Dataset Information

Deep learned tissue "fingerprints" classify breast cancers by ER/PR/Her2 status from H&E images.

ABSTRACT: Because histologic types are subjective and difficult to reproduce between pathologists, tissue morphology often takes a back seat to molecular testing for the selection of breast cancer treatments. This work explores whether a deep-learning algorithm can learn objective histologic H&E features that predict the clinical subtypes of breast cancer, as assessed by immunostaining for estrogen, progesterone, and Her2 receptors (ER/PR/Her2). Translating deep learning to this and related problems in histopathology presents a challenge due to the lack of large, well-annotated data sets, which are typically required for the algorithms to learn statistically significant discriminatory patterns. To overcome this limitation, we introduce the concept of "tissue fingerprints," which leverages large, unannotated datasets in a label-free manner to learn H&E features that can distinguish one patient from another. The hypothesis is that training the algorithm to learn the morphological differences between patients will implicitly teach it about the biologic variation between them. Following this training internship, we used the features the network learned, which we call "fingerprints," to predict ER, PR, and Her2 status in two datasets. Despite the discovery dataset being relatively small by the standards of the machine learning community (n?=?939), fingerprints enabled the determination of ER, PR, and Her2 status from whole slide H&E images with 0.89 AUC (ER), 0.81 AUC (PR), and 0.79 AUC (Her2) on a large, independent test set (n?=?2531). Tissue fingerprints are concise but meaningful histopathologic image representations that capture biological information and may enable machine learning algorithms that go beyond the traditional ER/PR/Her2 clinical groupings by directly predicting theragnosis.

SUBMITTER: Rawat RR

PROVIDER: S-EPMC7190637 | biostudies-literature | 2020 Apr

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Deep learned tissue "fingerprints" classify breast cancers by ER/PR/Her2 status from H&E images.

Rawat Rishi R RR Ortega Itzel I Roy Preeyam P Sha Fei F Shibata Darryl D Ruderman Daniel D Agus David B DB

Scientific reports 20200429 1

Because histologic types are subjective and difficult to reproduce between pathologists, tissue morphology often takes a back seat to molecular testing for the selection of breast cancer treatments. This work explores whether a deep-learning algorithm can learn objective histologic H&E features that predict the clinical subtypes of breast cancer, as assessed by immunostaining for estrogen, progesterone, and Her2 receptors (ER/PR/Her2). Translating deep learning to this and related problems in hi ...[more]

PMID: 32350370

Similar Datasets

Project description:BackgroundGenetic alterations in cellular signaling networks are a hallmark of cancer, however, effective methods to discover them are lacking. A novel form of abnormality called acquired uniparental disomy (aUPD) was recently found to pinpoint the region of mutated genes in various cancers, thereby identifying the region for next-generation sequencing.Methods/principal findingsWe retrieved large genomic data sets from the Gene Expression Omnibus database to perform genome-wide analysis of aUPD in breast tumor samples and cell lines using approaches that can reliably detect aUPD. aUPD was identified in 52.29% of the tumor samples. The most frequent aUPD regions were located at chromosomes 2q, 3p, 5q, 9p, 9q, 10q, 11q, 13q, 14q and 17q. We evaluated the data for any correlation between the most frequent aUPD regions and HER2/neu, ER, and PR status, and found a statistically significant correlation between the recurrent regions of aUPD and triple negative (TN) breast cancers. aUPD at chromosome 17q (VEZF1, WNT3), 3p (SUMF1, GRM7), 9p (MTAP, NFIB) and 11q (CASP1, CASP4, CASP5) are predictors for TN. The frequency of aUPD was found to be significantly higher in TN breast cancer cases compared to HER2/neu-positive and/or ER or PR-positive cases. Furthermore, using previously published mutation data, we found TP53 homozygously mutated in cell lines having aUPD in that locus.Conclusions/significanceWe conclude that aUPD is a common and non-random molecular feature of breast cancer that is most prominent in triple negative cases. As aUPD regions are different among the main pathological subtypes, specific aUPD regions may aid the sub-classification of breast cancer. In addition, we provide statistical support using TP53 as an example that identifying aUPD regions can be an effective approach in finding aberrant genes. We thus conclude that a genome-wide scale analysis of aUPD regions for homozygous sequence alterations can provide valuable insights into breast tumorigenesis.

Project description:Background: Extensive amounts of archived formalin fixed paraffin embedded (FFPE) human tumor tissues are the ultimate resource to investigate signaling network underlying tumorigenesis in human. Yet, their usage is severely limited for lacking of suitable protein techniques. In this study, a quantitative, objective, absolute, and high throughput immunoblot method, quantitative dot blot (QDB), was explored to address this issue by investigating the putative relationship between estrogen receptor (ER)/progesterone receptor (PR) and human epidermal growth factor receptor 2 (Her2) pathways in breast cancer tumorigenesis. Methods: In this descriptive observational retrospective study, ER, PR, Her2, and Ki67 protein levels were measured absolutely and quantitatively in 852 FFPE breast cancer tissues using the QDB method. ER, PR, and Her2 levels were charted on the X, Y, and Z-axes to observe samples distribution in a 3D scatterplot. Results: A "seesaw" relationship between ER/PR and Her2 pathways was observed in ER-PR-Her2 space, characterized by the expression levels of these 3 proteins. Specimens with strong expressions of ER/PR proteins were found spreading along the ER/PR floor while those with strong Her2 expression were found wrapping around the Her2 axis. Those lacking strong expressions of all 3 proteins were found accumulating at the intersection of the ER, PR, and Her2 axes. Few specimens floated in the ER-PR-Her2 space to suggest the lack of co-expression of all 3 proteins simultaneously. Ki67 levels were found to be significantly reduced in specimens spreading in the ER-PR space. Conclusions: The unique distribution of specimens in ER-PR-Her2 space prior to any clinical intervention provided visual support of bidirectional talk between ER/PR and Her2 pathways in breast cancer specimens. Clinical interventions to suppress these 2 pathways alternatively warrant further exploration for breast cancer patients accordingly. Our study also demonstrated that the QDB method is an effective tool to analyze archived FFPE cancer specimens in biomedical research.

Dataset Information

Deep learned tissue "fingerprints" classify breast cancers by ER/PR/Her2 status from H&E images.

Publications

Deep learned tissue "fingerprints" classify breast cancers by ER/PR/Her2 status from H&E images.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets