使用AI对非小细胞肺癌中肿瘤浸润淋巴细胞进行定量评估及其与免疫治疗反应的关联

Tumor-infiltrating lymphocytes (TILs) have been associated with treatment outcomes in non-small cell lung cancer (NSCLC), but challenges in visual and digital assessment have limited their clinical use. In this study, we used a quantitative AI-based approach on digitized H&E-stained sections to quantify intratumoral (iTIL) and stromal (sTIL) TILs and to evaluate optimal digital cutoff points associated with clinical outcomes across NSCLC cohorts treated with non-neoadjuvant and neoadjuvant immunotherapy. Two independently collected, annotated retrospective NSCLC cohorts from surgical resections treated with immune checkpoint inhibitors (ICIs) were included: the Yale 471 cohort (training set, n=42) and the Yale 592 cohort (validation set, n=41). Quantification of iTIL and sTIL was performed on whole-tissue H&E sections using the SCOPE IO™ platform developed by Lunit. A cutoff associated with clinical outcome was identified in the training cohort and applied to the validation cohort to assess associations with progression-free survival (PFS) and overall survival (OS). The two cohorts were then combined into a larger non-neoadjuvant training set (n=83). An optimal cutoff point was applied to the neoadjuvant validation set collected from pretreatment transbronchial and needle-core biopsies of patients with surgical resection after neoadjuvant therapy (n=76). This neoadjuvant cohort provided a clinically enriched population for assessing associations with PFS and OS. In the Yale 471 training cohort, higher sTIL scores were significantly associated with longer progression-free survival (PFS) at an optimal cutoff point of 2.4% (HR = 2.7). This cutoff remained significant in Yale 592 (HR = 5.8; P < 0.0001). No significance was found when assessing iTIL in this manner. In the non-neoadjuvant training set, higher iTIL scores were linked to longer PFS with an optimal cutoff point of 0.28%, dividing patients into equal groups (HR = 0.36). The neoadjuvant cohort showed significance (HR = 0.28; P = 0.0045). Higher sTIL scores were also associated with improved PFS in the non-neoadjuvant training cohort using the same 2.4% cutoff point, splitting patients into 70% high and 30% low groups (HR = 0.21). The neoadjuvant validation set showed significance (HR = 0.34; P = 0.0066). Although higher iTIL and sTIL values showed consistent trends for overall survival (OS) in the training sets, these associations did not reach statistical significance. Quantitative assessment of tumor-infiltrating lymphocytes (TILs) may provide an objective biomarker for immunotherapy response in NSCLC, particularly for intratumoral TILs (iTIL), which are not assessable by pathologists. Future work is planned to include additional analyses of neoadjuvant and prospective NSCLC cohorts to further validate associations between TIL assessment and clinical outcomes.