Deep Learning Spatial Profiling of CD103⁺CD8⁺ T Cells and Survival in Rectal Cancer After Neoadjuvant Chemoradiotherapy

Abstract Background CD8⁺ tumor-infiltrating lymphocytes (TILs) are established prognostic markers in colorectal cancer, yet the clinical significance of CD103⁺CD8⁺ tissue-resident memory–like (T RM -like) T cells in locally advanced rectal cancer (LARC) after neoadjuvant chemoradiotherapy (NACRT) remains unknown. Methods We quantified CD8⁺ and CD103⁺CD8⁺ T-cell densities in stromal and intratumoral compartments of post-NACRT resection specimens from 40 LARC patients using Cu-Cyto, a deep learning–based imaging cytometry platform. Associations with survival, pathological response, and adjuvant chemotherapy (AC) were examined. Treatment-induced T-cell dynamics were assessed in paired pretreatment biopsies and post-NACRT resections ( n = 9). Results High stromal CD103⁺CD8⁺ density independently predicted better 5-year RFS (67.4% vs. 12.1%, p < 0.001) and OS (80.0% vs. 26.6%, p = 0.016); intratumoral density showed no prognostic significance. Pathological response correlated with stromal CD8⁺ but not CD103⁺CD8⁺ density. Paired analysis revealed a selective non-expansion of the CD103⁺ subset: stromal CD8⁺ T cells increased significantly after NACRT while CD103⁺CD8⁺ density remained unchanged. AC may preferentially benefit patients with low stromal CD103⁺CD8⁺ density. Conclusions Stromal CD103⁺CD8⁺ T-cell density is a robust independent prognostic biomarker in rectal cancer after NACRT that appears to reflect pre-existing rather than treatment-induced immunity. Given its stability across NACRT, pretreatment biopsy assessment may provide equivalent prognostic information, with potential implications for patient stratification before treatment initiation.