Article Text

Download PDFPDF
Continuous measurement of breast tumour hormone receptor expression: a comparison of two computational pathology platforms
  1. Thomas P Ahern1,
  2. Andrew H Beck2,
  3. Bernard A Rosner3,4,
  4. Ben Glass2,
  5. Gretchen Frieling2,
  6. Laura C Collins2,
  7. Rulla M Tamimi3,5
  1. 1Departments of Surgery and Biochemistry, University of Vermont College of Medicine, Burlington, Vermont, USA
  2. 2Department of Pathology, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, Massachusetts, USA
  3. 3Channing Division of Network Medicine, Brigham and Women's Hospital & Harvard Medical School, Boston, Massachusetts, USA
  4. 4Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
  5. 5Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA
  1. Correspondence to Dr Thomas P Ahern, 89 Beaumont Avenue, Given D317A, Burlington VT 05405, USA; 02tahern{at}


Aims Computational pathology platforms incorporate digital microscopy with sophisticated image analysis to permit rapid, continuous measurement of protein expression. We compared two computational pathology platforms on their measurement of breast tumour oestrogen receptor (ER) and progesterone receptor (PR) expression.

Methods Breast tumour microarrays from the Nurses' Health Study were stained for ER (n=592) and PR (n=187). One expert pathologist scored cases as positive if ≥1% of tumour nuclei exhibited stain. ER and PR were then measured with the Definiens Tissue Studio (automated) and Aperio Digital Pathology (user-supervised) platforms. Platform-specific measurements were compared using boxplots, scatter plots and correlation statistics. Classification of ER and PR positivity by platform-specific measurements was evaluated with areas under receiver operating characteristic curves (AUC) from univariable logistic regression models, using expert pathologist classification as the standard.

Results Both platforms showed considerable overlap in continuous measurements of ER and PR between positive and negative groups classified by expert pathologist. Platform-specific measurements were strongly and positively correlated with one another (r≥0.77). The user-supervised Aperio workflow performed slightly better than the automated Definiens workflow at classifying ER positivity (AUCAperio=0.97; AUCDefiniens=0.90; difference=0.07, 95% CI 0.05 to 0.09) and PR positivity (AUCAperio=0.94; AUCDefiniens=0.87; difference=0.07, 95% CI 0.03 to 0.12).

Conclusions Paired hormone receptor expression measurements from two different computational pathology platforms agreed well with one another. The user-supervised workflow yielded better classification accuracy than the automated workflow. Appropriately validated computational pathology algorithms enrich molecular epidemiology studies with continuous protein expression data and may accelerate tumour biomarker discovery.


Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • Handling editor Runjan Chetty

  • Contributors All authors contributed to the design, analysis and interpretation of the study and to manuscript preparation.

  • Funding This study was supported by the following grants from the National Cancer Institute at the US NIH: UM1 CA186107 and P01 CA87969. TPA was supported by a Career Catalyst Award from Susan G Komen for the Cure (CCR13264024) and by the Mary Kay Foundation (003-14).

  • Competing interests AHB has an equity interest in PathAI.

  • Ethics approval Brigham and Women’s Hospital Institutional Review Board.

  • Provenance and peer review Not commissioned; externally peer reviewed.