Epithelium segmentation using deep learning in H&E-stained prostate specimens with immunohistochemistry as reference standard

Abstract

Given the importance of gland morphology in grading prostate cancer (PCa), automatically differentiating between epithelium and other tissues is an important prerequisite for the development of automated methods for detecting PCa. We propose a new deep learning method to segment epithelial tissue in digitised hematoxylin and eosin (H&E) stained prostatectomy slides using immunohistochemistry (IHC) as reference standard. We used IHC to create a precise and objective ground truth compared to manual outlining on H&E slides, especially in areas with high-grade PCa. 102 tissue sections were stained with H&E and subsequently restained with P63 and CK8/18 IHC markers to highlight epithelial structures. Afterwards each pair was co-registered. First, we trained a U-Net to segment epithelial structures in IHC using a subset of the IHC slides that were preprocessed with color deconvolution. Second, this network was applied to the remaining slides to create the reference standard used to train a second U-Net on H&E. Our system accurately segmented both intact glands and individual tumour epithelial cells. The generalisation capacity of our system is shown using an independent external dataset from a different centre. We envision this segmentation as the first part of a fully automated prostate cancer grading pipeline.

Quantifying the effects of data augmentation and stain color normalization in convolutional neural networks for computational pathology

“…The goal of this binary classification task was to identify patches containing epithelial cells in prostate tissue. We trained the classifier with 25 H&E WSIs of prostate resections from the Radboudumc scanned at 0.5 µm/pixel resolution, with annotations of epithelial tissue as described in (Bulten et al (2019)). We split this cohort into training (13), validation (6) and test (6), and extracted a total of 250K patches.…”

“…1). This test set was manually annotated as described in (Bulten et al (2019)) and named prostate-rumc2. We extracted 75K patches from these WSIs.…”

“…We extracted 75K patches from these WSIs. Second, we used publicly available images from 20 H&E slides of prostatectomy specimens with manual annotations of epithelial tissue obtained as described in (Bulten et al (2019); Gertych et al (2015)). We extracted 65K patches from them and named the test set prostate-cedar.…”

Improving the diagnosis and classification of Ph-negative myeloproliferative neoplasms through deep phenotyping

“…Recent advances in computational image analysis have already shown significant promise in various histological diagnosis / classification tasks in common human malignancies. [13][14][15][16][17][18][19] In the latter context, image analysis has even demonstrated the potential to provide robust predictions about key, targetable mutations. 20 Here we describe a computational method of subtyping megakaryocytes based on their cytomorphological features and determining their relative association with an underlying diagnosis of MPN (ET) or a reactive / non-neoplastic condition.…”

Clinically applicable histopathological diagnosis system for gastric cancer detection using deep learning

“…While recent studies have validated the effectiveness of pathology AI for tumor detection in various organ systems, such as lung 20 , stomach 21 , lymph node metastases in breast cancer 22-24 , prostate core needle biopsies [24][25][26] , and mesothelioma 27 , we identify many non-trivial challenges should be addressed before considering application in the clinical setting. First, a deep learning model should be able to sustain a thorough test with a substantial number (i.e., thousands) of slides over a continuous time period and with WSIs procured by various brands of digital scanners.…”