Abstract:Background: Tissue Microarrays (TMAs) stand to significantly improve the efficiency of multiplex spatial studies by including core samples taken from many tumor blocks in a single slide. However, as TMAs become increasingly prevalent, the implication of core selection within the tumor microenvironment (TME) on downstream discoveries is not well understood. This raises the question of which TME contexts are most predictive for scientific analyses. Methods: We segment whole tissue section (WTS) slides with a U-Net trained on pathologist annotations of tumor, DCIS, LCIS, normal, stromal, immune, and necrotic microenvironment contexts in 100 TCGA BRCA slides. We focus on 5-way PAM50 subtype prediction in breast cancer. We produce image embeddings for each core using a vision transformer pre-trained on histopathology images and use an attention-based aggregation model to predict PAM50 subtype. To evaluate the impact of core TME contexts in a held-out test set, we sample 0.6mm radius circular sections from 281 WTS images from 54 patients to produce one dataset of "majority non-cancerous" cores that the TME annotation model predicts as majority normal breast epithelial or necrotic tissue and one "majority cancerous" TMA dataset for the remaining cores. Results: We find that core microenvironment context directly influences diagnostic accuracy. Maximizing tumor content improves diagnostic accuracy on the PAM50 task, with the majority cancerous TMAs obtaining an AUROC of 0.80 (95% confidence interval 0.75-0.85), relative to 0.72 (95% CI 0.66-0.78) for majority non-cancerous TMAs and 0.82 (95% CI 0.80-0.84) for WTS images. Notably, our context-aware TMA sampling achieves accuracy within 2% of the WTS images, while poorly selected TMA cores underperform the WTS baseline by 10% on average. While TCGA is curated to contain patients with a very high tumor burden, and consequently random TMA selection performs well on this cohort (AUROC 0.77; 95% CI 0.74-0.79), our study indicates that tissue context is an important consideration when selecting cores from real donor tissue blocks. Our results demonstrate that TMA core sampling protocols should be informed by TME context to maximize analysis accuracy, and that intelligent core sampling can largely close the gap to WTS image accuracies on clinically-relevant tasks like PAM50 subtype prediction. These findings highlight the need for thoughtful TMA dataset construction and usage, especially as core selection becomes increasingly automated. We argue both for more targeted core selection from donor tissue blocks and for filtering of existing datasets. To avoid performance degradation, pipelines should identify and exclude these less task-relevant TMEs, where simulation serves as an efficient and task-specific approximation of TME context relevance. Citation Format: Addie Woicik, Zachary R. McCaw, Ben Dulken, insitro Research Team, Christopher Probert. Exploring the role of microenvironment context on diagnostic accuracy in tissue microarray core selection [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular s); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl) nr 6183.

Incorporating deep features in the analysis of tissue microarray images

Learning Low-dimensional Manifolds for Scoring of Tissue Microarray Images

A Deep Learning Approach for Tissue Spatial Quantification and Genomic Correlations of Histopathological Images

Automatically Score Tissue Images Like a Pathologist by Transfer Learning

Error-robust and Label-efficient Deep Learning for Understanding Tumor Microenvironment from Spatial Transcriptomics

Harnessing TME depicted by histological images to improve cancer prognosis through a deep learning system

Prediction of Tumor-Associated Macrophages and Immunotherapy Benefits Using Weakly Supervised Contrastive Learning in Breast Cancer Pathology Images

Computational staining of pathology images to study tumor microenvironment in lung cancer

Applications of machine learning in tumor-associated macrophages

Deep learning-based tumor microenvironment analysis in colon adenocarcinoma histopathological whole-slide images

Deep Representation Learning of Tissue Metabolome and Computed Tomography Images Annotates Non-invasive Classification and Prognosis Prediction of NSCLC

Abstract 6191: Simultaneous prediction of tumor microenvironment biomarkers from pathology slides using multi-task deep regression

Deep Learning Assessment of Tumor Proliferation in Breast Cancer Histological Images

Deep representation learning of tissue metabolome and computed tomography annotates NSCLC classification and prognosis

Sampling Based Tumor Recognition in Whole-Slide Histology Image with Deep Learning Approaches.

Toward Robust Histology-Prior Embedding for Endomicroscopy Image Classification

Long short-term memory model – A deep learning approach for medical data with irregularity in cancer predication with tumor markers

Abstract 6183: Exploring the role of microenvironment context on diagnostic accuracy in tissue microarray core selection

Simultaneously Segmenting and Classifying Cell Nuclei by Using Multi-Task Learning in Multiplex Immunohistochemical Tissue Microarray Sections

TopOC: Topological Deep Learning for Ovarian and Breast Cancer Diagnosis

Characterizing the Survival-Associated Interactions between Tumor-infiltrating Lymphocytes and Tumors from Pathological Images and Multi-omics Data