Related Experiment Video
Updated: Jan 11, 2026
15:17
Colorectal Cancer Cell Surface Protein Profiling Using an Antibody Microarray and Fluorescence Multiplexing
Published on: September 25, 2011
14.3K
Deep learning-enabled multiphoton microscopy predicts colorectal cancer recurrence from routine FFPE specimens.
Yabing Yang1,2, Chanchan Xiao3,4, Dehua Zou5,6
1Department of Rheumatology and Immunology, The Affiliated Guangdong Second Provincial General Hospital of Jinan University, Guangzhou, China.
NPJ Digital Medicine
|November 18, 2025
Summary
A new deep learning model, MPMRecNet, accurately predicts colorectal cancer recurrence from standard pathology slides. This tool aids in early risk assessment and personalized patient treatment planning after surgery.
Area of Science:
- Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Colorectal cancer (CRC) recurrence post-curative resection is a significant clinical challenge.
- Accurate prediction of CRC recurrence is crucial for patient stratification and personalized therapy.
- Existing prediction tools often lack sufficient accuracy for early risk assessment.
Purpose of the Study:
- To develop and validate a novel deep learning model for predicting colorectal cancer recurrence.
- To assess the performance of the model using multiphoton microscopy (MPM) imaging of tissue samples.
- To compare the model's predictive power against traditional clinical predictors.
Main Methods:
- Development of MPMRecNet, a dual-stream deep learning model utilizing MaxViT encoders and cross-modal attention fusion.
- Training and validation on formalin-fixed paraffin-embedded (FFPE) tissue sections from 1071 patients across two institutions.
- Performance evaluation using ROC-AUC and PR-AUC metrics, alongside multivariable analysis and nomogram integration.
Main Results:
- MPMRecNet demonstrated strong external validation performance with ROC-AUC of 0.849 and PR-AUC of 0.664.
- The model significantly outperformed traditional clinical predictors in identifying recurrence risk.
- Multivariable analysis identified MPMRecNet as the most potent independent predictor (OR=5.66, p<0.001); a combined nomogram achieved ROC-AUC of 0.872.
Conclusions:
- MPMRecNet provides a non-destructive, accurate method for predicting colorectal cancer recurrence from routine pathology slides.
- The model facilitates precise risk stratification and supports enhanced postoperative surveillance strategies.
- This AI-driven approach holds promise for improving personalized management of patients treated for colorectal cancer.
