Abstract 898: Deep learning-based cell types scores in tumor microenvironment estimated by H&E images associated with PD-L1 status in lung adenocarcinoma

Abstract

Abstract Purpose: The spatial distribution of cell types in the tumor microenvironment (TME) is associated with functional status of tumor immunology and eventually affects the response to immuno-oncology treatment. We have developed a deep learning model that was trained by integrating H&E images of lung adenocarcinoma with spatial transcriptomic data. We applied this model to predict various cell types within the TME using only hematoxylin and eosin (H&E) images to correlate with the PD-L1 status. Methods: A deep learning model to predict five cell types enrichment score maps from H&E images was trained by spatial transcriptomics data combined with matched H&E images. The five cell types of TME included B cells, T/NK cells, myeloid cells, fibroblasts, and epithelial cells. wAnother dataset based on tissue microarray (TMA) data of H&E images of lung cancer patients (n = 94) was used to predict cell type enrichment scores associated with PD-L1 status estimated by TPS score. The cell type scores derived from the H&E-stained lung cancer images were correlated with the PD-L1 status. Results: The cell types predicted by the model using H&E image patches showed a significant correlation with those determined through spatial transcriptomic data, serving as an internal validation. All cell types (B cells, T/NK cells, myeloid cells, fibroblasts, and epithelial cells) from TMA cores showed significant differences according to the PD-L1 expression groups. The enrichment scores of B cells, T/NK cells, myeloid cells and fibroblasts were significantly higher in the PD-L1 high group. Conclusions: Our research introduces a deep learning model for precise cell type mapping within the tumor microenvironment and applied to H&E-stained TMA cores. The relationship between PD-L1 status and the cell type enrichment scores within the tumor microenvironment, as predicted by the deep learning model analyzing H&E images, demonstrates the potential for using H&E-based characterization of the tumor microenvironment as a biomarker in immuno-oncology treatments. Citation Format: Seungho Cook, Haenara Shin, Jaemoon Koh, Hongyoon Choi, Kwon Joong Na. Deep learning-based cell types scores in tumor microenvironment estimated by H&E images associated with PD-L1 status in lung adenocarcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 898.