Abstract 2440: Correlation analysis between AI-based H-score and clinical data in MET IHC stained WSIs

Abstract Background: MET protein expression is considered a potential biomarker in lung cancer, but there is no guideline for analysis, and its clinical significance has not been sufficiently analyzed. In order to explore its potential as a biomarker, it is necessary to analyze the relationship between the expression level of the biomarker and clinical data. In this study, we analyze the relationship between H-score, which is automatically measured based on cell detection detected by a deep learning model, and clinical data. Design: The dataset included 270 whole slide images (WSIs) of non-small cell lung cancer (NSCLC) immunohistochemically stained with a MET antibody (SP44), obtained through resection, biopsy, or endobronchial ultrasound (EBUS). For each of these WSIs, pathologists estimated the proportions for 0, 1+, 2+, and 3+ according to duty, and calculated the H-score based on these values. In addition, the analysis areas were manually specified to exclude the internal controls. We developed analysis models based on deep learning. The cancer area localization model identified the cancer areas for each patch, and the cell analysis model identified the cell membrane, cytoplasm, and nucleus of each cell. Additionally, the MET expression intensity of each cell was classified as 0, 1+, 2+, and 3+ based on the mean diaminobenzidine (DAB) optical density (OD) measured in the cell membrane. The H-score was calculated accordingly. Result: For each WSI, the correlation between the H-score estimated by AI and the pathologist's H-score was calculated as Pearson’s correlation coefficient = 0.6909 (p<0.0001). In addition, when analyzing the H-score estimated by AI by tumor subtype of clinical data, the mean H-score of adenocarcinoma samples was estimated to be 177, which was significantly higher than the mean H-score of 116 estimated for squamous cell carcinoma samples (Mann-Whitney test, p < 0.0001). Conclusion: For MET IHC stained WSIs, H-scores were estimated through cell analysis using deep learning and image analysis, and the correlation with pathologist's H-scores was confirmed, and H-score differences between tumor subtypes were confirmed. Semi-automated AI analysis can be used to analyze the correlation between clinical data and MET staining expression. Citation Format: Hyeon Seok Yang, Yunseob Hwang, DoKyung Kim, Yoon-La Choi, Kyungsoo Jung, Minjung Sung, Young Kee Shin, Ji-Hye Nam, Tae-Yeong Kwak, Sun Woo Kim, Hyeyoon Chang. Correlation analysis between AI-based H-score and clinical data in MET IHC stained WSIs [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 2440.