Determining breast cancer biomarker status and associated morphological features using deep learning

Paul Gamble,James H Wren,Carrie Robinson,David F Steiner,Michael S Toss ,Hongwu Wang,Fraser Elisabeth Tan ,Isabelle Flament-Auvigne,Trissia Brown,Lily Peng ,Yun Liu,Niels Olson,Ronnachai Jaroensri,Emad A Rakha,Craig H Mermel,David J Dabbs,Melissa Moran,Greg S Corrado,Peter Regitnig,Po-Hsuan Cameron Chen

doi:10.1038/s43856-021-00013-3

Abstract

BackgroundBreast cancer management depends on biomarkers including estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 (ER/PR/HER2). Though existing scoring systems are widely used and well-validated, they can involve costly preparation and variable interpretation. Additionally, discordances between histology and expected biomarker findings can prompt repeat testing to address biological, interpretative, or technical reasons for unexpected results.MethodsWe developed three independent deep learning systems (DLS) to directly predict ER/PR/HER2 status for both focal tissue regions (patches) and slides using hematoxylin-and-eosin-stained (H&E) images as input. Models were trained and evaluated using pathologist annotated slides from three data sources. Areas under the receiver operator characteristic curve (AUCs) were calculated for test sets at both a patch-level (>135 million patches, 181 slides) and slide-level (n = 3274 slides, 1249 cases, 37 sites). Interpretability analyses were performed using Testing with Concept Activation Vectors (TCAV), saliency analysis, and pathologist review of clustered patches.ResultsThe patch-level AUCs are 0.939 (95%CI 0.936–0.941), 0.938 (0.936–0.940), and 0.808 (0.802–0.813) for ER/PR/HER2, respectively. At the slide level, AUCs are 0.86 (95%CI 0.84–0.87), 0.75 (0.73–0.77), and 0.60 (0.56–0.64) for ER/PR/HER2, respectively. Interpretability analyses show known biomarker-histomorphology associations including associations of low-grade and lobular histology with ER/PR positivity, and increased inflammatory infiltrates with triple-negative staining.ConclusionsThis study presents rapid breast cancer biomarker estimation from routine H&E slides and builds on prior advances by prioritizing interpretability of computationally learned features in the context of existing pathological knowledge.

Highlights

Breast cancer management depends on biomarkers including estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 (ER/PR/HER2)
We developed a separate deep learning systems (DLS) for each biomarker (ER, PR, and HER2) to enable exploration of the possibility that different morphological features might be associated with each biomarker
The second stage of the DLS predicts the slide-level biomarker status using the predictions of the first stage across every patch in the slide (Fig. 1, Methods)

Summary

Methods

We developed three independent deep learning systems (DLS) to directly predict ER/PR/HER2 status for both focal tissue regions (patches) and slides using hematoxylin-andeosin-stained (H&E) images as input. Models were trained and evaluated using pathologist annotated slides from three data sources. Interpretability analyses were performed using Testing with Concept Activation Vectors (TCAV), saliency analysis, and pathologist review of clustered patches. Results The patch-level AUCs are 0.939 (95%CI 0.936–0.941), 0.938 (0.936–0.940), and 0.808 (0.802–0.813) for ER/PR/HER2, respectively. AUCs are 0.86 (95% CI 0.84–0.87), 0.75 (0.73–0.77), and 0.60 (0.56–0.64) for ER/PR/HER2, respectively. De-identified breast cancer data for this study came from three sources (Table 1): a tertiary teaching hospital, a medical laboratory, and TCGA13, 14. The medical laboratory contributed only tissue blocks and TCGA represents only archival H&E slides. Inclusion criteria for the H&E images required the presence of invasive carcinoma in primary breast tissue specimens, as determined by pathologist review

Results

Discussion

Conclusion