Breast Density Classification Research Articles

Enhancing Accuracy in Breast Density Assessment Using Deep Learning: A Multicentric, Multi-Reader Study

The evaluation of mammographic breast density, a critical indicator of breast cancer risk, is traditionally performed by radiologists via visual inspection of mammography images, utilizing the Breast Imaging-Reporting and Data System (BI-RADS) breast density categories. However, this method is subject to substantial interobserver variability, leading to inconsistencies and potential inaccuracies in density assessment and subsequent risk estimations. To address this, we present a deep learning-based automatic detection algorithm (DLAD) designed for the automated evaluation of breast density. Our multicentric, multi-reader study leverages a diverse dataset of 122 full-field digital mammography studies (488 images in CC and MLO projections) sourced from three institutions. We invited two experienced radiologists to conduct a retrospective analysis, establishing a ground truth for 72 mammography studies (BI-RADS class A: 18, BI-RADS class B: 43, BI-RADS class C: 7, BI-RADS class D: 4). The efficacy of the DLAD was then compared to the performance of five independent radiologists with varying levels of experience. The DLAD showed robust performance, achieving an accuracy of 0.819 (95% CI: 0.736–0.903), along with an F1 score of 0.798 (0.594–0.905), precision of 0.806 (0.596–0.896), recall of 0.830 (0.650–0.946), and a Cohen’s Kappa (κ) of 0.708 (0.562–0.841). The algorithm achieved robust performance that matches and in four cases exceeds that of individual radiologists. The statistical analysis did not reveal a significant difference in accuracy between DLAD and the radiologists, underscoring the model’s competitive diagnostic alignment with professional radiologist assessments. These results demonstrate that the deep learning-based automatic detection algorithm can enhance the accuracy and consistency of breast density assessments, offering a reliable tool for improving breast cancer screening outcomes.

Fair evaluation of federated learning algorithms for automated breast density classification: The results of the 2022 ACR-NCI-NVIDIA federated learning challenge

The correct interpretation of breast density is important in the assessment of breast cancer risk. AI has been shown capable of accurately predicting breast density, however, due to the differences in imaging characteristics across mammography systems, models built using data from one system do not generalize well to other systems. Though federated learning (FL) has emerged as a way to improve the generalizability of AI without the need to share data, the best way to preserve features from all training data during FL is an active area of research. To explore FL methodology, the breast density classification FL challenge was hosted in partnership with the American College of Radiology, Harvard Medical Schools’ Mass General Brigham, University of Colorado, NVIDIA, and the National Institutes of Health National Cancer Institute. Challenge participants were able to submit docker containers capable of implementing FL on three simulated medical facilities, each containing a unique large mammography dataset. The breast density FL challenge ran from June 15 to September 5, 2022, attracting seven finalists from around the world. The winning FL submission reached a linear kappa score of 0.653 on the challenge test data and 0.413 on an external testing dataset, scoring comparably to a model trained on the same data in a central location.

Cancer detection rate of breast-MR in supplemental screening after negative mammography in women with dense breasts. Preliminary results of the MA-DETECT-Study after 200 participants

BACKGROUNDDue to increased cancer detection rates (CDR), breast MR (breast MRI) can reduce underdiagnosis of breast cancer compared to conventional imaging techniques, particularly in women with dense breasts. The purpose of this study is to report the additional breast cancer yield by breast MRI in women with dense breasts after receiving a negative screening mammogram. METHODSFor this study we invited consecutive participants of the national German breast cancer Screening program with breast density categories ACR C & D and a negative mammogram to undergo additional screening by breast MRI. Endpoints were CDR and recall rates. This study reports interim results in the first 200 patients. At a power of 80% and considering an alpha error of 5%, this preliminary population size is sufficient to demonstrate a 4/1000 improvement in CDR. RESULTSIn 200 screening participants, 8 women (40/1000, 17.4–77.3/1000) were recalled due to positive breast MRI findings. Image-guided biopsy revealed 5 cancers in 4 patients (one bilateral), comprising four invasive cancers and one case of DCIS. 3 patients revealed 4 invasive cancers presenting with ACR C breast density and one patient non-calcifying DCIS in a woman with ACR D breast density, resulting in a CDR of 20/1000 (95%-CI 5.5–50.4/1000) and a PPV of 50% (95%-CI 15.7–84.3%). CONCLUSIONOur initial results demonstrate that supplemental screening using breast MRI in women with heterogeneously dense and very dense breasts yields an additional cancer detection rate in line with a prior randomized trial on breast MRI screening of women with extremely dense breasts. These findings are highly important as the population investigated constitutes a much higher proportion of women and yielded cancers particularly in women with heterogeneously dense breasts.

Breast density classification in mammograms using VGG convolutional networks

In recent years, multiple studies have highlighted the growing correlation between breast density and the risk of developing breast cancer. In this research, the performance of two convolutional neural network architectures, VGG16 and VGG19, was evaluated for breast density classification across three distinct scenarios aimed to compare the masking effect on the models performance. These scenarios encompass both binary classification (fatty and dense) and multi-class classification based on the BI-RADS categorization, utilizing a subset of the ABC-Digital Mammography Dataset. In the first experiment, focusing on cases with no masses, VGG16 achieved an accuracy of 93.33% and 90.00% for two and four-class classification. The second experiment, which involved cases with benign masses, yielded a remarkable accuracy of 95.83% and 93.33% with VGG16, respectively. In the third and last experiment, an accuracy of 88.00% was obtained using VGG16 for the two-class classification, while VGG19 delivered an accuracy of 93.33% for the four-class classification. These findings underscore the potential of deep learning models in enhancing breast density classification, with implications for breast cancer risk assessment and early detection.

Screening mammography frequency following dense breast notification among a predominantly Hispanic/Latina screening cohort.

Nationally legislated dense breast notification (DBN) informs women of their breast density (BD) and the impact of BD on breast cancer risk and detection, but consequences for screening participation are unclear. We evaluated the association of DBN in New York State (NYS)with subsequent screening mammography in a largely Hispanic/Latina cohort. Women aged 40-60 were surveyed in their preferred language (33% English, 67% Spanish) during screening mammography from 2016 to 2018. We used clinical BD classification from mammography records from 2013 (NYSDBN enactment) through enrollment (baseline) to create a 6-category variable capturing prior and new DBN receipt (sent only after clinically dense mammograms). We used this variable to compare the number of subsequent mammograms (0, 1, ≥ 2) from 10 to 30months after baseline using ordinal logistic regression. In a sample of 728 women (78% foreign-born, 72% Hispanic, 46% high school education or less), first-time screeners and women who received DBN for the first timeafter prior non-dense mammograms had significantly fewer screening mammograms within 30months of baseline (Odds Ratios range: 0.33 (95% Confidence Interval (CI) 0.12-0.85) to 0.38 (95% CI 0.17-0.82)) compared to women with prior mammography but no DBN. There were no differences in subsequent mammogram frequency between women with multiple DBN and those who never received DBN. Findings were consistent across age, language, health literacy, and educationgroups. Women receiving their first DBN after previous non-dense mammograms have lower mammography participation within 2.5years. DBN has limited influence on screening participation of first-time screeners and those with persistent dense mammograms.

Tailored 3D breast models for development of microwave based breast tumor screening

Portable breast monitoring devices, which could be used outside the hospital and even for self-monitoring of risk groups, are considered as promising eHealth applications for future telemedicine. Microwave technique is one of the most promising emerging techniques for portable monitoring devices since it enables low-cost, high accuracy, and user-friendly devices. The technique is based on detecting differences in radio channel responses since tumors have different dielectric properties than the breast glandular or fat tissues. Breast density categories affect the detectability of the tumors with also microwave technique. This paper presents a study on breast tumor detectability with different simulation and measurement models tailored to correspond different breast types. The simulations are carried out using electromagnetic simulation software with human voxel models as well as developed breast models. The evaluation results show that even small-sized breast tumors can be detected with microwave technique within all the breast density categories. Differences in channel responses caused by tumors are breast-type dependent. The results highlight the importance of developing extensive reference databank covering all the breast density categories for microwave-based breast tumor detection applications.

Application of Artificial Intelligence (AI) System in Opportunistic Screening and Diagnostic Population in a Middle-income Nation.

This study evaluates the effectiveness of artificial intelligence (AI) in mammography in a diverse population from a middle-income nation and compares it to traditional methods. A retrospective study was conducted on 543 mammograms of 467 Malays, 48 Chinese, and 28 Indians in a middle-income nation. Three breast radiologists interpreted the examinations independently in two reading sessions (with and without AI support). Breast density and BI-RADS categories were assessed, comparing the accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) results. Of 543 mammograms, 69.2% had lesions detected. Biopsies were performed on 25%(n=136), with 66(48.5%) benign and 70(51.5%) malignant. Substantial agreement in density assessment between the radiologist and AI software (κ =0.606, p < 0.001) and the BI-RADS category with and without AI (κ =0.74, p < 0.001). The performance of the AI software was comparable to the traditional methods. The sensitivity, specificity, PPV, and NPV or radiologists alone, radiologist + AI, and AI alone were 81.9%,90.4%,56.0%, and 97.1%; 81.0%, 93.1%,55.5%, and 97.0%; and 90.0%,76.5%,36.2%, and 98.1%, respectively. AI software enhances the accuracy of lesion diagnosis and reduces unnecessary biopsies, particularly for BI-RADS 4 lesions. The AI software results for synthetic were almost similar to the original 2D mammography, with AUC of 0.925 and 0.871, respectively. AI software may assist in the accurate diagnosis of breast lesions, enhancing the efficiency of breast lesion diagnosis in a mixed population of opportunistic screening and diagnostic patients. • The use of artificial intelligence (AI) in mammography for population-based breast cancer screening has been validated in high-income nations, with reported improved diagnostic performance. Our study evaluated the usage of an AI tool in an opportunistic screening setting in a multi-ethnic and middle-income nation. • The application of AI in mammography enhances diagnostic accuracy, potentially leading to reduced unnecessary biopsies. • AI integration into the workflow did not disrupt the performance of trained breast radiologists, as there is a substantial inter-reader agreement for BI-RADS category assessment and breast density.

Abstract B078: Artificial-intelligence-driven breast density assessment in the transition from full-field digital mammograms to digital breast tomosynthesis

Abstract Introduction: To enhance reproducibility and robustness in mammographic density assessment, various artificial intelligence (AI) models have been proposed to automatically classify mammographic images into BI-RADS density categories. Despite their promising performances, so far density AI models have been assessed primarily in traditional full-field digital mammography (FFDM) images. Our study aims to assess the potential of AI in breast density assessment in FFDM versus the newer synthetic mammography (SM) images acquired with digital breast tomosynthesis. Methods: We retrospectively analyzed negative (BI-RADS 1 or 2) routine mammographic screening exams (Selenia or Selenia Dimensions; Hologic) acquired at sites within the Barnes-Jewish/Christian (BJC) Healthcare network in St. Louis, MO from 2015 to 2018. BI-RADS breast density assessments of radiologists were obtained from BJC’s mammography reporting software (Magview 7.1). For each mammographic imaging modality, a balanced dataset of 4,000 women was selected so there were equal numbers of women in each of the four BI-RADS density categories, and each woman had at least one mediolateral oblique (MLO) and one craniocaudal (CC) view per breast in that mammographic imaging modality. Previously validated pre-processing steps were applied to all FFDM and SM images to standardize image orientation and intensity. Images were then split into training, validation, and test sets at ratios of 80%, 10%, and 10%, respectively, while maintaining the distribution of breast density categories and ensuring that all images of the same woman appear only in one set. Our AI model was based on the widely used ResNet50 architecture and was designed to accept as an input a mammographic image and predict the BI-RADS breast density category that the image belongs to. Our AI model was optimized, trained, and evaluated separately for each mammographic imaging modality. We report on the AI model’s predictive accuracy on the test set for each mammographic imaging modality, for both views as well as separately for CC and MLO; accuracy differences in FFDM versus SM were assessed via bootstrapping. Results: A batch size of 32, learning rate of e-6, and Adam optimizer were chosen as the optimal hyperparameters for our AI model. Using the same hyperparameters, the AI model demonstrated substantially higher accuracy on the test set for FFDM than for SM (FFDM: accuracy = 71% ± 4.5% versus SM: accuracy = 66% ± 4.2%; p-value&amp;lt;0.001 for comparison). Similar conclusion held when CC and MLO views were evaluated separately (accuracy = 72% ± 4.6% versus 66% ± 4.3% for CC; accuracy = 69% ± 4.5% versus 62% ± 4.3% for MLO; p-value&amp;lt;0.001 for both comparisons). Conclusions: AI performance in BI-RADS breast density assessment was significantly higher on FFDM versus SM, even under the same AI model design, dataset size and training process. Our preliminary findings suggest that further AI optimizations and adaptations may be needed as we translate AI models from FFDM to the newer SM format acquired with digital breast tomosynthesis. Citation Format: Krisha Anant, Juanita Hernandez Lopez, Debbie Bennett, Aimilia Gastounioti. Artificial-intelligence-driven breast density assessment in the transition from full-field digital mammograms to digital breast tomosynthesis [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Breast Cancer Research; 2023 Oct 19-22; San Diego, California. Philadelphia (PA): AACR; Cancer Res 2024;84(3 Suppl_1):Abstract nr B078.

Development of Auto Dense-breast Classification on Mammography Images Using Image Similarity

In Japan, radiologists perform qualitative visual classification to define four categories of mammary gland density. However, an objective estimation of mammary gland density is necessary. To address this, we developed an automatic classification software using image similarity. We prepared 741 cases of mediolateral oblique images (MLO) for evaluation, and they were diagnosed as normal among the mammography images taken at our hospital. Image matching was performed using the evaluation images and an image database for breast density determination. In this study, the image similarity used zero normalized cross-correlation (ZNCC) as an index. In addition, if the breast thickness is less than 30 mm and each breast density category ZNNC has the same value, the category is evaluated on the fat side. We compared the results of qualitative visual classification and automatic classification methods to assess consistency. The agreement with the subjective breast composition classification was 78.5%, and the weighted kappa coefficient was 0.98. One mismatched case was evaluated on the higher density side with the same ZNCC value between categories and a breast thickness greater than 30 mm. Image similarity provides an excellent estimation of quantification of breast density. This system could contribute to improving the efficiency of the mammography screening system.

Imaging characteristics of interval cancers detected on Full Field Digital Mammography (FFDM) versus Digital Breast Tomosynthesis (DBT)

ObjectiveTo compare imaging features of interval cancers detected in patients screened with full field digital mammography (FFDM) versus digital breast tomosynthesis (DBT). Materials/MethodsThis retrospective observational study consisted of female patients undergoing screening DM or FFDM at an academic medical center and two outpatient imaging facilities between January 2012 and June 2017. A natural language processing algorithm queried breast imaging reports for breast density and BI-RADS category. This was cross-referenced to an institutional breast cancer registry to identify interval cancers. Retrospective consensus review of the cases was done to categorize imaging features of interval cancers on FFDM vs DBT. ResultsThe rate of interval cancers was comparable in patients screened with FFDM (30/39793) and DBT (29/32180) (p = 0.58). There was no significant difference in the rate, histopathology, or imaging features of interval cancers in patients screened with FFDM versus DBT. The most common mammographic features on diagnostic imaging across both groups was the presence of a mass (13/47). Almost equally common was negative diagnostic mammogram with mass detected only on ultrasound (11/47). The rate of interval cancers detected by high-risk surveillance breast MRI was increased in patients who previously had screening with DBT relative to those who had screening with FFDM (p = 0.0419). ConclusionThere is no significant difference in rate of detection, histopathology, or imaging features of interval cancers in patients screened with FFDM versus DBT. However, across both cohorts, the most common features on diagnostic mammogram were either the presence of a mass or a negative mammogram.

Multi-view fusion-based local-global dynamic pyramid convolutional cross-tansformer network for density classification in mammography

Object. Breast density is an important indicator of breast cancer risk. However, existing methods for breast density classification do not fully utilise the multi-view information produced by mammography and thus have limited classification accuracy. Method. In this paper, we propose a multi-view fusion network, denoted local-global dynamic pyramidal-convolution transformer network (LG-DPTNet), for breast density classification in mammography. First, for single-view feature extraction, we develop a dynamic pyramid convolutional network to enable the network to adaptively learn global and local features. Second, we address the problem exhibited by traditional multi-view fusion methods, this is based on a cross-transformer that integrates fine-grained information and global contextual information from different views and thereby provides accurate predictions for the network. Finally, we use an asymmetric focal loss function instead of traditional cross-entropy loss during network training to solve the problem of class imbalance in public datasets, thereby further improving the performance of the model. Results. We evaluated the effectiveness of our method on two publicly available mammography datasets, CBIS-DDSM and INbreast, and achieved areas under the curve (AUC) of 96.73% and 91.12%, respectively. Conclusion. Our experiments demonstrated that the devised fusion model can more effectively utilise the information contained in multiple views than existing models and exhibits classification performance that is superior to that of baseline and state-of-the-art methods.

Breast density and the likelihood of malignant MRI-detected lesions in women diagnosed with breast cancer

ObjectivesTo assess whether mammographic breast density in women diagnosed with breast cancer correlates with the total number of incidental magnetic resonance imaging (MRI)-detected lesions and the likelihood of the lesions being malignant.MethodsPatients diagnosed with breast cancer meeting the EUSOBI and EUSOMA criteria for preoperative breast MRI routinely undergo mammography and ultrasound before MRI at our institution. Incidental suspicious breast lesions detected in MRI are biopsied. We included patients diagnosed with invasive breast cancers between 2014 and 2019 who underwent preoperative breast MRI. One reader retrospectively determined breast density categories according to the 5th edition of the BI-RADS lexicon.ResultsOf 946 patients with 973 malignant primary breast tumors, 166 (17.5%) had a total of 175 (18.0%) incidental MRI-detected lesions (82 (46.9%) malignant and 93 (53.1%) benign). High breast density according to BI-RADS was associated with higher incidence of all incidental enhancing lesions in preoperative breast MRIs: 2.66 (95% confidence interval: 1.03–6.86) higher for BI-RADS density category B, 2.68 (1.04–6.92) for category C, and 3.67 (1.36–9.93) for category D compared to category A (p < 0.05). However, high breast density did not predict higher incidence of malignant incidental lesions (p = 0.741). Incidental MRI-detected lesions in the contralateral breast were more likely benign (p < 0.001): 18 (27.3%)/48 (72.7%) vs. 64 (58.7%)/45 (41.3%) malignant/benign incidental lesions in contralateral vs. ipsilateral breasts.ConclusionWomen diagnosed with breast cancer who have dense breasts have more incidental MRI-detected lesions, but higher breast density does not translate to increased likelihood of malignant incidental lesions.Clinical relevance statementDense breasts should not be considered as an indication for preoperative breast MRI in women diagnosed with breast cancer.Key Points• The role of preoperative MRI of patients with dense breasts diagnosed with breast cancer is under debate.• Women with denser breasts have a higher incidence of all MRI-detected incidental breast lesions, but the incidence of malignant MRI-detected incidental lesions is not higher than in women with fatty breasts.• High breast density alone should not indicate preoperative breast MRI.

Automatic Mammographic breast density classification Using machine learning approach

Automatic Mammographic breast density classification Using machine learning approach

Standardization of Breast Dynamic Contrast-enhanced MRI Signal with Application to the Assessment of Background Parenchymal Enhancement Rate.

There is currently no clinically accepted method for quantifying background parenchymal enhancement (BPE), though a sensitive method might allow individualized risk management based on the response to cancer-preventative hormonal therapy. The objective of this pilot study is to demonstrate the utility of linear modeling of standardized dynamic contrast-enhanced MRI (DCEMRI) signal for quantifying changes in BPE rates. On a retrospective database search, 14 women with DCEMRI examinations pre- and post- treatment with tamoxifen were identified. DCEMRI signal was averaged over the parenchymal ROIs to obtain time-dependent signal curves S(t). The gradient echo signal equation was used to standardize scale S(t) to values of (FA) ̃ = 10° and (TR) ̃ = 5.5 ms, and obtain the standardized parameters of DCE-MRI signal S ̃_p (t). Relative signal enhancement (〖RSE〗_p ) ̃ was calculated from S ̃_p, and the reference tissue method for T1 calculation was used to standardize (〖RSE〗_p ) ̃ to gadodiamide as the contrast agent, obtaining (RSE) ̃. (RSE) ̃, in the first 6 minutes, post-contrast administration was fit to a linear model with the slope α ̃_RSE denoting the standardized rate relative BPE. Changes in α ̃_RSE were not found to be significantly correlated with the average duration of tamoxifen treatment, age at the initiation of preventative treatment, or pre-treatment BIRADS breast density category. The average change in α ̃_RSE showed a large effect size of -1.12, significantly higher than -0.86 observed without signal standardization (p < 0.01). Linear modeling of BPE in standardized DCEMRI can provide quantitative measurements of BPE rates, improving sensitivity to changes due to tamoxifen treatment.

Evaluation of artificial intelligent breast ultrasound on lesion detection and characterization compared with hand-held ultrasound in asymptomatic women.

To compare the accuracy of Artificial Intelligent Breast Ultrasound (AIBUS) with hand-held breast ultrasound (HHUS) in asymptomatic women and to offer recommendations for screening in regions with limited medical resources. 852 participants who underwent both HHUS and AIBUS were enrolled between December 2020 and June 2021. Two radiologists, who were unaware of the HHUS results, reviewed the AIBUS data and scored the image quality on a separate workstation. Breast imaging reporting and data system (BI-RADS) final recall assessment, breast density category, quantified lesion features, and examination time were evaluated for both devices. The statistical analysis included McNemar's test, paired t-test, and Wilcoxon test. The kappa coefficient and consistency rate were calculated in different subgroups. Subjective satisfaction with AIBUS image quality reached 70%. Moderate agreements were found between AIBUS with good quality images and HHUS for the BI-RADS final recall assessment (κ = 0.47, consistency rate = 73.9%) and breast density category (κ = 0.50, consistency rate = 74.8%). The lesions measured by AIBUS were statistically smaller and deeper than those measured by HHUS (P < 0.001), though they were not significant in clinical diagnosis (all < 3mm). The total time required for the AIBUS examination and image interpretation was 1.03 (95% CI (0.57, 1.50)) minutes shorter than that of HHUS per case. Moderate agreement was obtained for the description of the BI-RADS final recall assessment and breast density category. With image quality comparable to that of HHUS, AIBUS was superior for the efficiency of primary screening.

A magnification-independent method for breast cancer classification using transfer learning

Breast cancer is the most common and diagnosed cancer among women worldwide. Doctors use breast imaging reporting and data systems to classify breast density. This study proposes a transfer learning-based artificially intelligent (AI) system to classify breast cancer from the histopathological images of the breast. Three different Deep Convolutional Neural Networks (DCNN) architectures, namely, the Visual Geometry Group-16 (VGG-16), Depthwise Separable Convolutions (Xception), and Dense Convolutional Network-201 (Densenet-201), have been used as a base model in the transfer learning method used in this work. Each test image has been classified into benign or malignant class after extracting the features from each test image using the three pre-trained base models. The proposed classification system is independent of the magnification factors of the images. The performance of the proposed breast cancer classification system has been evaluated on two widely used public datasets known as the Invasive Ductal Carcinoma (IDC) dataset and BreaKHis dataset. The classification accuracies of 99.42% (IDC dataset) and 99.12% (BreaKHis dataset) have been obtained from the proposed system. The results demonstrate that the proposed breast cancer classification system outperforms the state-of-the-art methods in this research domain.

Global Radiomic Features from Mammography for Predicting Difficult-To-Interpret Normal Cases.

This work aimed to investigate whether global radiomic features (GRFs) from mammograms can predict difficult-to-interpret normal cases (NCs). Assessments from 537 readers interpreting 239 normal mammograms were used to categorise cases as 120 difficult-to-interpret and 119 easy-to-interpret based on cases having the highest and lowest difficulty scores, respectively. Using lattice- and squared-based approaches, 34 handcrafted GRFs per image were extracted and normalised. Three classifiers were constructed: (i) CC and (ii) MLO using the GRFs from corresponding craniocaudal and mediolateral oblique images only, based on the random forest technique for distinguishing difficult- from easy-to-interpret NCs, and (iii) CC + MLO using the median predictive scores from both CC and MLO models. Useful GRFs for the CC and MLO models were recognised using a scree test. The CC and MLO models were trained and validated using the leave-one-out-cross-validation. The models' performances were assessed by the AUC and compared using the DeLong test. A Kruskal-Wallis test was used to examine if the 34 GRFs differed between difficult- and easy-to-interpret NCs and if difficulty level based on the traditional breast density (BD) categories differed among 115 low-BD and 124 high-BD NCs. The CC + MLO model achieved higher performance (0.71 AUC) than the individual CC and MLO model alone (0.66 each), but statistically non-significant difference was found (all p > 0.05). Six GRFs were identified to be valuable in describing difficult-to-interpret NCs. Twenty features, when compared between difficult- and easy-to-interpret NCs, differed significantly (p < 0.05). No statistically significant difference was observed in difficulty between low- and high-BD NCs (p = 0.709). GRF mammographic analysis can predict difficult-to-interpret NCs.

Breast Density Evaluation According to BI-RADS 5th Edition on Digital Breast Tomosynthesis: AI Automated Assessment Versus Human Visual Assessment.

The assessment of breast density is one of the main goals of radiologists because the masking effect of dense fibroglandular tissue may affect the mammographic identification of lesions. The BI-RADS 5th Edition has revised the mammographic breast density categories, focusing on a qualitative evaluation rather than a quantitative one. Our purpose is to compare the concordance of the automatic classification of breast density with the visual assessment according to the latest available classification. A sample of 1075 digital breast tomosynthesis images from women aged between 40 and 86 years (58 ± 7.1) was retrospectively analyzed by three independent readers according to the BI-RADS 5th Edition. Automated breast density assessment was performed on digital breast tomosynthesis images with the Quantra software version 2.2.3. Interobserver agreement was assessed with kappa statistics. The distributions of breast density categories were compared and correlated with age. The agreement on breast density categories was substantial to almost perfect between radiologists (κ = 0.63-0.83), moderate to substantial between radiologists and the Quantra software (κ = 0.44-0.78), and the consensus of radiologists and the Quantra software (κ = 0.60-0.77). Comparing the assessment for dense and non-dense breasts, the agreement was almost perfect in the screening age range without a statistically significant difference between concordant and discordant cases when compared by age. The categorization proposed by the Quantra software has shown a good agreement with the radiological evaluations, even though it did not completely reflect the visual assessment. Thus, clinical decisions regarding supplemental screening should be based on the radiologist's perceived masking effect rather than the data produced exclusively by the Quantra software.

Effect of contraceptive hormonal therapy on mammographic breast density: A longitudinal cohort study

PurposeEvaluate the longitudinal relationship between mammographic density and hormonal contraceptive use in late reproductive-aged women. MethodsPatients aged 35–50 years old who underwent 5 or more screening mammograms within a 7.5-year period between 2004 and 2019 in a single urban tertiary care center were randomly selected. Patients were categorized into four cohorts based on hormonal contraceptive exposure during a 2-year lead-in period and a 7.5-year study period: 1) never exposed, 2) always exposed, 3) interval hormonal contraceptive start, and 4) interval hormonal contraceptive stop. The primary outcome was difference in BI-RADS breast density category between initial and final mammograms. ResultsOf the 708 patients included, long-term use of combined oral contraceptives or a levonorgestrel intrauterine device were not associated with an increase in breast density category over the 7.5-year study period, compared to those with no hormonal contraceptive exposure. Initiation of combined oral contraceptives was associated with an increase in breast density category (β = 0.31, P = 0.045); however, no difference in initial density category was noted between those exposed and those never exposed to combined oral contraceptives during the 2-year lead-in period, and discontinuation was not associated with a decrease in breast density category when compared to those with continuous exposure. Conclusion(s)Long-term use of combined oral contraceptives or a levonorgestrel intrauterine device was not associated with an increase in BI-RADS breast density category. Initiation of a combined oral contraceptive was associated with an increase in breast density category, although this may be a transient effect.

Abstract P3-03-02: Can artificial intelligence derived ultrasound breast density provide comparable breast cancer risk estimates to density derived from mammograms

Abstract Can artificial intelligence derived ultrasound breast density provide comparable breast cancer risk estimates to density derived from mammograms Dustin Valdez12, Arianna Bunnell2, Thomas Wolfgruber1, Aleen V. Altamirano3, Brandon Quon1, Gertraud Maskarinec1, Peter Sadowski2, John A. Shepherd1 1 University of Hawaii Cancer Center, Honolulu, HI 2 University of Hawaii at Manoa, Honolulu, HI 3 Instituto Radiodiagnóstico, Managua, Nicaragua Background: Breast cancer is the second leading cause of cancer-related death among women in Hawaii and the Pacific. However, while there are programs like the Breast and Cervical Cancer Early Detection Program (BCCEDP) implemented throughout the Pacific, the lack of access to mammography screening and low screening participation rates contributes to very high advanced breast cancer rates in most cases over 50%. Portable breast ultrasound is a promising screening technology for low resource areas. However, without mammography, mammographic density is not available for risk modeling to determine who should participate in screening programs or at what frequency. In this study, we ask if breast ultrasound (US) images can be used to derive an equivalent mammographic density for risk modeling. We utilized artificial intelligence to derive breast density from diagnostic ultrasound images and compared to BI-RADS mammographic density in an established breast cancer risk model1. Methods: We selected women with negative screening visit who either later developed cancer (positives) or did not (negatives) over a 10-year period. Temporally-matched negative mammographic and ultrasound images, cancer outcome status and cancer risk information were sourced from the Hawaii and Pacific Islands Mammography Registry. US images had to have occurred within a year of the mammogram. BI-RADS mammographic density was derived using an existing deep neural network model2. Mammographic density was estimated from US images by training a deep-learning convolutional neural network model. A hold out set of images (Test set of 20% of the total) was used to compare 10-year breast cancer risk using the Tyrer-Cuzick (TC) risk model1 when calculated using breast density from either mammograms or US. The AUC values, confidence intervals, ROC plots and Pearson correlation were calculated and compared. Results: Over the 10-year study period, 1337 had matched mammograms and US images and 65 went on to develop breast cancer. Using the test set, the Pearson’s correlation between breast density from mammography and US was 0.31 (moderate correlation). There were no covariates found to improve this association. The AUC for TC 10-year personal risk was higher when breast density from mammograms was used 0.71 (95% CI=0.57-0.86) versus US images 0.65 (95% CI=0.53-0.76). Conclusion: Overall breast cancer risk was similar when breast density was derived from either mammograms or US. The performance of our US breast density model is expected to improve further when more US training data becomes available. Breast cancer screening programs exclusively using US imaging may be able to provide equivalent risk modeling to clinics using mammography. 1.Tyrer J, Duffy SW, Cuzick J (2004). A breast cancer prediction model incorporating familial and personal risk factors. Stat Med. 2004 Apr 15;23(7):1111-30. doi: 10.1002/sim.1668. Erratum in: Stat Med. 2005 Jan 15;24(1):156. PMID: 15057881. 2. Wu, N., K. J. Geras, Y. Shen, J. Su, S. G. Kim, E. Kim, S. Wolfson, L. Moy and K. Cho (2018). Breast Density Classification with Deep Convolutional Neural Networks. 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE. Citation Format: Dustin Valdez, Arianna Bunnell, Thomas Wolfgruber, Aleen Altamirano, Brandon Quon, Gertraud Maskarinec, Peter Sadowski, John Shepherd. Can artificial intelligence derived ultrasound breast density provide comparable breast cancer risk estimates to density derived from mammograms [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P3-03-02.