Application of deep learning on mammographies to discriminate between low and high-risk DCIS for patient participation in active surveillance trials.

BACKGROUND: Ductal Carcinoma In Situ (DCIS) can progress to invasive breast cancer, but most DCIS lesions never will. Therefore, four clinical trials (COMET, LORIS, LORETTA, AND LORD) test whether active surveillance for women with low-risk Ductal carcinoma In Situ is safe (E. S. Hwang et al., BMJ Open, 9: e026797, 2019, A. Francis et al., Eur J Cancer. 51: 2296-2303, 2015, Chizuko Kanbayashi et al. The international collaboration of active surveillance trials for low-risk DCIS (LORIS, LORD, COMET, LORETTA),  L. E. Elshof et al., Eur J Cancer, 51, 1497-510, 2015). Low-risk is defined as grade I or II DCIS. Because DCIS grade is a major eligibility criteria in these trials, it would be very helpful to assess DCIS grade on mammography, informed by grade assessed on DCIS histopathology in pre-surgery biopsies, since surgery will not be performed on a significant number of patients participating in these trials. OBJECTIVE: To assess the performance and clinical utility of a convolutional neural network (CNN) in discriminating high-risk (grade III) DCIS and/or Invasive Breast Cancer (IBC) from low-risk (grade I/II) DCIS based on mammographic features. We explored whether the CNN could be used as a decision support tool, from excluding high-risk patients for active surveillance. METHODS: In this single centre retrospective study, 464 patients diagnosed with DCIS based on pre-surgery biopsy between 2000 and 2014 were included. The collection of mammography images was partitioned on a patient-level into two subsets, one for training containing 80% of cases (371 cases, 681 images) and 20% (93 cases, 173 images) for testing. A deep learning model based on the U-Net CNN was trained and validated on 681 two-dimensional mammograms. Classification performance was assessed with the Area Under the Curve (AUC) receiver operating characteristic and predictive values on the test set for predicting high risk DCIS-and high-risk DCIS and/ or IBC from low-risk DCIS. RESULTS: When classifying DCIS as high-risk, the deep learning network achieved a Positive Predictive Value (PPV) of 0.40, Negative Predictive Value (NPV) of 0.91 and an AUC of 0.72 on the test dataset. For distinguishing high-risk and/or upstaged DCIS (occult invasive breast cancer) from low-risk DCIS a PPV of 0.80, a NPV of 0.84 and an AUC of 0.76 were achieved. CONCLUSION: For both scenarios (DCIS grade I/II vs. III, DCIS grade I/II vs. III and/or IBC) AUCs were high, 0.72 and 0.76, respectively, concluding that our convolutional neural network can discriminate low-grade from high-grade DCIS.

Classification performance bias between training and test sets in a limited mammography dataset.

OBJECTIVES: To assess the performance bias caused by sampling data into training and test sets in a mammography radiomics study. METHODS: Mammograms from 700 women were used to study upstaging of ductal carcinoma in situ. The dataset was repeatedly shuffled and split into training (n = 400) and test cases (n = 300) forty times. For each split, cross-validation was used for training, followed by an assessment of the test set. Logistic regression with regularization and support vector machine were used as the machine learning classifiers. For each split and classifier type, multiple models were created based on radiomics and/or clinical features. RESULTS: Area under the curve (AUC) performances varied considerably across the different data splits (e.g., radiomics regression model: train 0.58-0.70, test 0.59-0.73). Performances for regression models showed a tradeoff where better training led to worse testing and vice versa. Cross-validation over all cases reduced this variability, but required samples of 500+ cases to yield representative estimates of performance. CONCLUSIONS: In medical imaging, clinical datasets are often limited to relatively small size. Models built from different training sets may not be representative of the whole dataset. Depending on the selected data split and model, performance bias could lead to inappropriate conclusions that might influence the clinical significance of the findings. ADVANCES IN KNOWLEDGE: Performance bias can result from model testing when using limited datasets. Optimal strategies for test set selection should be developed to ensure study conclusions are appropriate.

Ipsilateral Lesion Detection Refinement for Tomosynthesis.

Computer-aided detection (CAD) frameworks for breast cancer screening have been researched for several decades. Early adoption of deep-learning models in CAD frameworks has shown greatly improved detection performance compared to traditional CAD on single-view images. Recently, studies have improved performance by merging information from multiple views within each screening exam. Clinically, the integration of lesion correspondence during screening is a complicated decision process that depends on the correct execution of several referencing steps. However, most multi-view CAD frameworks are deep-learning-based black-box techniques. Fully end-to-end designs make it very difficult to analyze model behaviors and fine-tune performance. More importantly, the black-box nature of the techniques discourages clinical adoption due to the lack of explicit reasoning for each multi-view referencing step. Therefore, there is a need for a multi-view detection framework that can not only detect cancers accurately but also provide step-by-step, multi-view reasoning. In this work, we present Ipsilateral-Matching-Refinement Networks (IMR-Net) for digital breast tomosynthesis (DBT) lesion detection across multiple views. Our proposed framework adaptively refines the single-view detection scores based on explicit ipsilateral lesion matching. IMR-Net is built on a robust, single-view detection CAD pipeline with a commercial development DBT dataset of 24675 DBT volumetric views from 8034 exams. Performance is measured using location-based, case-level receiver operating characteristic (ROC) and case-level free-response ROC (FROC) analysis.

Classification performance bias between training and test sets in a limited mammography dataset.

Objectives: To assess the performance bias caused by sampling data into training and test sets in a mammography radiomics study. Methods: Mammograms from 700 women were used to study upstaging of ductal carcinoma in situ. The dataset was repeatedly shuffled and split into training (n=400) and test cases (n=300) forty times. For each split, cross-validation was used for training, followed by an assessment of the test set. Logistic regression with regularization and support vector machine were used as the machine learning classifiers. For each split and classifier type, multiple models were created based on radiomics and/or clinical features. Results: Area under the curve (AUC) performances varied considerably across the different data splits (e.g., radiomics regression model: train 0.58-0.70, test 0.59-0.73). Performances for regression models showed a tradeoff where better training led to worse testing and vice versa. Cross-validation over all cases reduced this variability, but required samples of 500+ cases to yield representative estimates of performance. Conclusions: In medical imaging, clinical datasets are often limited to relatively small size. Models built from different training sets may not be representative of the whole dataset. Depending on the selected data split and model, performance bias could lead to inappropriate conclusions that might influence the clinical significance of the findings. Optimal strategies for test set selection should be developed to ensure study conclusions are appropriate.

Utility of a Rule-Based Algorithm in the Assessment of Standardized Reporting in PI-RADS.

RATIONALE AND OBJECTIVES: Adoption of the Prostate Imaging Reporting & Data System (PI-RADS) has been shown to increase detection of clinically significant prostate cancer on prostate mpMRI. We propose that a rule-based algorithm based on Regular Expression (RegEx) matching can be used to automatically categorize prostate mpMRI reports into categories as a means by which to assess for opportunities for quality improvement. MATERIALS AND METHODS: All prostate mpMRIs performed in the Duke University Health System from January 2, 2015, to January 29, 2021, were analyzed. Exclusion criteria were applied, for a total of 5343 male patients and 6264 prostate mpMRI reports. These reports were then analyzed by our RegEx algorithm to be categorized as PI-RADS 1 through PI-RADS 5, Recurrent Disease, or "No Information Available." A stratified, random sample of 502 mpMRI reports was reviewed by a blinded clinical team to assess performance of the RegEx algorithm. RESULTS: Compared to manual review, the RegEx algorithm achieved overall accuracy of 92.6%, average precision of 88.8%, average recall of 85.6%, and F1 score of 0.871. The clinical team also reviewed 344 cases that were classified as "No Information Available," and found that in 150 instances, no numerical PI-RADS score for any lesion was included in the impression section of the mpMRI report. CONCLUSION: Rule-based processing is an accurate method for the large-scale, automated extraction of PI-RADS scores from the text of radiology reports. These natural language processing approaches can be used for future initiatives in quality improvement in prostate mpMRI reporting with PI-RADS.

Classification of Multiple Diseases on Body CT Scans Using Weakly Supervised Deep Learning.

PURPOSE: To design multidisease classifiers for body CT scans for three different organ systems using automatically extracted labels from radiology text reports. MATERIALS AND METHODS: This retrospective study included a total of 12 092 patients (mean age, 57 years ± 18 [standard deviation]; 6172 women) for model development and testing. Rule-based algorithms were used to extract 19 225 disease labels from 13 667 body CT scans performed between 2012 and 2017. Using a three-dimensional DenseVNet, three organ systems were segmented: lungs and pleura, liver and gallbladder, and kidneys and ureters. For each organ system, a three-dimensional convolutional neural network classified each as no apparent disease or for the presence of four common diseases, for a total of 15 different labels across all three models. Testing was performed on a subset of 2158 CT volumes relative to 2875 manually derived reference labels from 2133 patients (mean age, 58 years ± 18; 1079 women). Performance was reported as area under the receiver operating characteristic curve (AUC), with 95% CIs calculated using the DeLong method. RESULTS: Manual validation of the extracted labels confirmed 91%-99% accuracy across the 15 different labels. AUCs for lungs and pleura labels were as follows: atelectasis, 0.77 (95% CI: 0.74, 0.81); nodule, 0.65 (95% CI: 0.61, 0.69); emphysema, 0.89 (95% CI: 0.86, 0.92); effusion, 0.97 (95% CI: 0.96, 0.98); and no apparent disease, 0.89 (95% CI: 0.87, 0.91). AUCs for liver and gallbladder were as follows: hepatobiliary calcification, 0.62 (95% CI: 0.56, 0.67); lesion, 0.73 (95% CI: 0.69, 0.77); dilation, 0.87 (95% CI: 0.84, 0.90); fatty, 0.89 (95% CI: 0.86, 0.92); and no apparent disease, 0.82 (95% CI: 0.78, 0.85). AUCs for kidneys and ureters were as follows: stone, 0.83 (95% CI: 0.79, 0.87); atrophy, 0.92 (95% CI: 0.89, 0.94); lesion, 0.68 (95% CI: 0.64, 0.72); cyst, 0.70 (95% CI: 0.66, 0.73); and no apparent disease, 0.79 (95% CI: 0.75, 0.83). CONCLUSION: Weakly supervised deep learning models were able to classify diverse diseases in multiple organ systems from CT scans.Keywords: CT, Diagnosis/Classification/Application Domain, Semisupervised Learning, Whole-Body Imaging© RSNA, 2022.

Prediction of Upstaging in Ductal Carcinoma in Situ Based on Mammographic Radiomic Features.

Background Improving diagnosis of ductal carcinoma in situ (DCIS) before surgery is important in choosing optimal patient management strategies. However, patients may harbor occult invasive disease not detected until definitive surgery. Purpose To assess the performance and clinical utility of mammographic radiomic features in the prediction of occult invasive cancer among women diagnosed with DCIS on the basis of core biopsy findings. Materials and Methods In this Health Insurance Portability and Accountability Act-compliant retrospective study, digital magnification mammographic images were collected from women who underwent breast core-needle biopsy for calcifications that was performed at a single institution between September 2008 and April 2017 and yielded a diagnosis of DCIS. The database query was directed at asymptomatic women with calcifications without a mass, architectural distortion, asymmetric density, or palpable disease. Logistic regression with regularization was used. Differences across training and internal test set by upstaging rate, age, lesion size, and estrogen and progesterone receptor status were assessed by using the Kruskal-Wallis or χ2 test. Results The study consisted of 700 women with DCIS (age range, 40-89 years; mean age, 59 years ± 10 [standard deviation]), including 114 with lesions (16.3%) upstaged to invasive cancer at subsequent surgery. The sample was split randomly into 400 women for the training set and 300 for the testing set (mean ages: training set, 59 years ± 10; test set, 59 years ± 10; P = .85). A total of 109 radiomic and four clinical features were extracted. The best model on the test set by using all radiomic and clinical features helped predict upstaging with an area under the receiver operating characteristic curve of 0.71 (95% CI: 0.62, 0.79). For a fixed high sensitivity (90%), the model yielded a specificity of 22%, a negative predictive value of 92%, and an odds ratio of 2.4 (95% CI: 1.8, 3.2). High specificity (90%) corresponded to a sensitivity of 37%, positive predictive value of 41%, and odds ratio of 5.0 (95% CI: 2.8, 9.0). Conclusion Machine learning models that use radiomic features applied to mammographic calcifications may help predict upstaging of ductal carcinoma in situ, which can refine clinical decision making and treatment planning. © RSNA, 2022.

Anomaly Detection of Calcifications in Mammography Based on 11,000 Negative Cases.

In mammography, calcifications are one of the most common signs of breast cancer. Detection of such lesions is an active area of research for computer-aided diagnosis and machine learning algorithms. Due to limited numbers of positive cases, many supervised detection models suffer from overfitting and fail to generalize. We present a one-class, semi-supervised framework using a deep convolutional autoencoder trained with over 50,000 images from 11,000 negative-only cases. Since the model learned from only normal breast parenchymal features, calcifications produced large signals when comparing the residuals between input and reconstruction output images. As a key advancement, a structural dissimilarity index was used to suppress non-structural noises. Our selected model achieved pixel-based AUROC of 0.959 and AUPRC of 0.676 during validation, where calcification masks were defined in a semi-automated process. Although not trained directly on any cancers, detection performance of calcification lesions on 1,883 testing images (645 malignant and 1238 negative) achieved 75% sensitivity at 2.5 false positives per image. Performance plateaued early when trained with only a fraction of the cases, and greater model complexity or a larger dataset did not improve performance. This study demonstrates the potential of this anomaly detection approach to detect mammographic calcifications in a semi-supervised manner with efficient use of a small number of labeled images, and may facilitate new clinical applications such as computer-aided triage and quality improvement.

Multimodal Patient-Specific Registration for Breast Imaging Using Biomechanical Modeling with Reference to AI Evaluation of Breast Tumor Change.

BACKGROUND: The strategy to combat the problem associated with large deformations in the breast due to the difference in the medical imaging of patient posture plays a vital role in multimodal medical image registration with artificial intelligence (AI) initiatives. How to build a breast biomechanical model simulating the large-scale deformation of soft tissue remains a challenge but is highly desirable. METHODS: This study proposed a hybrid individual-specific registration model of the breast combining finite element analysis, property optimization, and affine transformation to register breast images. During the registration process, the mechanical properties of the breast tissues were individually assigned using an optimization process, which allowed the model to become patient specific. Evaluation and results: The proposed method has been extensively tested on two datasets collected from two independent institutions, one from America and another from Hong Kong. CONCLUSIONS: Our method can accurately predict the deformation of breasts from the supine to prone position for both the Hong Kong and American samples, with a small target registration error of lesions.

A Data Set and Deep Learning Algorithm for the Detection of Masses and Architectural Distortions in Digital Breast Tomosynthesis Images.

Importance: Breast cancer screening is among the most common radiological tasks, with more than 39 million examinations performed each year. While it has been among the most studied medical imaging applications of artificial intelligence, the development and evaluation of algorithms are hindered by the lack of well-annotated, large-scale publicly available data sets. Objectives: To curate, annotate, and make publicly available a large-scale data set of digital breast tomosynthesis (DBT) images to facilitate the development and evaluation of artificial intelligence algorithms for breast cancer screening; to develop a baseline deep learning model for breast cancer detection; and to test this model using the data set to serve as a baseline for future research. Design, Setting, and Participants: In this diagnostic study, 16 802 DBT examinations with at least 1 reconstruction view available, performed between August 26, 2014, and January 29, 2018, were obtained from Duke Health System and analyzed. From the initial cohort, examinations were divided into 4 groups and split into training and test sets for the development and evaluation of a deep learning model. Images with foreign objects or spot compression views were excluded. Data analysis was conducted from January 2018 to October 2020. Exposures: Screening DBT. Main Outcomes and Measures: The detection algorithm was evaluated with breast-based free-response receiver operating characteristic curve and sensitivity at 2 false positives per volume. Results: The curated data set contained 22 032 reconstructed DBT volumes that belonged to 5610 studies from 5060 patients with a mean (SD) age of 55 (11) years and 5059 (100.0%) women. This included 4 groups of studies: (1) 5129 (91.4%) normal studies; (2) 280 (5.0%) actionable studies, for which where additional imaging was needed but no biopsy was performed; (3) 112 (2.0%) benign biopsied studies; and (4) 89 studies (1.6%) with cancer. Our data set included masses and architectural distortions that were annotated by 2 experienced radiologists. Our deep learning model reached breast-based sensitivity of 65% (39 of 60; 95% CI, 56%-74%) at 2 false positives per DBT volume on a test set of 460 examinations from 418 patients. Conclusions and Relevance: The large, diverse, and curated data set presented in this study could facilitate the development and evaluation of artificial intelligence algorithms for breast cancer screening by providing data for training as well as a common set of cases for model validation. The performance of the model developed in this study showed that the task remains challenging; its performance could serve as a baseline for future model development.

A new method to accurately identify single nucleotide variants using small FFPE breast samples.

Most tissue collections of neoplasms are composed of formalin-fixed and paraffin-embedded (FFPE) excised tumor samples used for routine diagnostics. DNA sequencing is becoming increasingly important in cancer research and clinical management; however it is difficult to accurately sequence DNA from FFPE samples. We developed and validated a new bioinformatic pipeline to use existing variant-calling strategies to robustly identify somatic single nucleotide variants (SNVs) from whole exome sequencing using small amounts of DNA extracted from archival FFPE samples of breast cancers. We optimized this strategy using 28 pairs of technical replicates. After optimization, the mean similarity between replicates increased 5-fold, reaching 88% (range 0-100%), with a mean of 21.4 SNVs (range 1-68) per sample, representing a markedly superior performance to existing tools. We found that the SNV-identification accuracy declined when there was less than 40 ng of DNA available and that insertion-deletion variant calls are less reliable than single base substitutions. As the first application of the new algorithm, we compared samples of ductal carcinoma in situ of the breast to their adjacent invasive ductal carcinoma samples. We observed an increased number of mutations (paired-samples sign test, P < 0.05), and a higher genetic divergence in the invasive samples (paired-samples sign test, P < 0.01). Our method provides a significant improvement in detecting SNVs in FFPE samples over previous approaches.

Assessment of task-based performance from five clinical DBT systems using an anthropomorphic breast phantom.

PURPOSE: Digital breast tomosynthesis (DBT) is a limited-angle tomographic breast imaging modality that can be used for breast cancer screening in conjunction with full-field digital mammography (FFDM) or synthetic mammography (SM). Currently, there are five commercial DBT systems that have been approved by the U.S. FDA for breast cancer screening, all varying greatly in design and imaging protocol. Because the systems are different in technical specifications, there is a need for a quantitative approach for assessing them. In this study, the DBT systems are assessed using a novel methodology with an inkjet-printed anthropomorphic phantom and four alternative forced choice (4AFC) study scheme. METHOD: A breast phantom was fabricated using inkjet printing and parchment paper. The phantom contained 5-mm spiculated masses fabricated with potassium iodide (KI)-doped ink and microcalcifications (MCs) made with calcium hydroxyapatite. Images of the phantom were acquired on all five systems with DBT, FFDM, and SM modalities where available using beam settings under automatic exposure control. A 4AFC study was conducted to assess reader performance with each signal under each modality. Statistical analysis was performed on the data to determine proportion correct (PC), standard deviations, and levels of significance. RESULTS: For masses, overall detection was highest with DBT. The difference in PC was statistically significant between DBT and SM for most systems. A relationship was observed between increasing PC and greater gantry span. For MCs, performance was highest with DBT and FFDM compared to SM. The difference between PC of DBT and PC of SM was statistically significant for all manufacturers. CONCLUSIONS: This methodology represents a novel approach for evaluating systems. This study is the first of its kind to use an inkjet-printed anthropomorphic phantom with realistic signals to assess performance of clinical DBT imaging systems.

Mixed-Methods Study to Predict Upstaging of DCIS to Invasive Disease on Mammography.

BACKGROUND. The incidence of ductal carcinoma in situ (DCIS) has steadily increased, as have concerns regarding overtreatment. Active surveillance is a novel treatment strategy that avoids surgical excision, but identifying patients with occult invasive disease who should be excluded from active surveillance is challenging. Radiologists are not typically expected to predict the upstaging of DCIS to invasive disease, though they might be trained to perform this task. OBJECTIVE. The purpose of this study was to determine whether a mixed-methods two-stage observer study can improve radiologists' ability to predict upstaging of DCIS to invasive disease on mammography. METHODS. All cases of DCIS calcifications that underwent stereotactic biopsy between 2010 and 2015 were identified. Two cohorts were randomly generated, each containing 150 cases (120 pure DCIS cases and 30 DCIS cases upstaged to invasive disease at surgery). Nine breast radiologists reviewed the mammograms in the first cohort in a blinded fashion and scored the probability of upstaging to invasive disease. The radiologists then reviewed the cases and results collectively in a focus group to develop consensus criteria that could improve their ability to predict upstaging. The radiologists reviewed the mammograms from the second cohort in a blinded fashion and again scored the probability of upstaging. Statistical analysis compared the performances between rounds 1 and 2. RESULTS. The mean AUC for reader performance in predicting upstaging in round 1 was 0.623 (range, 0.514-0.684). In the focus group, radiologists agreed that upstaging was better predicted when an associated mass, asymmetry, or architectural distortion was present; when densely packed calcifications extended over a larger area; and when the most suspicious features were focused on rather than the most common features. Additionally, radiologists agreed that BI-RADS descriptors do not adequately characterize risk of invasion, and that microinvasive disease and smaller areas of DCIS will have poor prediction estimates. Reader performance significantly improved in round 2 (mean AUC, 0.765; range, 0.617-0.852; p = .045). CONCLUSION. A mixed-methods two-stage observer study identified factors that helped radiologists significantly improve their ability to predict upstaging of DCIS to invasive disease. CLINICAL IMPACT. Breast radiologists can be trained to better predict upstaging of DCIS to invasive disease, which may facilitate discussions with patients and referring providers.

Predicting Upstaging of DCIS to Invasive Disease: Radiologists's Predictive Performance.

RATIONALE AND OBJECTIVES: The purpose of this study is to quantify breast radiologists' performance at predicting occult invasive disease when ductal carcinoma in situ (DCIS) presents as calcifications on mammography and to identify imaging and histopathological features that are associated with radiologists' performance. MATERIALS AND METHODS: Mammographically detected calcifications that were initially diagnosed as DCIS on core biopsy and underwent definitive surgical excision between 2010 and 2015 were identified. Thirty cases of suspicious calcifications upstaged to invasive ductal carcinoma and 120 cases of DCIS confirmed at the time of definitive surgery were randomly selected. Nuclear grade, estrogen and progesterone receptor status, patient age, calcification long axis length, and breast density were collected. Ten breast radiologists who were blinded to all clinical and pathology data independently reviewed all cases and estimated the likelihood that the DCIS would be upstaged to invasive disease at surgical excision. Subgroup analysis was performed based on nuclear grade, long axis length, breast density and after exclusion of microinvasive disease. RESULTS: Reader performance to predict upstaging ranged from an area under the receiver operating characteristic curve (AUC) of 0.541-0.684 with a mean AUC of 0.620 (95%CI: 0.489-0.751). Performances improved for lesions smaller than 2 cm (AUC: 0.676 vs 0.500; p = 0.002). The exclusion of microinvasive cases also improved performance (AUC: 0.651 vs 0.620; p = 0.005). There was no difference in performance based on breast density (p = 0.850) or nuclear grade (p = 0.270) CONCLUSION: Radiologists were able to predict invasive disease better than chance, particularly for smaller DCIS lesions (<2 cm) and after the exclusion of microinvasive disease.

Prediction of Upstaged Ductal Carcinoma In Situ Using Forced Labeling and Domain Adaptation.

OBJECTIVE: The goal of this study is to use adjunctive classes to improve a predictive model whose performance is limited by the common problems of small numbers of primary cases, high feature dimensionality, and poor class separability. Specifically, our clinical task is to use mammographic features to predict whether ductal carcinoma in situ (DCIS) identified at needle core biopsy will be later upstaged or shown to contain invasive breast cancer. METHODS: To improve the prediction of pure DCIS (negative) versus upstaged DCIS (positive) cases, this study considers the adjunctive roles of two related classes: atypical ductal hyperplasia (ADH), a non-cancer type of breast abnormity, and invasive ductal carcinoma (IDC), with 113 computer vision based mammographic features extracted from each case. To improve the baseline Model A's classification of pure vs. upstaged DCIS, we designed three different strategies (Models B, C, D) with different ways of embedding features or inputs. RESULTS: Based on ROC analysis, the baseline Model A performed with AUC of 0.614 (95% CI, 0.496-0.733). All three new models performed better than the baseline, with domain adaptation (Model D) performing the best with an AUC of 0.697 (95% CI, 0.595-0.797). CONCLUSION: We improved the prediction performance of DCIS upstaging by embedding two related pathology classes in different training phases. SIGNIFICANCE: The three new strategies of embedding related class data all outperformed the baseline model, thus demonstrating not only feature similarities among these different classes, but also the potential for improving classification by using other related classes.

Growth Dynamics of Mammographic Calcifications: Differentiating Ductal Carcinoma in Situ from Benign Breast Disease.

Background Most ductal carcinoma in situ (DCIS) lesions are first detected on screening mammograms as calcifications. However, false-positive biopsy rates for calcifications range from 30% to 87%. Improved methods to differentiate benign from malignant calcifications are thus needed. Purpose To quantify the growth rates of DCIS and benign breast disease that manifest as mammographic calcifications. Materials and Methods All calcifications (n = 2359) for which a stereotactic biopsy was performed from 2008 through 2015 at Duke University Medical Center were retrospectively identified. Mammograms from all cases of DCIS (n = 404) were reviewed for calcifications that were visible on mammograms taken at least 6 months before biopsy. Women with at least one prior mammogram with visible calcifications were age- and race-matched 1:2 to women with a benign breast biopsy and calcifications visible on prior mammograms. The long axis of the calcifications was measured on all mammograms. Multivariable adjusted linear mixed-effects models estimated the association of calcification growth rates with patholo findings. Hierarchical clustering accounted for matching benign and DCIS groups. Results A total of 74 DCIS calcifications and 148 benign calcifications were included for final analysis. The median patient age was 62 years (interquartile range, 51-71 years). No significant difference in breast density (P > .05) or number of available mammograms (P > .05) was detected between groups. Calcifications associated with DCIS were larger than those associated with benign breast disease at biopsy (median, 10 mm vs 6 mm, respectively; P < .001). After adjustment, the relative annual increase in the long-axis length of DCIS calcifications was greater than that of benign breast calcifications (96% [95% confidence interval: 72%, 224%] vs 68% [95% confidence interval: 56%, 80%] per year, respectively; P < .001). Conclusion Ductal carcinoma in situ calcifications are more extensive at diagnosis and grow faster in extent than those associated with benign breast disease. The rate of calcification change may help to discriminate benign from malignant calcifications. © RSNA, 2019 Online supplemental material is available for this article.

A four-alternative forced choice (4AFC) methodology for evaluating microcalcification detection in clinical full-field digital mammography (FFDM) and digital breast tomosynthesis (DBT) systems using an inkjet-printed anthropomorphic phantom.

PURPOSE: The advent of three-dimensional breast imaging systems such as digital breast tomosynthesis (DBT) has great promise for improving the detection and diagnosis of breast cancer. With these new technologies comes an essential need for testing methods to assess the resultant image quality. Although randomized clinical trials are the gold standard for assessing image quality, phantom-based studies can provide a simpler and less burdensome approach. In this work, a complete framework is presented for task-based evaluation of microcalcification (MCs) detection performance for DBT imaging systems. METHODS: The framework consists of three parts. The first part is a realistic anthropomorphic physical breast phantom created through inkjet printing, with parchment paper and iodine-doped ink. The second is a method for inserting realistic MCs fabricated from calcium hydroxyapatite. The reproducibility and stability of the phantom materials were investigated through multiple samples of parchment and ink over 6 months. The final part is an analysis using a four-alternative forced choice (4AFC) reader study. To demonstrate the framework, a task-based 4AFC study was conducted using a clinical system to compare performance from DBT, synthetic mammography (SM), and full-field digital mammography (FFDM). Nine human observers read images containing MC clusters imaged with all three modalities and tried to correctly locate the MCs. The proportion correct (PC) was measured as the number of correctly detected clusters out of all trials. RESULTS: Overall, readers scored the highest with FFDM, (PC = 0.95 ± 0.03) then DBT (0.85 ± 0.04), and finally SM (0.44 ± 0.06). For the parchment and ink samples, the linear attenuation properties were very stable over 6 months. In addition, little difference was found between the various parchment and ink samples, indicating good reproducibility. CONCLUSIONS: This framework presents a promising methodology for evaluating diagnostic task performance of clinical breast DBT systems.

Can Digital Breast Tomosynthesis Replace Full-Field Digital Mammography? A Multireader, Multicase Study of Wide-Angle Tomosynthesis.

OBJECTIVE. The purpose of this study was to test the hypothesis whether two-view wide-angle digital breast tomosynthesis (DBT) can replace full-field digital mammography (FFDM) for breast cancer detection. SUBJECTS AND METHODS. In a multireader multicase study, bilateral two-view FFDM and bilateral two-view wide-angle DBT images were independently viewed for breast cancer detection in two reading sessions separated by more than 1 month. From a pool of 764 patients undergoing screening and diagnostic mammography, 330 patient-cases were selected. The endpoints were the mean ROC AUC for the reader per breast (breast level), ROC AUC per patient (subject level), noncancer recall rates, sensitivity, and specificity. RESULTS. Twenty-nine of 31 readers performed better with DBT than FFDM regardless of breast density. There was a statistically significant improvement in readers' mean diagnostic accuracy with DBT. The subject-level AUC increased from 0.765 (standard error [SE], 0.027) for FFDM to 0.835 (SE, 0.027) for DBT (p = 0.002). Breast-level AUC increased from 0.818 (SE, 0.019) for FFDM to 0.861 (SE, 0.019) for DBT (p = 0.011). The noncancer recall rate per patient was reduced by 19% with DBT (p < 0.001). Masses and architectural distortions were detected more with DBT (p < 0.001); calcifications trended lower (p = 0.136). Accuracy for detection of invasive cancers was significantly greater with DBT (p < 0.001). CONCLUSION. Reader performance in breast cancer detection is significantly higher with wide-angle two-view DBT independent of FFDM, verifying the robustness of DBT as a sole view. However, results of perception studies in the vision sciences support the inclusion of an overview image.

Three-dimensionally-printed anthropomorphic physical phantom for mammography and digital breast tomosynthesis with custom materials, lesions, and uniform quality control region.

Anthropomorphic breast phantoms mimic patient anatomy in order to evaluate clinical mammography and digital breast tomosynthesis system performance. Our goal is to create a modular phantom with an anthropomorphic region to allow for improved lesion and calcification detection as well as a uniform region to evaluate standard quality control (QC) metrics. Previous versions of this phantom used commercial photopolymer inkjet three-dimensional printers to recreate breast anatomy using four surfaces that were fabricated with commercial materials spanning only a limited breast density range of 36% to 64%. We use modified printers to create voxelized, dithered breast phantoms with continuous gradations between glandular and adipose tissues. Moreover, the new phantom replicates the low-end density (representing adipose tissue) using third party material, Jf Flexible, and increases the high-end density to the density of glandular tissue and beyond by either doping Jf Flexible with salts and nanoparticles or using a new commercial resin, VeroPureWhite. An insert design is utilized to add masses, calcifications, and iodinated objects into the phantom for increased utility. The uniform chest wall region provides a space for traditional QC objects such as line pair patterns for measuring resolution and scale bars for measuring printer linearity. Incorporating these distinct design modules enables us to create an improved, complete breast phantom to better evaluate clinical mammography systems for lesion and calcification detection and standard QC performance evaluation.

Prediction of Occult Invasive Disease in Ductal Carcinoma in Situ Using Deep Learning Features.

PURPOSE: The aim of this study was to determine whether deep features extracted from digital mammograms using a pretrained deep convolutional neural network are prognostic of occult invasive disease for patients with ductal carcinoma in situ (DCIS) on core needle biopsy. METHODS: In this retrospective study, digital mammographic magnification views were collected for 99 subjects with DCIS at biopsy, 25 of which were subsequently upstaged to invasive cancer. A deep convolutional neural network model that was pretrained on nonmedical images (eg, animals, plants, instruments) was used as the feature extractor. Through a statistical pooling strategy, deep features were extracted at different levels of convolutional layers from the lesion areas, without sacrificing the original resolution or distorting the underlying topology. A multivariate classifier was then trained to predict which tumors contain occult invasive disease. This was compared with the performance of traditional "handcrafted" computer vision (CV) features previously developed specifically to assess mammographic calcifications. The generalization performance was assessed using Monte Carlo cross-validation and receiver operating characteristic curve analysis. RESULTS: Deep features were able to distinguish DCIS with occult invasion from pure DCIS, with an area under the receiver operating characteristic curve of 0.70 (95% confidence interval, 0.68-0.73). This performance was comparable with the handcrafted CV features (area under the curve = 0.68; 95% confidence interval, 0.66-0.71) that were designed with prior domain knowledge. CONCLUSIONS: Despite being pretrained on only nonmedical images, the deep features extracted from digital mammograms demonstrated comparable performance with handcrafted CV features for the challenging task of predicting DCIS upstaging.