Our research employed machine learning and computer vision to analyze the prognostic value of features derived from computational pathology, specifically focusing on the spatial organization and morphological diversity of TILs and cancer nuclei. The investigation spanned five GI cancer types—colon, stomach, pancreatic, and rectal adenocarcinoma, as well as liver hepatocellular carcinoma—using a multi-institutional cohort of over 1,700 patients. From 2,236 features extracted from standard H&E-stained whole-slide images, a LASSO Cox model selected nine top predictors. These were primarily based on the spatial interplay between TILs and the nearest cancer nuclei, along with the shape and texture of tumor nuclei in local clusters.
The resulting model effectively stratified patients into "low-risk" and "high-risk" groups, with the low-risk group demonstrating significantly better overall survival across all cancer types. The identified “low risk” patients have significantly better OS than those identified as “high risk” with hazard ratio (HR) of 2.28 (95% confidence interval (CI):1.32-3.93, p=0.0032) in liver hepatocellular carcinoma (LIHC); HR of 2.79 (95% CI:1.66-4.68, p=0.0001) in pancreatic adenocarcinoma (PAAD); HR of 5.85 (95% CI:2.53-15.5, p=0.0002) in rectal adenocarcinoma (READ); HR of 1.81 (95% CI:1.07-3.07, p=0.0268) in gastric adenocarcinoma (STAD). Across three different external validation sets of CRC patients, our model yielded HR of 2.32 (95% CI: 1.67-3.23, p<0.0001) in TCGA-COAD, HR of 2.32 (95% CI: 1.67-3.23, p<0.0001) in PLCO-COAD and HR of 3.38 (95% CI: 1.99-5.71, p<0.0001) in Emory dataset.
Crucially, multivariable analysis established the model's prognostic value as independent of clinical factors like cancer stage, age, race, and sex, underscoring that the spatial relationship between TILs and cancer nuclei is a robust predictor of survival. Our study's limitations include its retrospective design and a focus on prognosis rather than predicting treatment response. Despite this, the research concludes that this machine learning approach provides a powerful, cost-effective tool for risk stratification in GI cancers, with potential to improve clinical decision-making for stage II colorectal cancer patients.
Written by: Chuheng Chen,1 Haider A. Mejbel,2,3 Tilak Pathak,4 Alyssa Krasinskas,2,3 Michelle Reid,2 German Corredor,4 Pingfu Fu,6 Joseph E. Willis,1,5 Anant Madabhushi,3,4,7
- Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH
- Department of Pathology and Laboratory Medicine, Emory University, Atlanta, GA
- Radiology and Imaging Sciences, Biomedical Informatics (BMI) and Pathology, Georgia Institute of Technology and Emory University, Atlanta, GA
- Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Atlanta, GA
- Department of Pathology, University Hospitals Cleveland Medical Center and Case Western Reserve University, Cleveland, OH
- Department of Population and Quantitative Health Sciences, Case Western Reserve University, Cleveland, OH
- Atlanta Veterans Administration Medical Center, Decatur, GA