Summary
This content was automatically synthesized by Credo's AI models directly from the original source text.
AI summaries can make mistakes — double-check important details against the original source.
1 Sample Definition And Size
The study is a retrospective multicenter cohort study including 1,604 patients with stage II colorectal cancer (CRC), represented by 6,950 H&E-stained whole-slide images (WSIs). The cohorts comprised: Internal‑CRCII (743 patients, 3,494 slides), External‑CRCII‑1 (352 patients, 1,315 slides), External‑CRCII‑2 (331 patients, 1,708 slides), and TCGA‑CRCII (178 patients, 433 slides) ([journals.plos.org](https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1004614)).
2 Study Type
Retrospective multicenter cohort study employing deep learning (SurvFinder framework) with multiple-instance learning, segmentation networks, multi-view fusion (MVNet), and multimodal fusion (MMF) integrating histopathological image features and clinical data ([journals.plos.org](https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1004614)).
3 Conflicts Of Interest
The authors declared that no competing interests exist ([journals.plos.org](https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1004614)).
4 Results Summary
Key findings: SurvFinder identified tertiary lymphoid structures (TLSs) as critical prognostic features. MVNet (multi-view fusion of spatial and morphological TLS features) achieved AUROCs of 0.827 (95% CI 0.789–0.864), 0.805 (95% CI 0.749–0.860), and 0.805 (95% CI 0.748–0.861) across Internal‑CRCII, External‑CRCII‑1, and External‑CRCII‑2 cohorts, respectively, outperforming WSINet and single-branch models ([journals.plos.org](https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1004614)). In multivariate Cox regression, MVNet had a hazard ratio (HR) of 8.23 (95% CI 5.43–12.47; p < 0.001), indicating strong independent prognostic value ([journals.plos.org](https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1004614)). The MMF model combining MVNet with clinical variables further improved AUROC compared to MVNet or clinical-only models ([journals.plos.org](https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1004614)). In high-risk patients as predicted by MVNet, adjuvant chemotherapy (ACT) was associated with significantly improved relapse-free survival in Internal‑CRCII (p = 0.023), External‑CRCII‑1 (p = 0.026), and External‑CRCII‑2 (p = 0.00081) cohorts; low-risk patients did not benefit from ACT ([journals.plos.org](https://journals.plos.org/plosmedicine/article?id=10.1371%2Fjournal.pmed.1004614)).