Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study.
Breast Cancer
Breast density
Deep learning
Interval Cancer
Mammography
Masking
Neural network
Transfer learning
Journal
Cancer imaging : the official publication of the International Cancer Imaging Society
ISSN: 1470-7330
Titre abrégé: Cancer Imaging
Pays: England
ID NLM: 101172931
Informations de publication
Date de publication:
22 Jun 2019
22 Jun 2019
Historique:
received:
04
03
2019
accepted:
13
06
2019
entrez:
24
6
2019
pubmed:
24
6
2019
medline:
4
9
2019
Statut:
epublish
Résumé
To determine if mammographic features from deep learning networks can be applied in breast cancer to identify groups at interval invasive cancer risk due to masking beyond using traditional breast density measures. Full-field digital screening mammograms acquired in our clinics between 2006 and 2015 were reviewed. Transfer learning of a deep learning network with weights initialized from ImageNet was performed to classify mammograms that were followed by an invasive interval or screen-detected cancer within 12 months of the mammogram. Hyperparameter optimization was performed and the network was visualized through saliency maps. Prediction loss and accuracy were calculated using this deep learning network. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were generated with the outcome of interval cancer using the deep learning network and compared to predictions from conditional logistic regression with errors quantified through contingency tables. Pre-cancer mammograms of 182 interval and 173 screen-detected cancers were split into training/test cases at an 80/20 ratio. Using Breast Imaging-Reporting and Data System (BI-RADS) density alone, the ability to correctly classify interval cancers was moderate (AUC = 0.65). The optimized deep learning model achieved an AUC of 0.82. Contingency table analysis showed the network was correctly classifying 75.2% of the mammograms and that incorrect classifications were slightly more common for the interval cancer mammograms. Saliency maps of each cancer case found that local information could highly drive classification of cases more than global image information. Pre-cancerous mammograms contain imaging information beyond breast density that can be identified with deep learning networks to predict the probability of breast cancer detection.
Sections du résumé
BACKGROUND
BACKGROUND
To determine if mammographic features from deep learning networks can be applied in breast cancer to identify groups at interval invasive cancer risk due to masking beyond using traditional breast density measures.
METHODS
METHODS
Full-field digital screening mammograms acquired in our clinics between 2006 and 2015 were reviewed. Transfer learning of a deep learning network with weights initialized from ImageNet was performed to classify mammograms that were followed by an invasive interval or screen-detected cancer within 12 months of the mammogram. Hyperparameter optimization was performed and the network was visualized through saliency maps. Prediction loss and accuracy were calculated using this deep learning network. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were generated with the outcome of interval cancer using the deep learning network and compared to predictions from conditional logistic regression with errors quantified through contingency tables.
RESULTS
RESULTS
Pre-cancer mammograms of 182 interval and 173 screen-detected cancers were split into training/test cases at an 80/20 ratio. Using Breast Imaging-Reporting and Data System (BI-RADS) density alone, the ability to correctly classify interval cancers was moderate (AUC = 0.65). The optimized deep learning model achieved an AUC of 0.82. Contingency table analysis showed the network was correctly classifying 75.2% of the mammograms and that incorrect classifications were slightly more common for the interval cancer mammograms. Saliency maps of each cancer case found that local information could highly drive classification of cases more than global image information.
CONCLUSIONS
CONCLUSIONS
Pre-cancerous mammograms contain imaging information beyond breast density that can be identified with deep learning networks to predict the probability of breast cancer detection.
Identifiants
pubmed: 31228956
doi: 10.1186/s40644-019-0227-3
pii: 10.1186/s40644-019-0227-3
pmc: PMC6589178
doi:
Types de publication
Journal Article
Langues
eng
Sous-ensembles de citation
IM
Pagination
41Subventions
Organisme : NCI NIH HHS
ID : P01 CA154292
Pays : United States
Organisme : National Science Foundation
ID : 1144247
Organisme : California Breast Cancer Research Program
ID : 21IB-0130
Organisme : NIH HHS
ID : P01CA154292
Pays : United States
Organisme : NIH HHS
ID : R01CA166945
Pays : United States
Références
Acad Radiol. 2010 Mar;17(3):382-6
pubmed: 20004597
Med Phys. 2009 Dec;36(12):5525-36
pubmed: 20095265
Ann Intern Med. 2011 Oct 18;155(8):493-502
pubmed: 22007043
Curr Oncol. 2012 Oct;19(5):249-53
pubmed: 23144572
Med Phys. 2013 Jan;40(1):014301
pubmed: 23298126
Med Phys. 2013 Jan;40(1):014302
pubmed: 23298127
CA Cancer J Clin. 2014 Jan-Feb;64(1):52-62
pubmed: 24114568
AJR Am J Roentgenol. 2015 Feb;204(2):W141-9
pubmed: 25615774
J Clin Oncol. 2015 Mar 20;33(9):1030-7
pubmed: 25646195
Comput Struct Biotechnol J. 2014 Nov 15;13:8-17
pubmed: 25750696
Breast Cancer. 2016 May;23(3):525-32
pubmed: 25763535
Ann Intern Med. 2015 May 19;162(10):673-81
pubmed: 25984843
Med Phys. 2016 Mar;43(3):1249-58
pubmed: 26936709
Invest Ophthalmol Vis Sci. 2016 Oct 1;57(13):5200-5206
pubmed: 27701631
Breast Cancer Res. 2016 Oct 5;18(1):100
pubmed: 27716311
JAMA. 2016 Dec 13;316(22):2402-2410
pubmed: 27898976
Radiology. 2017 Apr;283(1):49-58
pubmed: 27918707
Breast Cancer Res. 2016 Dec 6;18(1):122
pubmed: 27923387
Med Image Anal. 2017 Dec;42:60-88
pubmed: 28778026
Sci Rep. 2018 Mar 15;8(1):4165
pubmed: 29545529
Br J Radiol. 2018 Sep;91(1089):20170545
pubmed: 29565644
Ann Intern Med. 2018 Jun 5;168(11):757-765
pubmed: 29710124
Clin Cancer Res. 2018 Dec 1;24(23):5902-5909
pubmed: 30309858
Radiology. 2019 Jan;290(1):52-58
pubmed: 30325282
Radiology. 2019 Feb;290(2):456-464
pubmed: 30398430
Eur J Radiol. 2019 Feb;111:76-80
pubmed: 30691669