EUVIMEDEUVIMED entdecken
Crossref

The study of rectal cancer staging with magnetic resonance imaging radiomics

Σπυρίδων Χ. Νίκας

Aristotle University of Thessaloniki · 01.01.2026

DOI: 10.26262/heal.auth.ir.374610 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Aristotle University of Thessaloniki, erschienen am 01.01.2026. Verfasst von Σπυρίδων Χ. Νίκας.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Rectal cancer is a major cause of morbidity and mortality worldwide. Although rectal MRI is the reference standard for locoregional staging, radiologist assessment shows variability and offers limited prognostic biomarkers. Radiomics offers potential imaging biomarkers, while alternative peritumoral segmentation strategies may be biologically informative and more practical than tumor-only approaches.

Purpose

To explore radiomics-based ML classifiers that accurately diagnose T staging, N staging, MRF involvement and EMVI in T2WI rectal MRIs, and define the best segmentation approach to guide future radiomics and segmentation research. Materials and Methods: We conducted a retrospective, diagnostic accuracy feasibility study in the Radiology Department of the “Papageorgiou” General Hospital. 90 patients were included. Of those, 77/90 were advanced T-stage (T3 or T4), 69/90 were advanced N-stage (N1 or N2), 42/90 had an involved MRF, and 55/90 had a positive EMVI. The Radiologist’s report was the reference standard. Three segmentations were performed: tumor-only, whole-organ and MRF. Radiomics features were extracted with PyRadiomics. Stepwise feature selection was applied (filter and wrapper methods). Radiomics and clinical characteristics were combined. Several ML models were trained (5-fold cross-validation), for each combination and segmentation mask, with mean balanced accuracy as the primary metric.

Results

Best model for T staging: SVM of combined features for the tumor-only segmentation, with mean Balanced Accuracy = 74.3% (SD = 14.2%). Best model for N staging: SVM of combined features for the combined segmentation, with mean Balanced Accuracy = 73.3% (SD = 14.3%). Best model for MRF involvement: RF model of the radiomics features for the MRF segmentation, with mean Balanced Accuracy = 77.9% (SD = 10.3%). Best model for EMVI: RF model of combined features for the MRF segmentation, with mean Balanced Accuracy = 74.6% (SD = 6.6%).

Conclusion

For T staging, accurate tumor-only segmentation yields the best results. For N staging, features from a combination of segmentations are better. For MRF involvement and EMVI, features from the MRF segmentation mask performed best. Our models had overall a moderate, realistic performance. Nevertheless, for some outcomes, simpler segmentations like MRF may prove non-inferior or even superior options.

Schlagwörter

rectal cancerradiomicsMRI stagingT stagingN staginglymph node metastasismesorectal fascia involvementextramural venous invasionmachine learning classifierssupport vector machine

Zitieren

APA

Νίκας, Σ. Χ. (2026). The study of rectal cancer staging with magnetic resonance imaging radiomics. Aristotle University of Thessaloniki. https://doi.org/10.26262/heal.auth.ir.374610

Vancouver

Νίκας ΣΧ. The study of rectal cancer staging with magnetic resonance imaging radiomics. Aristotle University of Thessaloniki. 2026. doi:10.26262/heal.auth.ir.374610

BibTeX
@article{2026,
  author = {Νίκας, Σπυρίδων Χ.},
  title = {The study of rectal cancer staging with magnetic resonance imaging radiomics},
  journal = {Aristotle University of Thessaloniki},
  year = {2026},
  doi = {10.26262/heal.auth.ir.374610}
}
Kostenlos starten →