Artificial Intelligence Identifies Prognostic Markers for GI Cancers

Strategic Insights -

Researchers from Emory University have employed machine learning and computer vision to quantify tumor-infiltrating lymphocytes (TILs) in gastrointestinal (GI) cancers, including esophagus, stomach, colon, rectum, pancreas, and liver cancers. The study, which included 1700 patients from four different sites, aimed to evaluate the prognostic significance of computational pathology features. The researchers used top prognostic features identified by a Least Absolute Shrinkage and Selection Operator (LASSO) Cox model to train a model that distinguished between 'low-risk' and 'high-risk' patients. The study found that patients identified as 'high-risk' had significantly poorer overall survival rates compared to those identified as 'low-risk'.

Key Takeaways:

  • The study included 1700 patients from four different sites and evaluated the prognostic significance of computational pathology features relating to spatial arrangement and diversity in the appearance of TILs and cancer nuclei across five different types of GI cancers.
  • The researchers used computer vision and machine learning approaches to evaluate the prognostic significance of computational pathology features relating to spatial arrangement and diversity in the appearance of TILs and cancer nuclei.
  • The study focused on evaluating the prognostic significance of spatial relationships between TILs and the closest cancer nuclei, as well as tumor nuclei shape and texture features captured within local cellular clusters.
  • The trained model identified that patients with 'high-risk' features had significantly poorer overall survival rates compared to those with 'low-risk' features, with hazard ratios (HR) ranging from 1.81 to 5.85 across different types of GI cancers.
  • The study concluded that the spatial relationships of TILs and cancer nuclei are prognostic of survival across multiple GI cancer types.

Statistics:

  • 1700 patients were included in the study across four different sites.
  • The top 9 features were selected by LASSO Cox model from a total of 2236 pathomic features extracted from hematoxylin-eosin stained whole slide images.
  • Patients with 'high-risk' features had significantly poorer overall survival rates, with HRs ranging from 1.81 to 5.85 across different types of GI cancers.
  • The model yielded an HR of 2.32 (95% CI 1.67-3.23, P <0.0001) in external validation sets of colorectal cancer (CRC) patients.
  • The study found that the spatial relationships of TILs and cancer nuclei are prognostic of survival across multiple GI cancer types.

Sources:

  • "Artificial intelligence defines spatial patterns of tumor-infiltrating lymphocytes highly associated with outcome - a pan-GI cancer study." ESMO Open, 2025;10(10):105757.
  • Emory University, Dept. of Pathology and Laboratory Medicine, Atlanta, United States.
  • ESMO Open, Elsevier, Radarweg 29, 1043 Nx Amsterdam, Netherlands.