Use of artificial intelligence-powered spatial analysis of tumor microenvironment to predict the prognosis in resected gallbladder cancer
Young Hoon Choi, Hyemin Kim, So Jeong Yoon, Yeong Hak Bang, Kee-Taek Jang, Changhoon Yoo, Chang Ho Ahn, Soohyun Hwang, Sangwon Shin, Sang Hyun Shin, In Woong Han, Jin Seok Heo, Kwang Hyuck Lee, Jong Kyun Lee, Se-Hoon Lee, Kyu Taek Lee, Hongbeom Kim, Joo Kyung Park
ASCO, 2025
Background:
Gallbladder cancer (GBC) is a highly lethal disease with a lack of reliable biomarkers. The tumor microenvironment (TME) is closely associated with prognosis, but its clinical application as a prognostic marker is limited by evaluation challenges. This study assessed the prognostic significance of AI-powered TME analysis in resected GBC patients.
Methods:
A total of 225 GBC patients with an R0 resection were enrolled, and their hematoxylin & eosin (H&E)-stained GBC sections were analyzed using Lunit SCOPE IO, an artificial intelligence (AI)-powered whole-slide image (WSI) analyze, to evaluate TME-related features, including tumor-infiltrating lymphocyte (TIL) density, fibroblast (FB) density, and tertiary lymphoid structure (TLS) counts. Risk stratification was based on TME-related risk factors (low TIL, high FB, low TLS), and survival outcomes were assessed. External validation was conducted using 146 biliary tract cancer patients.
Results:
Overall survival (OS) and disease-free survival (DFS) declined as the number of TME-related risk factors increased. Patients with three risk factors had the poorest outcomes (median OS: 17.7 months [reference]; median DFS: 12.7 months [reference]), followed by those with two risk factors (median OS: 115.9 months, HR = 0.40, 95% CI: 0.19–0.85; median DFS: 57.8 months, HR = 0.37, 95% CI: 0.18–0.74) and one risk factor (median OS: 126.5 months, HR = 0.34, 95% CI: 0.16–0.74; median DFS: 117.2 months, HR = 0.30, 95% CI: 0.15–0.62). Patients with no risk factors had the best survival (median OS: not reached, HR = 0.20, 95% CI: 0.06–0.67; median DFS: not reached, HR = 0.13, 95% CI: 0.04–0.41). External validation confirmed consistent trends across all risk groups.
Conclusions:
AI-powered TME analysis shows promise as a practical tool for identifying TME-related risk factors using H&E-stained WSI, providing valuable prognostic information for resected GBC patients.