logologo

Popular Boards

1 Mentions
Chiara Maria Lavinia Loeffler, Hideaki Bando, Srividhya Sainath | Nature Communications | (2025)
Primary Link Pubmed DOI Openalex Full text (OA)

Abstract

Although surgical resection is the standard therapy for stage II/III colorectal cancer, recurrence rates exceed 30%. Circulating tumor DNA (ctNDA) detects molecular residual disease (MRD), but lacks spatial and tumor microenvironment information. Here, we develop a deep learning (DL) model to predict disease-free survival from hematoxylin & eosin stained whole slide images in stage II-IV colorectal cancer. The model is trained on the DACHS cohort (n = 1766) and validated on the GALAXY cohort (n = 1404). In GALAXY, the DL model categorizes 304 patients as DL high-risk and 1100 as low-risk (HR 2.31; p < 0.005). Combining DL scores with MRD status improves prognostic stratification in both MRD-positive (HR 1.58; p < 0.005) and MRD-negative groups (HR 2.1; p < 0.005). Notably, MRD-negative patients predicted as DL high-risk benefit from adjuvant chemotherapy (HR 0.49; p = 0.01) vs. DL low-risk (HR = 0.92; p = 0.64). Combining ctDNA with DL-based histology analysis significantly improves risk stratification, with the potential to improve follow-up and personalized adjuvant therapy decisions.

Tags

Sample Definition And Size

The study analyzed hematoxylin & eosin–stained whole slide images (WSIs) from two cohorts of colorectal cancer patients: the DACHS cohort (training set) comprising 1,766 patients (1,774 WSIs) and the GALAXY cohort (external validation) comprising 1,404 patients (1,404 WSIs) ([nature.com](https://www.nature.com/articles/s41467-025-62910-8?utm_source=openai)).

Study Type

This is a retrospective observational study employing a deep learning (transformer-based multiple instance learning) model trained on histopathological images, combined with circulating tumor DNA (ctDNA)–based molecular residual disease (MRD) status, to predict disease-free survival and stratify risk in colorectal cancer patients ([nature.com](https://www.nature.com/articles/s41467-025-62910-8?utm_source=openai)).

Conflicts Of Interest

No conflicts of interest are declared in the accessible metadata of the article ([nature.com](https://www.nature.com/articles/s41467-025-62910-8?utm_source=openai)).

Results Summary

In the GALAXY validation cohort, the deep learning (DL) model classified 304 patients as high-risk and 1,100 as low-risk, with a hazard ratio (HR) of 2.31 (p < 0.005) for disease-free survival ([nature.com](https://www.nature.com/articles/s41467-025-62910-8?utm_source=openai)). Combining DL risk scores with MRD status improved prognostic stratification: in MRD-positive patients, HR = 1.58 (p < 0.005); in MRD-negative patients, HR = 2.1 (p < 0.005) ([nature.com](https://www.nature.com/articles/s41467-025-62910-8?utm_source=openai)). Among MRD-negative patients, those classified as DL high-risk benefited from adjuvant chemotherapy (ACT) with HR = 0.49 (p = 0.01), whereas DL low-risk patients did not benefit (HR = 0.92; p = 0.64) ([nature.com](https://www.nature.com/articles/s41467-025-62910-8?utm_source=openai)).

Referenced In

Mercedes C.
2 months ago
AI Papers and Discussion

Colorectal Cancer (CRC) is the 3rd most common cancer worldwide 🌍 , accounting for approximately 10% of all cancer cases.

According to world health organisation (WHO) - in 2022, an estimated 1.9 million new cases of colorectal cancer and more than 900 000 deaths 💀 occurred worldwide.

With the wave of AI-powered innovations transforming how we predict patient outcomes, stratify risk, and personalise treatment: Here are THREE landmark studies from 2025-2026, targeting colorectal cancer.

🔬 SurvFinder: Deep Learning Discovers Novel Prognostic Biomarkers

The Innovation: Researchers built a multi-view deep learning system that analyzes routine H&E slides from multiple angles to identify something pathologists rarely assess systematically: tertiary lymphoid structures (TLSs). These organised immune cell clusters form in response to tumours, and their location and maturity strongly predict outcomes.

The Achievement:

  • 6,950 slides from 1,604 patients across 4 independent cohorts

  • AUROC of 0.827 for predicting recurrence risk

  • High-risk patients showed 8.23× higher hazard for relapse than low-risk

Why It Matters: SurvFinder discovered a clinically actionable biomarker hidden in ordinary microscope slides of tumour tissue. In a disease where 20% relapse post-surgery, this system allows for better treatment plans.

🩸 ColoLDB: Making Precision Medicine Accessible

The Innovation: This is a machine learning model that uses only standard laboratory parameters — complete blood counts, liver function tests, inflammatory markers, and basic metabolic panels — to predict CRC risk and outcomes.

The Achievement:

  • Uses XGBoost and ensemble methods for robust predictions

  • Requires no specialized equipment beyond existing lab infrastructure, which enables population-level screening in resource-limited settings

  • Demonstrated comparable performance to more complex multi-modal systems in specific prediction tasks

Why It Matters: While digital pathology AI grabs headlines, ColoLDB addresses global health equity. By leveraging existing lab infrastructure, it brings AI-powered risk prediction to low-resource settings where CRC mortality remains highest.

🧬 HIBRID: When Two Tests Become One Smart Prediction

The Innovation: First system to combine AI analysis of microscope slides with circulating tumour DNA (ctDNA) blood tests — fusing analysis on how the tumour looks like and whether tumour DNA remains in the body.

The Achievement:

  • Tested on 1,023 stage II patients and Identified four distinct risk strata enabling nuanced treatment decisions

  • Found patients with scary-looking tumours but clean blood tests still face 18% recurrence risk

  • Found patients with benign-looking tumours but positive blood tests face 31% recurrence risk

Why It Matters: HIBRID establishes multi-modal AI as clinically viable, not experimental. It transforms binary treatment decisions into probabilistic, personalised medicine — sparing low-risk patients chemotherapy toxicity while ensuring high-risk patients receive intervention.

Read Full Post
3