AI Maps MRI Signs of Vascular Invasion in Rectal Cancer
A multicentre retrospective study tested automated MRI segmentation of extramural vascular invasion, with stronger overlap for tumor regions than for the smaller vascular signals.
Making a subtle MRI finding more measurable
Extramural vascular invasion (EMVI) describes tumor within vessels outside the bowel wall and can provide prognostic information in rectal cancer. On MRI, identifying the relevant vascular signal can be difficult and may vary between readers.
A study in PLOS Digital Health trained an nnU-Net segmentation model to locate and visualize MRI-detected EMVI. The multicentre retrospective dataset included 2,501 patients: 1,830 for model development and 671 in two independent external test cohorts.
The model’s performance was uneven by region
In internal cross-validation, reported Dice scores were 0.850 for the tumor region, 0.442 for the intravascular tumor signal, and 0.335 for the dilated vessel region. The corresponding scores in combined external testing were lower: 0.748, 0.340, and 0.256. The study also reported external classification accuracy of 81.5% and 84.7% in the two test cohorts, with substantial agreement with senior radiologists.
The difference between tumor and vascular-region scores matters. The clinically specific signal is harder to segment than the surrounding tumor, so a single headline metric would hide an important limitation.
A research aid, not an autonomous read
The retrospective results suggest that automated localization could help standardize a difficult imaging task. They do not establish that the model can replace radiologist review or improve treatment decisions. The external results also show why validation on new cohorts is essential.
Future work should test the model prospectively, assess performance across scanners and institutions, and evaluate how visual overlays affect reader decisions. For now, the model is best understood as a candidate support tool whose most challenging target remains the one of greatest clinical interest.
Source and credit
Independent summary of “Artificial intelligence for evaluation of magnetic resonance imaging-detected extramural vascular invasion in rectal cancer,” PLOS Digital Health, published 1 October 2026. Original paper, DOI: 10.1371/journal.pdig.0001763. Source article license: CC BY 4.0. No source figures or tables are reproduced here.
Questions about this article? Contact the author. Research discussion is welcome; please do not submit personal medical information.