Journal: NPJ digital medicine
This publication reports the development of PRISM-CRC, a deep learning framework that integrates histopathology images, radiologic data, and clinical variables to improve both diagnosis and prognosis in colorectal cancer.
Key points:
- Multimodal integration: By combining pathology, imaging, and clinical data, PRISM-CRC outperforms single-modality models in both survival prediction and biomarker classification.
- Prognostic performance: The model achieves a concordance index of 0.82 for predicting 5-year disease-free survival, and its composite risk score is an independent predictor of outcome, providing more refined risk stratification than standard TNM staging.
- Diagnostic performance: PRISM-CRC predicts microsatellite instability (MSI) status with an AUC of 0.91, suggesting potential utility in guiding immunotherapy decisions.
- Clinical implications: The refined risk stratification may help identify high-risk stage II patients who could benefit from adjuvant chemotherapy and better tailor treatment intensity across stages.
- Limitations: Model performance decreases on external cohorts due to domain shift, and errors occur in morphologically ambiguous cases. The authors emphasize the need for prospective, multi-center validation before clinical adoption.