.png){kind=icon}
.png){kind=icon}
Return
Clinical target volume radiomics from planning CT for pretreatment response prediction in rectal cancer undergoing chemoradiotherapy
.png){kind=icon}
PRE
AI
Y
S
J
S
DOI：10.1007/s00066-026-02569-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Response to neoadjuvant chemoradiotherapy (nCRT) in rectal cancer remains heterogeneous. This exploratory study evaluated whether radiomics features extracted from the clinical target volume (CTV) on pretreatment non-contrast radiotherapy planning CT could provide additional information for pretreatment prediction of poor treatment response. This retrospective single-center study included 60 patients with rectal cancer treated with nCRT between 2008 and 2024, including 50 good responders and 10 poor responders. A total of 1148 radiomics features were extracted from the original pretreatment treatment-planning CTV and combined with clinical variables, including age, sex, cT stage, cN stage, log-transformed carcinoembryonic antigen, and tumor-to-anal verge distance. Models were internally evaluated using repeated nested cross-validation with fold-wise ComBat harmonization. Performance was assessed from pooled repeated out-of-fold predictions, including discrimination, apparent calibration, exploratory threshold-based operating characteristics, and decision curve analysis. The best-performing combined clinical–radiomics model was Extra Trees, with an area under the receiver operating characteristic curve of 0.754 (95% confidence interval [CI], 0.713–0.789), compared with 0.507 for the clinical-only model. At a post hoc sensitivity-prioritized operating threshold, the combined Extra Trees model achieved sensitivity of 0.94 (95% CI, 0.89–0.98), specificity of 0.47 (95% CI, 0.42–0.51), and negative predictive value of 0.98 in pooled repeated out-of-fold analysis. Decision curve analysis suggested potential net benefit across low-to-moderate threshold probabilities in internal analysis. Clinical target volume-based radiomics from routine pretreatment non-contrast planning CT may provide exploratory information for prediction of poor response to nCRT in rectal cancer. Because treatment-planning CTVs are routinely generated before radiotherapy, this workflow may be practical for radiotherapy-based research. The findings are hypothesis generating and require external validation before clinical use.
Keywords:
Rectal cancer
Neoadjuvant chemoradiotherapy
CT radiomics
Clinical target volume
Machine learning
Decision curve analysis
Journal
IF:
2.5
Papers:
270
Citations:
3.4K
Organization
Cited Papers