Deep learning for risk stratification and identification of prognostic features in patients with colorectal metastases
Metastatic disease remains a leading cause of cancer-related mortality for patients diagnosed with colorectal cancer. While surgical resection of the primary tumor is a primary treatment, clinicians currently lack the precise tools needed to identify which patients will develop metastases or face relapse.
Supported by funding from Kreftforeningen (the Norwegian Cancer Society), and Helse Sør-Øst (South-Eastern Norway Regional Health Authority), two parallel projects running from July 2025 to July 2027 utilise digital pathology and deep learning to develop explainable prognostic models based on routine tissue slides. One project focuses on colorectal metastases to the liver, while the other focuses on metastases to the peritoneum. These are among the most common metastatic sites in colorectal cancer and together account for about half of all metastatic-site occurrences. The goal is to improve risk stratification for patients with metastatic colorectal cancer, providing a foundation for more personalised treatment.
Methodology and Clinical Innovation
The projects include tissue slides from both primary colorectal tumors and metastatic samples collected from Norwegian and international cohorts. The liver metastases project includes more than 7000 patients from 12 cohorts, while the peritoneal metastases project includes more than 5600 patients from 9 cohorts.
Deep models will be developed to predict metastatic spread, as well as disease relapse and patient survival after metastatic disease has been diagnosed. These models identify biologically meaningful patterns in the tumor and its microenvironment linked to prognosis that are often not captured in routine pathology practice. By correlating model outputs with specific cell types and tissue structures, we aim to identify biological changes and new biomarker associated with metastasis. This transparency, which can be defined as "explainable AI", is essential for building the clinical trust required for safe, real-world implementation.
Future Development and Systemic Impact
These projects may have an impact beyond the development of prognostic models alone. In the longer term, the models could be combined with other clinicopathological information in clinical decision-support tools for treatment planning and follow-up. The dedicated biobanks established through these projects will create important infrastructure for future research on colorectal metastases. Together, these efforts may improve our understanding of the biological mechanisms underlying metastatic disease, help identify new biomarkers and therapeutic targets and support more personalised treatment.
A Strategic Asset for Health Systems
This research addresses key healthcare challenges linked to an ageing population, a rising volume of samples for analysis, and a global shortage of pathologists. By automating the analysis of tissue samples from colorectal cancer patients, we can provide rapid, objective risk assessments at a low cost.
Better targeting of intensive therapies to high-risk patients—while sparing low-risk individuals from unnecessary toxicity—directly improves survival and quality of life. In this way, the projects may contribute to more efficient use of healthcare resources and help establish a new diagnostic standard, aligning with the future of precision medicine.