Abstract 3205: Artificial intelligence (AI) -powered H&E whole-slide image (WSI) analysis of tertiary lymphoid structure(TLS) to predict response to immunotherapy in non-small cell lung cancer (NSCLC)

Background: The efficacy of ICIs in treating NSCLC is limited to a subset of patients, necessitating reliable predictive biomarkers. TLSs within the tumor microenvironment have emerged as potential indicators. This study aims to investigate the association between an AI-powered TLS analysis and treatment response of NSCLC patients treated with ICIs. Method: We collected H&E-stained WSI of primary and metastatic tumors and clinical data in a retrospective study of advanced-stage NSCLC patients treated with ICI (with or without chemotherapy) at at Northwestern University from May 2015 to November 2023. Following the assessment of the presence of TLS by a pathologist, the identical WSIs were evaluated by the AI-powered H&E WSI analyzer. We used Lunit SCOPE IO, an AI-powered H&E WSI analyzer developed using 21,096 H&E stained WSIs, which can segment various classes of tissue area including TLS within the TME. Survival analysis was performed using Kaplan-Meier curves, and differences in survival outcomes among TLS groups were evaluated using the log-rank. Result: A total of 102 patients with advanced-stage NSCLC were included in the study. 30.3% (31/102) and 22.5% (23/102) of cases were found to contain TLS as determined by the AI analyzer and a pathologist respectively. The AI model achieved an accuracy of 88.2% (95% confidence interval [CI], 80.4%-93.8%) in identifying the presence of TLS compared to the pathologist’s interpretation. The concordance between the pathologist's assessment and the AI model, as evaluated by Cohen’s kappa, was 0.70. The median follow-up duration was 40.0 months. Based on TLS presence as assessed by a pathologist, the TLS (+) group demonstrated significantly improved PFS (HR: 0.29, 95% CI: 0.12-0.68, p < 0.01) and OS (HR: 0.43, 95% CI: 0.23-0.78, p < 0.01) compared to the TLS (-) group. When TLS presence was assessed by an AI analyzer, the TLS (+) group had a median PFS (mPFS) that was not reached (95% CI: 31 months-not reached), whereas the TLS (-) group had a mPFS of 11 months (95% CI: 6-28 months), with an HR of 0.39 (95% CI: 0.20-0.79, p < 0.01). The median OS (mOS) for the TLS (+) group was 26.2 months (95% CI: 17 months-not reached) compared to 13 months (95% CI: 9-21 months) in the TLS (-) group, with an HR of 0.55 (95% CI: 0.33-0.92, p = 0.02). There was no significant different in overall response rate between the two groups (17.7% in TLS (+) and 17.4% in TLS (-)) Conclusions: AI-powered TLS analysis can be potentially utilized as a predictive biomarker of survival outcomes of NSCLC patients treated with ICI. Citation Format: Young Kwang Chae, Leeseul Kim, Sangwon Shin, Sooick Cho, Sanghoon Song, Borislav Alexiev, Behtash Nezami. Artificial intelligence (AI) -powered H&E whole-slide image (WSI) analysis of tertiary lymphoid structure(TLS) to predict response to immunotherapy in non-small cell lung cancer (NSCLC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3205.