Quantitative image profiling of the tumor microenvironment on double stained immunohistochemistry images using deep learning.

• DOI: 10.1200/jco.2019.37.15_suppl.e14619
• Publication Date: 2019-05-27T16:00:33Z
• Authors: Mate Levente Nagy; Anna Juncker-Jensen; Ben Ovadia; Robert Smale; Karen Yamamoto; Josette William; Nicholas Hoe; Raghav Padmanabhan
• Keywords: 03 medical and health sciences; 0302 clinical medicine; 3. Good health

e14619 Background: Spatial locations of immune cells in the tumor microenvironment (TME) correlate with clinical outcome cancers. A worse patient has been reported oral squamous cancer for individuals an increased number regulatory T within 30µm CD8+ [1]. The spatial relationship between and PD-L1+ become area interest understanding PD-L1 inhibition. combined assessment CD8 NSCLC tumors was shown to outperform or alone as prognostic markers predicting treatment checkpoint inhibitors [2]. Methods: As these cases, quantifying relationships two biomarkers is valuable providing insights. analytics TME requires accurate cell segmentation classification. To that end, NeoGenomics developed a deep learning pipeline automatically segment classify from whole slide double stained IHC images. Results: In this study, we performed staining classification analysis on sets assays (CD8+ FoxP3+) by modifying our MultiOmyx [3]. addition image outputs, also generate tables morphological information, phenotype counts densities, biomarker intensities can be used define H-score-like measures. Advanced identify clustering patterns various phenotypes. These analyses enable investigation complex interactions TMEs. Conclusions: quantitative assay compatible any even if they are expressed same long sub-cellular localization different (membrane CD8, nuclear FoxP3). combination FoxP3+ capable needed context guide decisions. References: Feng Z et al., JCI Insight. 2017 Jul 20:e93652 Steele KE J Immunother Cancer 2018;6:20 Nagy ML AACR Annual Meeting 2018; April 14-18, Chicago, IL