AI-powered spatial tumor microenvironment analysis in metastatic microsatellite-stable colorectal cancer receiving immunotherapy.

2532 Background: Microsatellite-stable (MSS) colorectal cancer (CRC) is typically immune-cold, limiting the routine use of immune checkpoint inhibitors (ICIs). However, a subset of MSS CRC patients may still derive benefit from ICIs. Leveraging an AI-powered whole-slide image (WSI) analyzer applied to hematoxylin and eosin (H&E)–stained slides, we sought to identify subsets of MSS CRC more likely to benefit from ICIs. Methods: Between August 2016 and May 2024, pretreatment H&E–stained WSIs were collected from patients with metastatic MSS CRC treated at Asan Medical Center, Seoul, Korea. After quality control, 91 WSIs (93.8%) from 51 patients were included in the final analysis. Patients were treated with anti-PD-1/PD-L1 ICIs in clinical trials, with the majority treated in the third-line setting (n = 46, 90.2%). An AI-powered WSI analyzer (Lunit SCOPE IO, Lunit, Seoul, Korea) segmented cancer area (CA) and stromal area and identified tumor-infiltrating lymphocytes (TILs), tertiary lymphoid structures (TLS), fibroblasts, and endothelial cells within tumor tissue. We then evaluated the associations between CA-specific densities of TILs, macrophages, and endothelial cells, as well as the summed TLS area, and progression-free survival (PFS) and overall survival (OS). Results: Using maximally selected rank statistics based on PFS, patients were dichotomized into high and low groups according to the summed TLS area and the densities of TILs, macrophages, endothelial cells, and fibroblasts within the CA. Notably, patients with a high summed TLS area had significantly improved PFS and OS compared with those with a low TLS area (median PFS, 3.6 vs. 1.6 months; P=0.026; median OS, 11.6 vs. 6.6 months; P=0.024). High endothelial cell and fibroblast densities within the CA were not significantly associated with PFS (P=0.160, P=0.112, respectively), but were associated with worse OS, with a significant association for endothelial cells (median, 7.6 vs. 14.7 months; P=0.024) and a trend toward worse OS for fibroblasts (median, 8.9 vs. 12.2 months; P=0.056). Higher macrophage density within the CA showed a favorable trend toward improved PFS (median, 10.2 vs. 8.5 months, P=0.092), but was not associated with OS (P=0.399). In contrast, none of these biomarkers were associated with clinical outcomes in the first-line chemotherapy. Conclusions: Despite microsatellite-stable status, a higher summed TLS area was associated with improved progression-free and overall survival following immunotherapy, with endothelial cell and fibroblast densities providing additional prognostic information. AI-powered WSI analysis enabled effective stratification of these features.