Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study
Abstract 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 Full-field digital screening mammograms acquired in our c...
Main Authors: | , , , , , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
BMC
2019-06-01
|
Series: | Cancer Imaging |
Subjects: | |
Online Access: | http://link.springer.com/article/10.1186/s40644-019-0227-3 |
id |
doaj-9bdeae64271043d19c14f4fb7b4b081b |
---|---|
record_format |
Article |
spelling |
doaj-9bdeae64271043d19c14f4fb7b4b081b2021-04-02T12:50:19ZengBMCCancer Imaging1470-73302019-06-011911910.1186/s40644-019-0227-3Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case studyBenjamin Hinton0Lin Ma1Amir Pasha Mahmoudzadeh2Serghei Malkov3Bo Fan4Heather Greenwood5Bonnie Joe6Vivian Lee7Karla Kerlikowske8John Shepherd9Department of Bioengineering, University of California-San Francisco Berkeley Joint ProgramKaiser Permanente Division of ResearchAccentureApplied MaterialsDepartment of Bioengineering, University of California-San Francisco Berkeley Joint ProgramDepartment of Radiology and Biomedical Imaging, UC-San FranciscoDepartment of Radiology and Biomedical Imaging, UC-San FranciscoResearch Advocate, UCSF Breast Science Advocacy CoreDepartments of Medicine and Epidemiology and Biostatistics, UCSFCancer Epidemiology, University of Hawaii Cancer CenterAbstract 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 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 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 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.http://link.springer.com/article/10.1186/s40644-019-0227-3Breast CancerMaskingMammographyInterval CancerDeep learningTransfer learning |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Benjamin Hinton Lin Ma Amir Pasha Mahmoudzadeh Serghei Malkov Bo Fan Heather Greenwood Bonnie Joe Vivian Lee Karla Kerlikowske John Shepherd |
spellingShingle |
Benjamin Hinton Lin Ma Amir Pasha Mahmoudzadeh Serghei Malkov Bo Fan Heather Greenwood Bonnie Joe Vivian Lee Karla Kerlikowske John Shepherd Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study Cancer Imaging Breast Cancer Masking Mammography Interval Cancer Deep learning Transfer learning |
author_facet |
Benjamin Hinton Lin Ma Amir Pasha Mahmoudzadeh Serghei Malkov Bo Fan Heather Greenwood Bonnie Joe Vivian Lee Karla Kerlikowske John Shepherd |
author_sort |
Benjamin Hinton |
title |
Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study |
title_short |
Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study |
title_full |
Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study |
title_fullStr |
Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study |
title_full_unstemmed |
Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study |
title_sort |
deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study |
publisher |
BMC |
series |
Cancer Imaging |
issn |
1470-7330 |
publishDate |
2019-06-01 |
description |
Abstract 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 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 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 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. |
topic |
Breast Cancer Masking Mammography Interval Cancer Deep learning Transfer learning |
url |
http://link.springer.com/article/10.1186/s40644-019-0227-3 |
work_keys_str_mv |
AT benjaminhinton deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT linma deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT amirpashamahmoudzadeh deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT sergheimalkov deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT bofan deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT heathergreenwood deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT bonniejoe deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT vivianlee deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT karlakerlikowske deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy AT johnshepherd deeplearningnetworksfinduniquemammographicdifferencesinpreviousnegativemammogramsbetweenintervalandscreendetectedcancersacasecasestudy |
_version_ |
1721567442408308736 |