Linking features of the tumor microenvironment (TME) to patient outcome is ambitious yet crucial to understand cancer progression and prognosis in order to develop new therapies. The current method to assess histopathological images is visual inspection by a professional, generally a pathologist. While this is an effective approach for diagnosis, it does not scale well for discovery biology and population-based studies for drug discovery purposes. An automated approach would allow for a more rapid, systematic, and comprehensive analysis of several morphological features of the TME by taking advantage of thousands of already existing slides within databases and biobanks. Recent studies have shown intriguing relationships between innervation in tumors (tumor exoneural biology) and patient outcomes. Here we present an automated tool that can detect and quantify nerve presence in tumors. We manually annotated a set of digital slides from The Cancer Genome Atlas (TCGA) in order to develop a deep learning model to quantify the presence of nerves in head and neck tumors stained with H&E. This tool is generalizable and was applied to identify patterns in the appearance of tumor-infiltrating lymphocytes (TILs) in and around tumors. It may be further applied to other structural features such as blood vessels in order to characterize and correlate these features within the TME in hundreds of images across cancer types. This enables integration of imaging features with multi-omics data to uncover potential biological pathways that are upregulated in groups with dense innervation compared to sparse innervation in cancer. The main advantage of this approach is the ability to utilize many public databases in which the features of interest can be correlated with reported patient comorbidities, treatments, and phenotypes. This platform-based methodology can be expanded to other disease areas and could ultimately provide valuable insight about exoneural biology and its role in disease physiology to identify new avenues for therapies.
Citation Format: Alison R. Miller, Daniel Krasnonosenkikh, Monica Thanawala, Kai Chih Huang, George V. Thomas, Alexandra B. Lantermann, Hongyue Dai, Masoud Sadaghiani, John A. Wagner, Pearl Huang. Automated nerve identification in histopathology slides enables comprehensive analysis of innervation in cancer and tumor neurobiology [abstract]. In: Proceedings of the AACR-NCI-EORTC Virtual International Conference on Molecular Targets and Cancer Therapeutics; 2021 Oct 7-10. Philadelphia (PA): AACR; Mol Cancer Ther 2021;20(12 Suppl):Abstract nr P262.