Determining breast cancer biomarker status and associated morphological features using deep learning.
| Autor: | Gamble P; Google Health, Palo Alto, CA USA., Jaroensri R; Google Health, Palo Alto, CA USA., Wang H; Google Health, Palo Alto, CA USA., Tan F; Google Health, Palo Alto, CA USA., Moran M; Google Health, Palo Alto, CA USA., Brown T; Google Health via Vituity, Emeryville, CA USA., Flament-Auvigne I; Google Health via Vituity, Emeryville, CA USA., Rakha EA; Department of Pathology, School of Medicine, University of Nottingham, Nottingham, UK., Toss M; Department of Pathology, School of Medicine, University of Nottingham, Nottingham, UK., Dabbs DJ; John A. Burns University of Hawaii Cancer Center, Honolulu, HI USA.; Department of Pathology, Magee-Womens Hospital of UPMC, Pittsburgh, PA USA., Regitnig P; Diagnostic and Research Institute of Pathology, Medical University of Graz, Graz, Austria., Olson N; Defense Innovation Unit, Mountain View, CA USA., Wren JH; Henry M. Jackson Foundation, Bethesda, MD USA., Robinson C; Laboratory Department, Naval Medical Center San Diego, San Diego, CA USA., Corrado GS; Google Health, Palo Alto, CA USA., Peng LH; Google Health, Palo Alto, CA USA., Liu Y; Google Health, Palo Alto, CA USA., Mermel CH; Google Health, Palo Alto, CA USA., Steiner DF; Google Health, Palo Alto, CA USA., Chen PC; Google Health, Palo Alto, CA USA. |
|---|---|
| Jazyk: | angličtina |
| Zdroj: | Communications medicine [Commun Med (Lond)] 2021 Jul 14; Vol. 1, pp. 14. Date of Electronic Publication: 2021 Jul 14 (Print Publication: 2021). |
| DOI: | 10.1038/s43856-021-00013-3 |
| Abstrakt: | Background: Breast 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. 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-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. 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. 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. Conclusions: This 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. Competing Interests: Competing interestsThis study was funded by Google LLC and Verily Life Sciences. P.G., R.J., H.W., F.T., M.M., G.S.C., L.H.P., Y.L., C.H.M., D.F.S., and P.-H.C.C. are employees of Google LLC and own Alphabet stock. I.F.-A. and T.B. are consultants of Google LLC. M.T., D.J.D., E.A.R., P.R., N.O., J.H.W., and C.R. declare no competing interests. (© The Author(s) 2021.) |
| Databáze: | MEDLINE |
| Externí odkaz: |
|