logologo

Popular Boards

1 Mentions
Xu Dong Zhang, Xuedong Tong, Jiangtao Mou | Journal of Gastrointestinal Oncology | (2026)
Primary Link Pubmed DOI Openalex Full text (OA)

Key Takeaways

Sample Definition And Size

The study retrospectively analyzed patients from The Third Affiliated Hospital of Chongqing Medical University between 2022 and 2024. It included individuals diagnosed with colorectal cancer (CRC) and those with benign colorectal diseases. After excluding invalid or non-numerical records and retaining only the first diagnostic test result per parameter per patient, 78 laboratory indicators remained from an initial 371. From these, eight key parameters were selected to construct the ColoLDB model. The exact number of patients is not specified in the accessible metadata.

Study Type

Observational retrospective diagnostic study using machine learning model development and validation (random forest, LightGBM, logistic regression, XGBoost), following TRIPOD reporting guidelines.

Conflicts Of Interest

One author, C.Z., is an employee of Roche Diagnostics Ltd. The other authors declared no conflicts of interest. ([pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC12972017/?utm_source=openai))

Results Summary

The random forest (RF) ColoLDB model, based on eight parameters (specific gravity, CA19‑9, CEA, age, albumin, CYFRA21‑1, HDL‑C, CA72‑4), achieved in the test set: AUC 0.863 (95% CI: 0.792–0.922), accuracy 0.900, sensitivity 0.225, specificity 0.997, PPV 0.917, NPV 0.900. When specificity was set at 0.903, sensitivity increased to 0.694. In comparison, a model combining CEA and CA19‑9 had AUC 0.688, sensitivity 0.429, specificity 0.947. ([pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC12972017/?utm_source=openai))

Abstract

Our research findings indicate that eight laboratory test indicators may be related the risk of developing CRC. Our RF diagnostic ColoLDB model is an innovative and practical tool that effectively predicts the occurrence of CRC, enhancing the diagnostic efficiency for this disease. This method holds promise as a valuable tool for diagnosing CRC.

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