Artificial intelligence (AI) –powered spatial analysis of tumor-infiltrating lymphocytes (TILs) for prediction of prognosis in resectable pancreatic adenocarcinoma (PDAC).

Abstract

4162 Background: The degree of TIL infiltration in the tumor microenvironment has been suggested as a prognostic factor in multiple cancer types, but the significance in PDAC is not well known. The aim of this study is to assess the prognostic role of AI-based spatial TIL density assessment and immune phenotype (IP) classification in resectable PDAC. Methods: We collected hematoxylin and eosin (H&E)-stained whole slide images (WSI) of tumor tissues and clinical data from stage I – III PDAC patients who received curative surgery at Samsung Medical Center, in Seoul, Korea, between Jan 2017 and Dec 2018. For spatial TIL analysis, we used Lunit SCOPE IO, an AI-powered pathology slide analyzer that can identify TIL and segment tumor epithelium and stroma from H&E-stained WSI. It then estimates intratumoral (iTIL) and stromal TIL (sTIL) densities, as well as IP classification of each case. IPs were defined as follows: inflamed IP (IIP) as high iTIL and sTIL; immune-excluded as low iTIL and high sTIL; immune-desert as low TIL overall. Predetermined cutoffs prior to this study were used for the determination of IPs. Results: A total of 209 patients were identified, 182 treated with upfront surgery and 27 with preoperative chemo(radio)therapies followed by surgery. The median recurrence-free survival (RFS) and overall survival (OS) of all patients was 16.8 months (95% CI 12.1 - 21.4), and 36.8 months (95% CI 31.2 - 42.4), respectively. In the 182 patients treated with upfront surgery, the median iTIL and sTIL densities were 85.3/mm2 (IQR 56.9 – 133.0) and 672.2/mm2 (IQR 445.9 – 912.9). 85.7% of the cases were immune-excluded, while 11.5% were IIP. The patients having iTIL higher than the median exhibited longer RFS (HR 0.62, 95% CI 0.42 - 0.92, p = 0.017). Also, the patients having IIP exhibited longer OS (p = 0.045, HR 2.05, 95% CI 1.00 – 4.22) and a marginally longer RFS (p = 0.051, HR 2.03, 95% CI 0.98 – 4.18) compared to those with non-IIP. In the 27 patients who were treated with preoperative therapies, the iTIL densities in the surgical specimens were significantly higher compared to the 182 untreated samples (mean iTIL 168.8/mm2 after chemo(radio)therapies vs. 112.6/mm2 in upfront surgery, p = 0.012), although the increase in the proportion of IIP was not significant. Using the iTIL median of the 182 upfront surgery cases, 17/27(63.0%) patients had high iTIL in WSIs obtained after preoperative therapies, and these patients were observed to have a significantly better RFS (HR 0.26. 95% CI 0.09 - 0.73, p = 0.006). Conclusions: While most cases of resectable PDAC were found to have TIL infiltration mainly in stroma, relatively higher iTIL infiltration was associated with better prognosis. An increase in iTIL after chemo(radio)therapies may be associated with better outcomes.