Predicting survival from colorectal cancer histology slides using deep learning: A retrospective multicenter study.
Journal
PLoS medicine
ISSN: 1549-1676
Titre abrégé: PLoS Med
Pays: United States
ID NLM: 101231360
Informations de publication
Date de publication:
01 2019
01 2019
Historique:
received:
23
05
2018
accepted:
17
12
2018
entrez:
25
1
2019
pubmed:
25
1
2019
medline:
14
5
2019
Statut:
epublish
Résumé
For virtually every patient with colorectal cancer (CRC), hematoxylin-eosin (HE)-stained tissue slides are available. These images contain quantitative information, which is not routinely used to objectively extract prognostic biomarkers. In the present study, we investigated whether deep convolutional neural networks (CNNs) can extract prognosticators directly from these widely available images. We hand-delineated single-tissue regions in 86 CRC tissue slides, yielding more than 100,000 HE image patches, and used these to train a CNN by transfer learning, reaching a nine-class accuracy of >94% in an independent data set of 7,180 images from 25 CRC patients. With this tool, we performed automated tissue decomposition of representative multitissue HE images from 862 HE slides in 500 stage I-IV CRC patients in the The Cancer Genome Atlas (TCGA) cohort, a large international multicenter collection of CRC tissue. Based on the output neuron activations in the CNN, we calculated a "deep stroma score," which was an independent prognostic factor for overall survival (OS) in a multivariable Cox proportional hazard model (hazard ratio [HR] with 95% confidence interval [CI]: 1.99 [1.27-3.12], p = 0.0028), while in the same cohort, manual quantification of stromal areas and a gene expression signature of cancer-associated fibroblasts (CAFs) were only prognostic in specific tumor stages. We validated these findings in an independent cohort of 409 stage I-IV CRC patients from the "Darmkrebs: Chancen der Verhütung durch Screening" (DACHS) study who were recruited between 2003 and 2007 in multiple institutions in Germany. Again, the score was an independent prognostic factor for OS (HR 1.63 [1.14-2.33], p = 0.008), CRC-specific OS (HR 2.29 [1.5-3.48], p = 0.0004), and relapse-free survival (RFS; HR 1.92 [1.34-2.76], p = 0.0004). A prospective validation is required before this biomarker can be implemented in clinical workflows. In our retrospective study, we show that a CNN can assess the human tumor microenvironment and predict prognosis directly from histopathological images.
Sections du résumé
BACKGROUND
For virtually every patient with colorectal cancer (CRC), hematoxylin-eosin (HE)-stained tissue slides are available. These images contain quantitative information, which is not routinely used to objectively extract prognostic biomarkers. In the present study, we investigated whether deep convolutional neural networks (CNNs) can extract prognosticators directly from these widely available images.
METHODS AND FINDINGS
We hand-delineated single-tissue regions in 86 CRC tissue slides, yielding more than 100,000 HE image patches, and used these to train a CNN by transfer learning, reaching a nine-class accuracy of >94% in an independent data set of 7,180 images from 25 CRC patients. With this tool, we performed automated tissue decomposition of representative multitissue HE images from 862 HE slides in 500 stage I-IV CRC patients in the The Cancer Genome Atlas (TCGA) cohort, a large international multicenter collection of CRC tissue. Based on the output neuron activations in the CNN, we calculated a "deep stroma score," which was an independent prognostic factor for overall survival (OS) in a multivariable Cox proportional hazard model (hazard ratio [HR] with 95% confidence interval [CI]: 1.99 [1.27-3.12], p = 0.0028), while in the same cohort, manual quantification of stromal areas and a gene expression signature of cancer-associated fibroblasts (CAFs) were only prognostic in specific tumor stages. We validated these findings in an independent cohort of 409 stage I-IV CRC patients from the "Darmkrebs: Chancen der Verhütung durch Screening" (DACHS) study who were recruited between 2003 and 2007 in multiple institutions in Germany. Again, the score was an independent prognostic factor for OS (HR 1.63 [1.14-2.33], p = 0.008), CRC-specific OS (HR 2.29 [1.5-3.48], p = 0.0004), and relapse-free survival (RFS; HR 1.92 [1.34-2.76], p = 0.0004). A prospective validation is required before this biomarker can be implemented in clinical workflows.
CONCLUSIONS
In our retrospective study, we show that a CNN can assess the human tumor microenvironment and predict prognosis directly from histopathological images.
Identifiants
pubmed: 30677016
doi: 10.1371/journal.pmed.1002730
pii: PMEDICINE-D-18-01784
pmc: PMC6345440
doi:
Substances chimiques
Coloring Agents
0
Eosine Yellowish-(YS)
TDQ283MPCW
Hematoxylin
YKM8PY2Z55
Types de publication
Journal Article
Multicenter Study
Research Support, Non-U.S. Gov't
Langues
eng
Sous-ensembles de citation
IM
Pagination
e1002730Déclaration de conflit d'intérêts
The authors have declared that no competing interests exist.
Références
Oncotarget. 2015 Aug 7;6(22):19163-76
pubmed: 26061817
IEEE Rev Biomed Eng. 2009;2:147-71
pubmed: 20671804
Br J Cancer. 2017 Nov 7;117(10):1427-1441
pubmed: 28982110
Nature. 2015 May 28;521(7553):436-44
pubmed: 26017442
PLoS One. 2013 Jul 29;8(7):e70221
pubmed: 23922958
Nature. 2017 Feb 2;542(7639):115-118
pubmed: 28117445
Sci Rep. 2018 Feb 21;8(1):3395
pubmed: 29467373
Lancet Oncol. 2008 Aug;9(8):766-76
pubmed: 18672212
Magn Reson Imaging. 2012 Nov;30(9):1301-12
pubmed: 22898682
Sci Rep. 2016 Jun 16;6:27988
pubmed: 27306927
Neural Netw. 2012 Aug;32:333-8
pubmed: 22386783
Nat Med. 2015 Nov;21(11):1350-6
pubmed: 26457759
Cancer Res. 2017 Nov 15;77(22):6442-6452
pubmed: 28923860
Nat Biotechnol. 2012 May;30(5):413-21
pubmed: 22544022
Science. 2006 Sep 29;313(5795):1960-4
pubmed: 17008531
Ann Oncol. 2013 Jan;24(1):179-85
pubmed: 22865778
J Natl Cancer Inst. 2015 Mar 13;107(6):djv045
pubmed: 25770147
J Clin Oncol. 2016 Jun 20;34(18):2157-64
pubmed: 27138577
Med Image Anal. 2017 May;38:90-103
pubmed: 28314191
JAMA. 2017 Dec 12;318(22):2199-2210
pubmed: 29234806
Nature. 2016 Jan 28;529(7587):484-9
pubmed: 26819042
Lancet Oncol. 2018 Mar;19(3):356-369
pubmed: 29402700
Br J Cancer. 2015 Sep 29;113(7):1075-80
pubmed: 26348443
Sci Rep. 2017 Dec 4;7(1):16878
pubmed: 29203879
Ann Oncol. 2018 Mar 1;29(3):616-623
pubmed: 29293881
Br J Cancer. 2015 Jan 20;112(2):251-9
pubmed: 25562432
J Pathol. 2018 Apr;244(4):421-431
pubmed: 29282718
Nat Genet. 2015 Apr;47(4):312-9
pubmed: 25706627
Nat Commun. 2014 Jun 03;5:4006
pubmed: 24892406
Oncoimmunology. 2018 Mar 29;7(7):e1444412
pubmed: 29900054
J Clin Oncol. 2011 Oct 1;29(28):3761-7
pubmed: 21876077
Nature. 2012 Jul 18;487(7407):330-7
pubmed: 22810696
Oncotarget. 2016 Oct 18;7(42):68954-68965
pubmed: 27661111
Semin Cancer Biol. 2018 Oct;52(Pt 2):189-197
pubmed: 29501787
Sci Transl Med. 2012 Oct 24;4(157):157ra143
pubmed: 23100629
J Pathol Inform. 2016 Sep 01;7:38
pubmed: 27688929
Clin Cancer Res. 2016 Aug 15;22(16):4057-66
pubmed: 26994146
Nat Rev Clin Oncol. 2009 Aug;6(8):445-54
pubmed: 19546864
Cancer Res. 2011 Sep 1;71(17):5670-7
pubmed: 21846824
Clin Cancer Res. 2009 Jan 15;15(2):431-40
pubmed: 19147747
Eur J Surg Oncol. 2012 Apr;38(4):307-13
pubmed: 22264965