Proceedings: AACR Annual Meeting 2019; March 29-April 3, 2019; Atlanta, GA
Introduction: Tumor purity and deconvolution of cell subsets in tumor environments has been developed based on DNA sequencing data or gene expression profiles. We have developed the artificial intelligence named Lunit SCOPE, identifying and quantifying various histologic components using deep learning from H&E-stained whole slide images. The Lunit SCOPE can define the proportions of cancer epithelium, cancer stroma and immune cells infiltration. We hypothesized that cell proportions analyzed by Lunit SCOPE can accurately predict cell subsets in tumor microenvironment with biologically-reliable results.
Method: Deep learning-based H&E image analyzer, Lunit SCOPE, has been trained by 1,191 H&E-stained breast cancer whole slide images with histological components such as cancer epithelium, cancer stroma, normal cells, fat, necrosis, lymphocyte, and fibroblast, annotated by over 10 pathologists. Lunit SCOPE trained by the annotated dataset can profile the cancer stroma ratio and tumor infiltrating lymphocyte (TIL) density in cancer epithelium and cancer stroma. We have selected 21 types of cancers excluding mesenchymal origin cancer in The Cancer Genome Atlas (TCGA). Finally, H&E slides from 7,837 cases have been analyzed by Lunit SCOPE.
Result: Pan-cancer analysis has shown that the cancer stroma ratio from Lunit SCOPE was the highest in pancreatic adenocarcinoma. Especially, Consensus Molecular Subtype 4 (CMS4) has shown the highest cancer stroma ratio among colorectal cancer (P < 2.2 x 10-16). The density of tumor infiltrating lymphocyte (TIL) was highest in renal clear cell carcinoma, melanoma and bladder cancer in that order, and lung cancer, esophageal cancer, and head and neck cancer have also presented high TIL density. Moreover, high mutational burden (> 10/Mb) was significantly related to higher TIL density across all cancer types (P = 4.06 x 10-5). In addition, the stromal TIL density was higher in basal type than other subtypes of breast cancer (P = 2.45 x 10-7). Pan-cancer survival analysis showed that higher cancer stroma ratio (HR 1.14, 95% CI 1.04-1.25, P = 0.00419) and higher stromal TIL density (HR 1.17, 95% CI 1.07-1.28, P = 0.000732) were associated with poor prognosis, but higher intratumoral TIL density had good prognosis (HR 0.89, 95% CI 0.82-0.98, P = 0.0126).
Conclusion: In this study, TCGA pan-cancer dataset have been analyzed by deep learning-based H&E image analyzer, Lunit SCOPE. As a result, it has been proven that the ratio of cancer stroma is high in pancreatic cancer and CMS4 colorectal cancer, and immunogenic tumors and high mutational burden is related to high TIL density. In summary, Lunit SCOPE has significantly correlated with previously well-defined biological features in pan-cancer analysis. Furthermore, Lunit SCOPE can be developed for the novel tissue-agnostic predictive biomarkers of cancer immunotherapy.
Citation Format: Kyunghyun Paeng, Geunyoung Jung, Sarah Lee, Soo Youn Cho, Eun Yoon Cho, Sang Yong Song. Pan-cancer analysis of tumor microenvironment using deep learning-based cancer stroma and immune profiling in H&E images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 2445.
Kyunghyun Paeng, Geunyoung Jung, Sarah Lee, Soo Youn Cho, Eun Yoon Cho and Sang Yong Song
Abstract: Deep learning-based predictive biomarker for adjuvant chemotherapy in early-stage hormone receptor-positive breast cancer
Abstract: Deep learning-based predictive biomarker for immune checkpoint inhibitor response in metastatic non-small cell lung cancer
Abstract: Comprehensive deep learning analysis of H&E tissue phenomics reveals distinct immune landscape and transcriptomic enrichment profile among immune inflamed, excluded and desert subtypes...
Abstract: Deep-learning analysis of H&E images to define three immune phenotypes to reveal loss-of-target in excluded immune cells as a novel resistance mechanism of immune checkpoint inhibitor...
Abstract: Deep learning-based immune phenotype analysis reveals distinct resistance pattern of immune checkpoint inhibitor in non-small cell lung cancer
Predicting breast tumor proliferation from whole-slide images: The TUPAC16 challenge
Accuracy and efficiency of an artificial intelligence tool when counting breast mitoses
Abstract: Artificial intelligence-powered spatial analysis of tumor infiltrating lymphocytes (TIL) to reflect target gene expressions of novel immuno-oncology agents.
Abstracts: Distinct subset of immune cells assessed by multiplex immunohistochemistry correlates with immune phenotype classified by ...
Abstracts : Pathologic validation of artificial intelligence-powered prediction of combined positive score of PD-L1 immunohistochemistry in urothelial carcinoma.
Abstracts : Clinical performance of artificial intelligence-powered annotation of tumor cell PD-L1 expression for treatment of immune-checkpoint inhibitor (ICI) ...
Abstracts : Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes predicts survival after immune checkpoint inhibitor therapy across multiple cancer types.
Abstracts : Deep learning based radiomic biomarker for predicting the presence of high-grade histologic patterns in lung adenocarcinoma
Abstract : Artificial intelligence-powered tissue analysis reveals distinct tumor-infiltrating lymphocyte profile as a potential biomarker of molecular subtypes in endometrial cancer
Abstract : Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes reveals distinct genomic profile of immune excluded phenotype in pan-carcinoma
Deep learning from HE slides predicts the clinical benefit from adjuvant chemotherapy in hormone receptor-positive breast cancer patients