123P - Artificial intelligence (AI)-powered analysis of human epidermal growth factor receptor-2 (HER2) and tumor-infiltrating lymphocytes (TILs) in advanced biliary tract cancer (BTC)

Background

HER2 is overexpressed in a subset of BTC patients, and recent developments in HER2-targeted agents are expanding therapeutic options. However, accurately assessing HER2 expression is challenging in BTC, especially at a low-expression level. We evaluated HER2 expression as well as TILs density through AI models in BTC.

Methods

Whole-slide images (WSI) of pre-treatment HER2 immunohistochemistry (IHC) slides, matched H&E-stained slides and clinicopathological data were obtained from advanced or metastatic BTC patients at CHA Bundang Medical Center in South Korea, between 2019 and 2022. An AI-powered HER2 analyzer classifies tumor cells by HER2 staining intensity (H0, H1, H2, or H3) and quantifies the cells in each staining category. Then, it calculates HER2 categories (0/1+/2+/3+) following the latest ASCO/CAP HER2 IHC guidelines. For evaluating immune landscape, a separate WSI analyzer identified and quantified. TILs within the cancer epithelium or stroma from H&E-stained slides.

Results

In a total of 328 patients, the AI-scored HER2 categories were 0 in 92 (28.9%), 1+ in 137 (43.1%), 2+ in 68 (21.4%), and 3+ in 21 (6.6%). The overall concordance of HER2 categories between the AI and pathologists’ readings was 75.3%. HER2 3+ rates by AI in intrahepatic, extrahepatic, gallbladder, and Ampulla of Vater are estimated to be 3.1%, 3.4%, 21.4%, and 5.9%, respectively. HER2-expressing tumors (1+, 2+, and 3+) have significantly higher densities of stromal TILs (983.2 ± 1347.5 / mm2) than HER2-negative tumors (683.5 ± 739.7 / mm2, p=0.012) and there were no significant TIL density differences between HER2-high expressing (3+) and low expressing tumors (1+, 2+), which were consistent with either by the AI or pathologists.

Conclusions

AI-powered HER2 scoring showed good agreement with pathologists’ scoring, and HER2-expressing tumors harbored high stromal TIL in BTC. Objective expression analysis and the associated differences in immunogenicity will be important for future treatment strategies.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Lunit Inc.

Disclosure

All authors have declared no conflicts of interest.