Dataset Information

Deep Learning to Improve Breast Cancer Detection on Screening Mammography.

ABSTRACT: The rapid development of deep learning, a family of machine learning techniques, has spurred much interest in its application to medical imaging problems. Here, we develop a deep learning algorithm that can accurately detect breast cancer on screening mammograms using an "end-to-end" training approach that efficiently leverages training datasets with either complete clinical annotation or only the cancer status (label) of the whole image. In this approach, lesion annotations are required only in the initial training stage, and subsequent stages require only image-level labels, eliminating the reliance on rarely available lesion annotations. Our all convolutional network method for classifying screening mammograms attained excellent performance in comparison with previous methods. On an independent test set of digitized film mammograms from the Digital Database for Screening Mammography (CBIS-DDSM), the best single model achieved a per-image AUC of 0.88, and four-model averaging improved the AUC to 0.91 (sensitivity: 86.1%, specificity: 80.1%). On an independent test set of full-field digital mammography (FFDM) images from the INbreast database, the best single model achieved a per-image AUC of 0.95, and four-model averaging improved the AUC to 0.98 (sensitivity: 86.7%, specificity: 96.1%). We also demonstrate that a whole image classifier trained using our end-to-end approach on the CBIS-DDSM digitized film mammograms can be transferred to INbreast FFDM images using only a subset of the INbreast data for fine-tuning and without further reliance on the availability of lesion annotations. These findings show that automatic deep learning methods can be readily trained to attain high accuracy on heterogeneous mammography platforms, and hold tremendous promise for improving clinical tools to reduce false positive and false negative screening mammography results. Code and model available at: https://github.com/lishen/end2end-all-conv .

SUBMITTER: Shen L

PROVIDER: S-EPMC6715802 | biostudies-other | 2019 Aug

REPOSITORIES: biostudies-other

ACCESS DATA

Publications

Deep Learning to Improve Breast Cancer Detection on Screening Mammography.

Shen Li L Margolies Laurie R LR Rothstein Joseph H JH Fluder Eugene E McBride Russell R Sieh Weiva W

Scientific reports 20190829 1

The rapid development of deep learning, a family of machine learning techniques, has spurred much interest in its application to medical imaging problems. Here, we develop a deep learning algorithm that can accurately detect breast cancer on screening mammograms using an "end-to-end" training approach that efficiently leverages training datasets with either complete clinical annotation or only the cancer status (label) of the whole image. In this approach, lesion annotations are required only in ...[more]

PMID: 31467326

Similar Datasets

Project description:BackgroundBreast cancer screening through mammography is crucial for early detection, yet the demand for mammography services surpasses the capacity of radiologists. Artificial intelligence (AI) can assist in evaluating microcalcifications on mammography. We developed and tested an AI model for localizing and characterizing microcalcifications.MethodsThree expert radiologists annotated a dataset of mammograms using histology-based ground truth. The dataset was partitioned for training, validation, and testing. Three neural networks (AlexNet, ResNet18, and ResNet34) were trained and evaluated using specific metrics including receiver operating characteristics area under the curve (AUC), sensitivity, and specificity. The reported metrics were computed on the test set (10% of the whole dataset).ResultsThe dataset included 1,000 patients aged 21-73 years and 1,986 mammograms (180 density A, 220 density B, 380 density C, and 220 density D), with 389 malignant and 611 benign groups of microcalcifications. AlexNet achieved the best performance with 0.98 sensitivity, 0.89 specificity of, and 0.98 AUC for microcalcifications detection and 0.85 sensitivity, 0.89 specificity, and 0.94 AUC of for microcalcifications classification. For microcalcifications detection, ResNet18 and ResNet34 achieved 0.96 and 0.97 sensitivity, 0.91 and 0.90 specificity and 0.98 and 0.98 AUC, retrospectively. For microcalcifications classification, ResNet18 and ResNet34 exhibited 0.75 and 0.84 sensitivity, 0.85 and 0.84 specificity, and 0.88 and 0.92 AUC, respectively.ConclusionsThe developed AI models accurately detect and characterize microcalcifications on mammography.Relevance statementAI-based systems have the potential to assist radiologists in interpreting microcalcifications on mammograms. The study highlights the importance of developing reliable deep learning models possibly applied to breast cancer screening.Key points• A novel AI tool was developed and tested to aid radiologists in the interpretation of mammography by accurately detecting and characterizing microcalcifications. • Three neural networks (AlexNet, ResNet18, and ResNet34) were trained, validated, and tested using an annotated dataset of 1,000 patients and 1,986 mammograms. • The AI tool demonstrated high accuracy in detecting/localizing and characterizing microcalcifications on mammography, highlighting the potential of AI-based systems to assist radiologists in the interpretation of mammograms.

Project description:ContextAnnual ultrasound screening may detect small, node-negative breast cancers that are not seen on mammography. Magnetic resonance imaging (MRI) may reveal additional breast cancers missed by both mammography and ultrasound screening.ObjectiveTo determine supplemental cancer detection yield of ultrasound and MRI in women at elevated risk for breast cancer.Design, setting, and participantsFrom April 2004-February 2006, 2809 women at 21 sites with elevated cancer risk and dense breasts consented to 3 annual independent screens with mammography and ultrasound in randomized order. After 3 rounds of both screenings, 612 of 703 women who chose to undergo an MRI had complete data. The reference standard was defined as a combination of pathology (biopsy results that showed in situ or infiltrating ductal carcinoma or infiltrating lobular carcinoma in the breast or axillary lymph nodes) and 12-month follow-up.Main outcome measuresCancer detection rate (yield), sensitivity, specificity, positive predictive value (PPV3) of biopsies performed and interval cancer rate.ResultsA total of 2662 women underwent 7473 mammogram and ultrasound screenings, 110 of whom had 111 breast cancer events: 33 detected by mammography only, 32 by ultrasound only, 26 by both, and 9 by MRI after mammography plus ultrasound; 11 were not detected by any imaging screen. Among 4814 incidence screens in the second and third years combined, 75 women were diagnosed with cancer. Supplemental incidence-screening ultrasound identified 3.7 cancers per 1000 screens (95% CI, 2.1-5.8; P < .001). Sensitivity for mammography plus ultrasound was 0.76 (95% CI, 0.65-0.85); specificity, 0.84 (95% CI, 0.83-0.85); and PPV3, 0.16 (95% CI, 0.12-0.21). For mammography alone, sensitivity was 0.52 (95% CI, 0.40-0.64); specificity, 0.91 (95% CI, 0.90-0.92); and PPV3, 0.38 (95% CI, 0.28-0.49; P < .001 all comparisons). Of the MRI participants, 16 women (2.6%) had breast cancer diagnosed. The supplemental yield of MRI was 14.7 per 1000 (95% CI, 3.5-25.9; P = .004). Sensitivity for MRI and mammography plus ultrasound was 1.00 (95% CI, 0.79-1.00); specificity, 0.65 (95% CI, 0.61-0.69); and PPV3, 0.19 (95% CI, 0.11-0.29). For mammography and ultrasound, sensitivity was 0.44 (95% CI, 0.20-0.70, P = .004); specificity 0.84 (95% CI, 0.81-0.87; P < .001); and PPV3, 0.18 (95% CI, 0.08 to 0.34; P = .98). The number of screens needed to detect 1 cancer was 127 (95% CI, 99-167) for mammography; 234 (95% CI, 173-345) for supplemental ultrasound; and 68 (95% CI, 39-286) for MRI after negative mammography and ultrasound results.ConclusionThe addition of screening ultrasound or MRI to mammography in women at increased risk of breast cancer resulted in not only a higher cancer detection yield but also an increase in false-positive findings.Trial registrationclinicaltrials.gov Identifier: NCT00072501.

Project description:ObjectiveTo evaluate the effectiveness of contemporary mammography screening using individual information about screening history and breast cancer mortality from public screening programmes.DesignProspective cohort study of Norwegian women who were followed between 1986 and 2009. Within that period (1995-2005), a national mammography screening programme was gradually implemented, with biennial invitations sent to women aged 50-69 years.ParticipantsAll Norwegian women aged 50-79 between 1986 and 2009.Main outcome measuresMultiple Poisson regression analysis was used to estimate breast cancer mortality rate ratios comparing women who were invited to screening (intention to screen) with women who were not invited, with a clear distinction between cases of breast cancer diagnosed before (without potential for screening effect) and after (with potential for screening effect) the first invitation for screening. We took competing causes of death into account by censoring women from further follow-up who died from other causes. Based on the observed mortality reduction combined with the all cause and breast cancer specific mortality in Norway in 2009, we used the CISNET (Cancer Intervention and Surveillance Modeling Network) Stanford simulation model to estimate how many women would need to be invited to biennial mammography screening in the age group 50-69 years to prevent one breast cancer death during their lifetime.ResultsDuring 15 193 034 person years of observation (1986-2009), deaths from breast cancer occurred in 1175 women with a diagnosis after being invited to screening and 8996 women who had not been invited before diagnosis. After adjustment for age, birth cohort, county of residence, and national trends in deaths from breast cancer, the mortality rate ratio associated with being invited to mammography screening was 0.72 (95% confidence interval 0.64 to 0.79). To prevent one death from breast cancer, 368 (95% confidence interval 266 to 508) women would need to be invited to screening.ConclusionInvitation to modern mammography screening may reduce deaths from breast cancer by about 28%.

Project description:To improve the performance of a computer-aided detection (CAD) system for mass detection by using four-view information in screening mammography.The authors developed a four-view CAD system that emulates radiologists' reading by using the craniocaudal and mediolateral oblique views of the ipsilateral breast to reduce false positives (FPs) and the corresponding views of the contralateral breast to detect asymmetry. The CAD system consists of four major components: (1) Initial detection of breast masses on individual views, (2) information fusion of the ipsilateral views of the breast (referred to as two-view analysis), (3) information fusion of the corresponding views of the contralateral breast (referred to as bilateral analysis), and (4) fusion of the four-view information with a decision tree. The authors collected two data sets for training and testing of the CAD system: A mass set containing 389 patients with 389 biopsy-proven masses and a normal set containing 200 normal subjects. All cases had four-view mammograms. The true locations of the masses on the mammograms were identified by an experienced MQSA radiologist. The authors randomly divided the mass set into two independent sets for cross validation training and testing. The overall test performance was assessed by averaging the free response receiver operating characteristic (FROC) curves of the two test subsets. The FP rates during the FROC analysis were estimated by using the normal set only. The jackknife free-response ROC (JAFROC) method was used to estimate the statistical significance of the difference between the test FROC curves obtained with the single-view and the four-view CAD systems.Using the single-view CAD system, the breast-based test sensitivities were 58% and 77% at the FP rates of 0.5 and 1.0 per image, respectively. With the four-view CAD system, the breast-based test sensitivities were improved to 76% and 87% at the corresponding FP rates, respectively. The improvement was found to be statistically significant (p < 0.0001) by JAFROC analysis.The four-view information fusion approach that emulates radiologists' reading strategy significantly improves the performance of breast mass detection of the CAD system in comparison with the single-view approach.

Dataset Information

Deep Learning to Improve Breast Cancer Detection on Screening Mammography.

Publications

Deep Learning to Improve Breast Cancer Detection on Screening Mammography.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets