Search

Background: Recently, anti-PD-L1 in combination with cytotoxic chemotherapy has shown significant survival benefit in a randomized phase 3 trial for unresectable or metastatic biliary tract cancer (BTC). However, no biomarker including PD-L1 expression has been established to predict clinical outcomes, and there is an unmet need for a novel predictive biomarker for anti-PD-1 or PD-L1 therapy. Here, we assessed TILs using artificial intelligence (AI)-powered spatial analysis in advanced BTC treated with anti-PD-1 beyond 1st line treatment. Methods: An AI-powered whole-slide image (WSI) analyzer (Lunit SCOPE IO, Lunit, Seoul, Korea) was used to segment tumor epithelium and stroma, and identification of intratumor TIL (iTIL) and stromal TIL (sTIL). H&E stained WSI from pre-treatment samples was acquired from Asan Medical Center (n =166), and a total of 154 samples (92.8%) after quality control were used for the final analysis. Immune phenotypes (IP) were defined as follow: inflamed as high iTIL and sTIL; immune-excluded as low iTIL and high sTIL; immune-desert as low TIL overall. Among them, 20 patients were available for multi-color flow cytometry analysis (FACS) using peripheral blood mononuclear cells, collected at baseline, C1D8, and C2D1. Results: All patients (n=154) were treated with anti-PD-1 (pembrolizumab or nivolumab) monotherapy, and 72 of 154 patients (46.8%) were treated as 2nd line. Gemcitabine plus cisplatin (GemCis) was used prior to anti-PD-1 as first-line therapy in all patients. Overall, 15 (9.7%) patients showed inflamed IP. With median follow-up duration of 15.4 months, the inflamed IP group showed better overall survival (17.2 vs. 6.6 months, P=0.03), and progression-free survival (PFS; 4.5 vs. 2.6 months, P=0.09) along with higher PFS rate at 12 months (33.3% vs. 11.5%, P=0.035), and overall response rate (26. 7% vs. 8.6%, P=0.053) than other phenotype groups. There was no significant difference in median PFS with GemCis among IP groups (P=0.74). In the FACS available subgroup, inflamed IP showed higher baseline central memory T (Tcm)+effector memory T (Tem)/Tnaive ratio than other IPs. With the administration of anti-PD-1, Tcm+Tem/Tnaive ratio was increased, while the proportion of PD1+CD8+T, CD39+CD8+T, CD103+CD8+T and Treg were decreased in the inflamed IP group than other phenotype groups. Conclusions: Immune phenotypes classified by AI-powered spatial TIL analysis was effective to predict the clinical outcomes of patients with advanced BTC treated with anti-PD-1 therapy. Citation Format: Yeong Hak Bang, Kyunghye Bang, Jin Ho Shin, Hyunseok Yoon, Kyu-Pyo Kim, Inkeun Park, Jae Ho Jeong, Heung-Moon Chang, Baek-Yeol Ryoo, Chiyoon Oum, Seulki Kim, Yoojoo Lim, Gahee Park, Chan-Young Ock, Changhoon Yoo. Spatial analysis of tumor-infiltrating lymphocytes in tumor microenvironment as biomarker for immune checkpoint inhibition in biliary tract cancer. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4333.

601 Background: Anti-PD-(L)1 monotherapy or combination with bevacizumab or ipilimumab are approved treatment options in HCC. Nevertheless, not all patients benefit from IO treatment, and discovery of predictive biomarkers for IO outcome has not been fully investigated. Immune phenotype analysis by AI-powered spatial tumor-infiltrating lymphocyte (TIL) analyzer in H&E whole-slide images (WSI), Lunit SCOPE IO, has shown predictive effect across multiple cancer types. Here, we investigated the correlation between immune phenotype (IP) and the real-world outcomes in HCC. Methods: Real-world dataset of 177 patients with HCC were retrospectively collected from CHA Bundang Medical Center. Inflamed score was defined by the proportion of area with high intra-tumoral TIL density per 0.5 mm2-sized grid. IP status was sub-analyzed by PD-L1 combined positive score (CPS) and prognostic factors including Child-Pugh class and Barcelona Clinic Liver Cancer (BCLC) stage. We evaluated the correlation between IP and IO outcome by regimen (atezolizumab+bevacizumab [Atez+Bev], n=100; nivolumab+/-ipilimumab [Niv+/-Ipi], n=77). Results: The majority of patients (75.3%) were Child-Pugh class A (75.3%) and BCLC stage C (86.5%), respectively. Atez+Bev was mostly received as first-line therapy (92.0%), whereas Niv+/-Ipi as second-line or beyond (98.7%). The inflamed score was significantly higher in patients who had lymph node metastasis ( p=0.008), but it was not correlated with other metastatic sites, or any other prognostic factors. Of 177 patients, 105 (59.3%) patients were classified as having a high inflamed score (≥10%) and 72 (40.7%) patients as having lower than 10%. PD-L1 CPS was significantly higher in the sample with high score compared to those with low inflamed score ( p=0.01387). PFS of Niv+/-Ipi were significantly favorable in the patients with high inflamed score (3-m PFS rate; 44.0% vs 3.61%, 12-m PFS rate; 17.8% vs 0%, Hazard ratio [HR] 0.37, 95% confidence interval [CI] 0.22-0.63, p

e12543 Background: Human epidermal growth factor receptor 2 (HER2) expression is a predictive marker for HER2-targeted therapy in breast cancer patients. Interobserver variation in the interpretation of HER2 levels exists among pathologists, thus a method to increase the consistency of evaluation is needed. This study aimed to evaluate the performance of the artificial intelligence (AI)-based Lunit SCOPE HER2 in assisting pathologists to evaluate HER2 expression levels in breast cancer. Methods: Lunit SCOPE HER2 was developed with a 1.04 x 1010 μm2 area and 7.31 x 105 tumor cells from 1,133 HER2 immunohistochemistry stained whole-slide images (WSI) of breast cancer, annotated by 113 board-certified pathologists. The AI model was developed based on a semantic segmentation algorithm, which consists of two atrous spatial pyramid pooling blocks for tissue level classification and for tumor cell level classification. To validate the model, a total of 209 HER2 WSIs diagnosed with breast cancer were obtained from Kyung Hee University Hospital in Korea and were assigned as an external validation set. Three board-certified pathologists evaluated slide level HER2 expression (3+, 2+, 1+, and 0) twice, first without AI assistance and second, with it. The second reading was performed for WSIs where the pathologist's reading showed discrepancy with the AI model. Results: In the external validation set, all pathologists scored the same HER2 grade in 103 WSIs (49.3%), and the Fleiss kappa value was 0.512. The HER2 grade from the AI model and pathologists was the same in 151 WSIs (72.2%), and the weighted kappa value was 0.844. The pathologists re-evaluate 43, 63, and 83 WSIs, respectively. After AI assistance, all pathologists scored the same HER2 grade in 156 WSIs (74.6%), and the Fleiss kappa value increased to 0.762 (Table). Conclusions: This study demonstrates that an AI-powered HER2 analyzer can help achieve consistent HER2 expression level evaluation in breast cancer by reducing interobserver variability. Thus, the AI model can be applied as an assistance tool for pathologists in HER2 grade evaluation.[Table: see text]