Crist, A., Penberthy, L., Pritchett, N., Acheson, A., Deitesfeld, L., A, B., et al. (2025) · The Lancet · doi:10.1016/s0140-6736(25)01635-6 · Cited by 494
Automated Abdominal Segmentation of CT Scans for Body Composition Analysis Using Deep Learning
Weston, A. D., Korfiatis, P., Kline, T. L., Philbrick, K. A., Kostandy, P., Sakinis, T., Sugimoto, M., Takahashi, N., & Erickson, B. J. (2019) · Radiology · doi:10.1148/radiol.2018181432 · Cited by 358
Mutic, S., Palta, J. R., Butker, E. K., Das, I. J., Huq, M. S., Loo, L.-N. D., Salter, B. J., McCollough, C. H., & Van Dyk, J. (2003) · Medical Physics · doi:10.1118/1.1609271 · Cited by 344
Intermuscular adipose tissue in metabolic disease
Goodpaster, B. H., Bergman, B. C., Brennan, A. M., & Sparks, L. M. (2023) · Nature Reviews Endocrinology · doi:10.1038/s41574-022-00784-2 · Cited by 177
Koitka, S., Kroll, L., Malamutmann, E., Oezcelik, A., & Nensa, F. (2021) · European Radiology · doi:10.1007/s00330-020-07147-3 · Cited by 158
Bates, D. D. B., & Pickhardt, P. J. (2022) · American Journal of Roentgenology · doi:10.2214/ajr.22.27749 · Cited by 82
Mai, D. V. C., Drami, I., Pring, E. T., Gould, L. E., Lung, P., Popuri, K., Chow, V., Beg, M. F., Athanasiou, T., Jenkins, J. T., et al. (2023) · Journal of Cachexia, Sarcopenia and Muscle · doi:10.1002/jcsm.13310 · Cited by 78
Skeletal muscle adiposity, coronary microvascular dysfunction, and adverse cardiovascular outcomes
Souza, A. C. D. A. H., Troschel, A. S., Marquardt, J. P., Hadžić, I., Foldyna, B., Moura, F. A., Hainer, J., Divakaran, S., Blankstein, R., Dorbala, S., et al. (2025) · European Heart Journal · doi:10.1093/eurheartj/ehae827 · Cited by 68
Liu, Q., Li, J., Sun, X., Lin, J., Yu, Z., Xiao, Y., Li, D., Sun, B., Bao, H., & Liu, Y. (2025) · Journal of Hematology & Oncology · doi:10.1186/s13045-025-01735-w · Cited by 65
Liu, X., Qu, L., Xie, Z., Zhao, J., Shi, Y., & Song, Z. (2024) · BioMedical Engineering OnLine · doi:10.1186/s12938-024-01238-8 · Cited by 50
Body composition in advanced non-small cell lung cancer treated with immunotherapy
Chaunzwa, T. L., Qian, J. M., Li, Q., Ricciuti, B., Nuernberg, L., Johnson, J. W., Weiss, J., Zhang, Z., MacKay, J., Kagiampakis, I., et al. (2024) · JAMA Oncology · doi:10.1001/jamaoncol.2024.1120 · Cited by 50
Anwar, M. R., Yeretzian, S. T., Ayala, A. P., Matosyan, E., Breunis, H., Bote, K., Puts, M., Habib, M. H., Li, Q., Sahakyan, Y., et al. (2023) · JNCI: Journal of the National Cancer Institute · doi:10.1093/jnci/djad200 · Cited by 47
Automated deep learning-based detection and segmentation of lung tumors at CT Imaging
Kashyap, M., Wang, X., Panjwani, N., Hasan, M., Zhang, Q., Huang, C., Bush, K., Chin, A., Vitzthum, L. K., Dong, P., et al. (2025) · Radiology · doi:10.1148/radiol.233029 · Cited by 36
Knowledge-Informed Machine Learning for Cancer Diagnosis and Prognosis: A Review
Mao, L., Wang, H., Hu, L., Tran, N. L., Canoll, P., Swanson, K. R., & Li, J. (2024) · IEEE Transactions on Automation Science and Engineering · doi:10.1109/tase.2024.3515839 · Cited by 34
Farsijani, S., Santanasto, A. J., Miljkovic, I., Boudreau, R. M., Goodpaster, B. H., Kritchevsky, S. B., & Newman, A. B. (2021) · The Journals of Gerontology: Series A · doi:10.1093/gerona/glaa161 · Cited by 30
Prognostic value of comorbidities in older patients with cancer: the ELCAPA cohort study
Benderra, M.-A., Serrano, A. G., Paillaud, E., Tapia, C. M., Cudennec, T., Chouaïd, C., Lorisson, E., de la Taille, A., Laurent, M., Brain, E., et al. (2023) · ESMO Open · doi:10.1016/j.esmoop.2023.101831 · Cited by 30
Zhang, T., Li, J., Li, X., & Liu, Y. (2023) · Frontiers in Endocrinology · doi:10.3389/fendo.2023.1280853 · Cited by 21
Xing, H., Wu, C., Yang, W., Cai, S., Zhang, X., & Ye, X. (2025) · BMC Cancer · doi:10.1186/s12885-025-14363-x · Cited by 19
DeSeg: auto detector-based segmentation for brain metastases
Yu, H. J., Zhang, Z., Xia, W., Liu, Y., Liu, L., Luo, W., Zhou, J., & Zhang, Y. (2023) · Physics in Medicine and Biology · doi:10.1088/1361-6560/acace7 · Cited by 16
Ying, L., Xu, L., Yang, J., & Zhang, Q. (2024) · Scientific Reports · doi:10.1038/s41598-024-62825-2 · Cited by 15
Künnemann, M., Römer, C., Helfen, A., Bleckmann, A., Kemper, M., Heindel, W., Brix, T. J., Forsting, M., Haubold, J., Opitz, M., et al. (2025) · Journal of Cachexia, Sarcopenia and Muscle · doi:10.1002/jcsm.70021 · Cited by 14
Association of obesity and skeletal muscle with postoperative survival in non-small cell lung cancer
Lee, J. H., Kang, D., Lee, J., Jeon, Y. J., Park, S. Y., Cho, J. H., Choi, Y. S., Kim, J., Shim, Y. M., Kong, S., et al. (2025) · Radiology · doi:10.1148/radiol.241507 · Cited by 14
Brown, L. R., Soupashi, M., Yule, M. S., Clyde, D. R., Gardner, E., Smith, C., Dhaif, A., Laird, B. J. A., Wigmore, S. J., & Skipworth, R. J. E. (2025) · Journal of Cachexia, Sarcopenia and Muscle · doi:10.1002/jcsm.13673 · Cited by 12
Fu, L., Ding, H., Mo, L., Pan, X., Feng, L., Wen, S., Lan, Q., & Long, L. (2025) · Scientific Reports · doi:10.1038/s41598-025-87073-w · Cited by 10
Golban, C., Susa, S.-R., Varga, N.-I., Ivan, C.-S., Schirta, P. O., Schirta, N. C., Negru, A. G., Saftescu, S., & Negru, S. M. (2025) · Cancers · doi:10.3390/cancers17172765 · Cited by 9
Lung cancer among the Elderly in Denmark – A comprehensive population-based cohort study
Christensen, N. L., Gouliaev, A., McPhail, S., Lyratzopoulos, G., Rasmussen, T. R., & Jensen, H. (2024) · Lung Cancer · doi:10.1016/j.lungcan.2024.107555 · Cited by 7
Gu, X., Zhu, Y., Li, C., Xu, X., Jin, K., & Xu, L. (2025) · npj Digital Medicine · doi:10.1038/s41746-025-02041-y · Cited by 6
Sharma, R. K., Roy, S., Collaborators, A., & Force, L. M. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6378361 · Cited by 4
Cited by
Still being checked.
Metadata
- Authors
- Sung, S., Choi, H. J., Jeong, S., Cho, W., Kim, Y. J., Cha, H., Min, C. H., & Kim, M. S.
- Affiliations
- Yonsei University (KR)
- Oncosoft Inc.
- Corresponding author
- Kim, Mi Sun
- Venue
- Frontiers in Oncology
- Publisher
- Frontiers Media SA
- Published
- 2026-09-29
- Volume
- 16
- Pages
- 1946212-1946212
- ISSN
- 2234-943X (ISSN-L)
- PMC
- PMC13658111
- OpenAlex
- W7214769925
- Semantic Scholar
- d4db8aad6813d9b47aa0c412cfd67e493bc3c360
- Cited by
- 0
- References
- 35
- FWCI
- 0.00
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Frontiers in Oncology (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- Funders
- National Research Foundation of Korea (133561)
- Ministry of Education
- Topics
- Health Sciences › Medicine › Pulmonary and Respiratory Medicine › Lung Cancer Diagnosis and Treatment
- Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
- Health Sciences › Medicine › Physiology (medical) › Nutrition and Health in Aging
- Keywords
- lung cancer, thoracic radiotherapy, elderly patients, body composition, skeletal muscle, intermuscular fat, artificial intelligence segmentation, imaging biomarkers, overall survival, prognostic factors, concordance index, interobserver variability
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-11-21, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-21 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.

