Deep learning-driven MRI radiomics reveals biological subtypes and predicts recurrence risk in rectal cancer.
Machine learning analysis of routine MRI scans identifies three distinct rectal cancer subtypes with different prognosis, improving prediction of who needs aggressive post-surgery treatment.
A DL MRI radiomics approach on 2060 rectal cancer patients across 4 cohorts identified three biologically distinct subtypes (immune-excluded, lymphocyte-enriched, stromal-dominant) with independent prognostic value confirmed by scRNA-seq and bulk RNA-seq. The integrated recurrence risk score improved 5-year RFS prediction vs clinical model alone (AUC 0.834 vs 0.780 external validation), providing a noninvasive tool for individualized postoperative risk stratification.
What the study was
- Study design
- Multicenter retrospective study with external validation (4 independent cohorts)
- Population
- Rectal cancer patients without neoadjuvant therapy (4 cohorts, multicenter China)
- Sample size
- 2060
- Category
- Diagnostics
- Maturity
- Validated
- Journal
- NPJ Precision Oncology
Why it surfaced
Large multicenter study (n=2060) with external validation and multi-omics biological grounding; DL radiomics subtypes map to immune phenotypes relevant for treatment decisions (immunotherapy candidacy); externally validated AUC improvement over clinical staging.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.