Full-field Digital Mammography Images Research Articles

Generalisable deep learning method for mammographic density prediction across imaging techniques and self-reported race.

Breast density is an important risk factor for breast cancer complemented by a higher risk of cancers being missed during screening of dense breasts due to reduced sensitivity of mammography. Automated, deep learning-based prediction of breast density could provide subject-specific risk assessment and flag difficult cases during screening. However, there is a lack of evidence for generalisability across imaging techniques and, importantly, across race. This study used a large, racially diverse dataset with 69,697 mammographic studies comprising 451,642 individual images from 23,057 female participants. A deep learning model was developed for four-class BI-RADS density prediction. A comprehensive performance evaluation assessed the generalisability across two imaging techniques, full-field digital mammography (FFDM) and two-dimensional synthetic (2DS) mammography. A detailed subgroup performance and bias analysis assessed the generalisability across participants' race. Here we show that a model trained on FFDM-only achieves a 4-class BI-RADS classification accuracy of 80.5% (79.7-81.4) on FFDM and 79.4% (78.5-80.2) on unseen 2DS data. When trained on both FFDM and 2DS images, the performance increases to 82.3% (81.4-83.0) and 82.3% (81.3-83.1). Racial subgroup analysis shows unbiased performance across Black, White, and Asian participants, despite a separate analysis confirming that race can be predicted from the images with a high accuracy of 86.7% (86.0-87.4). Deep learning-based breast density prediction generalises across imaging techniques and race. No substantial disparities are found for any subgroup, including races that were never seen during model development, suggesting that density predictions are unbiased.

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<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<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.

YOLO-based CAD framework with ViT transformer for breast mass detection and classification in CESM and FFDM images

Breast cancer detection is considered a challenging task for the average experienced radiologist due to the variation of the lesions’ size and shape, especially with the existence of high fibro-glandular tissues. The revolution of deep learning and computer vision contributes recently in introducing systems that can provide an automated diagnosis for breast cancer that can act as a second opinion for doctors/radiologists. The most of previously proposed deep learning-based Computer-Aided Diagnosis (CAD) systems mainly utilized Convolutional Neural Networks (CNN) that focuses on local features. Recently, vision transformers (ViT) have shown great potential in image classification tasks due to its ability in learning the local and global spatial features. This paper proposes a fully automated CAD framework based on YOLOv4 network and ViT transformers for mass detection and classification of Contrast Enhanced Spectral Mammography (CESM) images. CESM is an evolution type of Full Field Digital Mammography (FFDM) images that provides enhanced visualization for breast tissues. Different experiments were conducted to evaluate the proposed framework on two different datasets that are INbreast and CDD-CESM that provides both FFDM and CESM images. The model achieved at mass detection a mean Average Precision (mAP) score of 98.69%, 81.52%, and 71.65% and mass classification accuracy of 95.65%, 97.61%, and 80% for INbreast, CE-CESM, and DM-CESM, respectively. The proposed framework showed competitive results regarding the state-of-the-art models in INbreast. It outperformed the previous work in the literature in terms of the F1-score by almost 5% for mass detection in CESM. Moreover, the experiments showed that the CESM could provide more morphological features that can be more informative, especially with the highly dense breast tissues.

MLN-net: A multi-source medical image segmentation method for clustered microcalcifications using multiple layer normalization

The accurate segmentation of clustered microcalcifications in mammography is crucial for the diagnosis and treatment of breast cancer. Despite exhibiting expert-level accuracy, recent deep learning advancements in medical image segmentation provide insufficient contribution to practical applications, due to the domain shift resulting from differences in patient postures, individual gland density, and imaging modalities, etc. In this paper, a novel framework named MLN-net is proposed for clustered microcalcification segmentation. It can segment multi-source images using only single source images. Specifically, to rich domain distribution information, we introduce a source domain image augmentation for generating multi-source images. A structure of multiple layer normalization (LN) layers is then used to construct the segmentation network, which can be found efficient for clustered microcalcification segmentation in different domains. Additionally, a branch selection strategy is designed for measuring the similarity of the source domain data and the target domain data. To validate the proposed MLN-net, extensive analyses including ablation experiments are performed, comparison of 12 baseline methods. MLN-net enhances segmentation quality of full-field digital mammography (FFDM) and digital breast tomosynthe (DBT) images from the FFDM-DBT dataset, achieving the average Dice similarity coefficient (DSC) of 86.52% and the average Hausdorff distance (HD) of 20.49 mm on the source domain DBT. And it outperforms the baseline models for the task in FFDM images from both the CBIS-DDSM and the FFDM-DBT dataset, achieving the average DSC of 50.78% and the average HD of 35.12 mm on the source domain CBIS-DDSM. Extensive experiments validate the effectiveness of MLN-net in segmenting clustered microcalcifications from different domains and its segmentation accuracy surpasses state-of-the-art methods. Code will be available at https://github.com/yezanting/MLN-NET-VERSON1.

Examination of fully automated mammographic density measures using LIBRA and breast cancer risk in a cohort of 21,000 non-Hispanic white women

BackgroundBreast density is strongly associated with breast cancer risk. Fully automated quantitative density assessment methods have recently been developed that could facilitate large-scale studies, although data on associations with long-term breast cancer risk are limited. We examined LIBRA assessments and breast cancer risk and compared results to prior assessments using Cumulus, an established computer-assisted method requiring manual thresholding.MethodsWe conducted a cohort study among 21,150 non-Hispanic white female participants of the Research Program in Genes, Environment and Health of Kaiser Permanente Northern California who were 40–74 years at enrollment, followed for up to 10 years, and had archived processed screening mammograms acquired on Hologic or General Electric full-field digital mammography (FFDM) machines and prior Cumulus density assessments available for analysis. Dense area (DA), non-dense area (NDA), and percent density (PD) were assessed using LIBRA software. Cox regression was used to estimate hazard ratios (HRs) for breast cancer associated with DA, NDA and PD modeled continuously in standard deviation (SD) increments, adjusting for age, mammogram year, body mass index, parity, first-degree family history of breast cancer, and menopausal hormone use. We also examined differences by machine type and breast view.ResultsThe adjusted HRs for breast cancer associated with each SD increment of DA, NDA and PD were 1.36 (95% confidence interval, 1.18–1.57), 0.85 (0.77–0.93) and 1.44 (1.26–1.66) for LIBRA and 1.44 (1.33–1.55), 0.81 (0.74–0.89) and 1.54 (1.34–1.77) for Cumulus, respectively. LIBRA results were generally similar by machine type and breast view, although associations were strongest for Hologic machines and mediolateral oblique views. Results were also similar during the first 2 years, 2–5 years and 5–10 years after the baseline mammogram.ConclusionAssociations with breast cancer risk were generally similar for LIBRA and Cumulus density measures and were sustained for up to 10 years. These findings support the suitability of fully automated LIBRA assessments on processed FFDM images for large-scale research on breast density and cancer risk.

Classifying presence or absence of calcifications on mammography using generative contribution mapping.

The purpose of this study was to verify the efficacy of generative contribution mapping (GCM), an explainable deep learning model for images, in classifying the presence or absence of calcifications on mammography. The learning dataset consisted of 303 full-field digital mammography (FFDM) images labeled with microcalcifications obtained from the public INbreast database without extremely dense images. FFDM images were divided into calcification and non-calcification patch images using a sliding window method with 25% overlap. The patch images of the mediolateral oblique (MLO) and craniocaudal (CC) views were divided into a training set of 70%, a validation set of 10%, and a testing set of 20%. The classification performance of GCM classifiers was evaluated and compared with that of EfficientNet classifiers. Visualization maps of GCM highlighted regions of interest more clearly than EfficientNet's gradient-weighted class activation maps. The results showed that GCM classifiers yielded an accuracy of 0.92 (CC), 0.91 (MLO), and an area under the receiver operating characteristic curve of 0.92 (CC), 0.94 (MLO). In conclusion, GCM could accurately classify the presence or absence of calcifications on mammograms and explain intuitively reasonable grounds for their classification with visualization maps highlighting regions of interest.

Abstract 5879: The impact of radiomic imaging features in predicting short term risk of breast cancer after a negative mammogram using machine learning

Abstract Introduction & Purpose: While mammography reduces breast cancer mortality, up to 20% of breast cancers are diagnosed within the screening interval of a prior negative mammogram. Advances in breast imaging and machine learning have enabled evaluation of radiomic markers from mammography imaging with respect to breast cancer risk. Identification of radiomic markers associated with short term risk of breast cancer would be useful to direct supplemental screening. The aim of this study was to evaluate the potential of mammographic imaging features beyond breast density to predict screening failure using machine learning. Methods: We utilized full-field digital mammography (FFDM) studies from a cohort of women aged 40-85 who had a negative screening mammogram at the University of Pennsylvania Health System from 2010 - 2015. BRCA1/2 carriers, women with prior breast cancer, and women with breast implants were excluded. We identified 125 breast cancer cases diagnosed within 2 years of a negative mammogram whom we matched to controls not diagnosed with breast cancer within 2 years in a 1:5 ratio on BI-RADS breast density category, age, race, and screening date, resulting in a total of 750 FFDM studies. For each FFDM image, a total of 344 radiomic features were extracted using a previously validated lattice-based computational approach for parenchymal pattern characterization. Radiomic features were summarized across both breasts and across mammographic image views (CC and MLO) using means and standard deviations of radiomic features. Principal component analysis was applied to z-score normalized radiomic features and four principal components (PCs) were retained. Widely used machine learning models, such as Support Vector Machines (SVMs) and Random Forests (RFs), were applied. The performance of the machine learning models in predicting screening failure on the basis of the radiomic PCs was evaluated via cross-validated calculations of the area under the ROC curve (AUC). Results: Upon fine-tuning, SVM models using the mean summaries of radiomic features demonstrated a cross-validated case-control discriminatory capacity of AUC = 0.61 (95% CI 0.52, 0.68) while RFs did not achieve significant performance (AUC = 0.53, 95% CI 0.45, 0.60). Neither SVM models, nor RFs achieved significant predictive performance as measured by the cross-validated AUC when the standard deviation summaries of radiomic features was used. Conclusions: Given that BI-RADS density is the standard measure of breast density used clinically to direct supplemental screening, measuring the performance of radiomic features beyond BI-RADS provides the most clinically useful findings. In this small sample, our results suggest that mammographic radiomic features provided only moderate discriminatory accuracy, when BI-RADS density, age, race, and screening date were controlled for via matching. Citation Format: Aimilia Gastounioti, Sarah Ehsan, Walter Mankowski, Emily F. Conant, Despina Kontos, Anne Marie McCarthy. The impact of radiomic imaging features in predicting short term risk of breast cancer after a negative mammogram using machine learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5879.

Lesion-specific exposure parameters for breast cancer diagnosis on digital breast tomosynthesis and full-field digital mammography

ObjectivesTo explore lesion-specific exposure parameters on digital breast tomosynthesis (DBT) and full-field digital mammography (FFDM) and evaluate their efficiency for breast cancer diagnosis. MethodsWe reviewed the DBT and FFDM images of 492 women with unilateral breast lesions and compared the findings with a healthy group comprising the contralateral breasts of all patients. The tube voltage (kVp), tube load (mAs), compressed breast thickness (CBT), and mean glandular dose (MGD) for each image were retrieved from the DICOM metadata, and the difference in MGD between DBT and FFDM images (△MGD) was calculated. Three models were developed to discriminate breast cancer: logistic model, comprising traditional risk factors alone; FFDM model, comprising exposure parameters from FFDM images alone, and a hybrid model, comprising traditional risk factors and exposure parameters from DBT and FFDM images. Model performance was assessed in an independent dataset of 189 women by determining the area under the receiver operating characteristic curve (AUC). ResultsThe malignant group showed greater kVp, mAs, CBT, and MGD values than the benign and healthy groups on both FFDM and DBT. Breast thickness, density, and lesion size were independently associated with △MGD. The hybrid model showed a significantly higher AUC value (0.78 ± 0.04) than the logistic (0.68 ± 0.05) and FFDM models (0.56 ± 0.05). ConclusionsThe breast lesion type has an influence on exposure parameters when using automatic exposure control. Our results provide evidence for lesion-specific exposure parameters, which can be helpful in distinguishing benign lesions from malignancies.

Comparative analysis of proficiencies of various textures and geometric features in breast mass classification using k-nearest neighbor

This paper introduces a comparative analysis of the proficiencies of various textures and geometric features in the diagnosis of breast masses on mammograms. An improved machine learning-based framework was developed for this study. The proposed system was tested using 106 full field digital mammography images from the INbreast dataset, containing a total of 115 breast mass lesions. The proficiencies of individual and various combinations of computed textures and geometric features were investigated by evaluating their contributions towards attaining higher classification accuracies. Four state-of-the-art filter-based feature selection algorithms (Relief-F, Pearson correlation coefficient, neighborhood component analysis, and term variance) were employed to select the top 20 most discriminative features. The Relief-F algorithm outperformed other feature selection algorithms in terms of classification results by reporting 85.2% accuracy, 82.0% sensitivity, and 88.0% specificity. A set of nine most discriminative features were then selected, out of the earlier mentioned 20 features obtained using Relief-F, as a result of further simulations. The classification performances of six state-of-the-art machine learning classifiers, namely k-nearest neighbor (k-NN), support vector machine, decision tree, Naive Bayes, random forest, and ensemble tree, were investigated, and the obtained results revealed that the best classification results (accuracy = 90.4%, sensitivity = 92.0%, specificity = 88.0%) were obtained for the k-NN classifier with the number of neighbors having k = 5 and squared inverse distance weight. The key findings include the identification of the nine most discriminative features, that is, FD26 (Fourier Descriptor), Euler number, solidity, mean, FD14, FD13, periodicity, skewness, and contrast out of a pool of 125 texture and geometric features. The proposed results revealed that the selected nine features can be used for the classification of breast masses in mammograms.

Diagnostic performance of digital breast tomosynthesis for predicting response to neoadjuvant systemic therapy in breast cancer patients: A comparison with magnetic resonance imaging, ultrasound, and full-field digital mammography.

BackgroundThe goals of neoadjuvant systemic therapy (NST) are to reduce tumor volume and to provide a prognostic indicator in assessing treatment response. Digital breast tomosynthesis (DBT) was developed and has increased interest in clinical settings due to its higher sensitivity for breast cancer detection compared to full-field digital mammography (FFDM).PurposeTo evaluate the accuracy of DBT in assessing response to NST compared to FFDM, ultrasound (US), and magnetic resonance imaging (MRI) in breast cancer patients.Material and MethodsIn this retrospective study, 95 stages II–III breast cancer patients undergoing NST and subsequent surgeries were enrolled. After NST, the longest diameter of residual tumor measured by DBT, FFDM, US, and MRI was compared with pathology. Agreements and correlations of tumor size were assessed, and the diagnostic performance for predicting pathologic complete response (pCR) was evaluated.ResultsMean residual tumor size after NST was 19.9 mm for DBT, 18.7 mm for FFDM, 16.0 mm for US, and 18.4 mm for MRI, compared with 17.9 mm on pathology. DBT and MRI correlated better with pathology than that of FFDM and US. The ICC values were 0.85, 0.87, 0.74, and 0.77, respectively. Twenty-five patients (26.3%) achieved pCR after NST. For predicting pCR, area under the receiver operating characteristic (ROC) curve for DBT, FFDM, US, and MRI were 0.79, 0.66, 0.68, and 0.77, respectively.ConclusionDBT has good correlation with histopathology for measuring residual tumor size after NST. DBT was comparable to MRI in assessing tumor response after completion of NST.

OD123 - Small-size details detection performance of digital breast tomosynthesis, synthetic 2D and conventional full-field digital mammography images for different mammography systems: a multicenter study

OD123 - Small-size details detection performance of digital breast tomosynthesis, synthetic 2D and conventional full-field digital mammography images for different mammography systems: a multicenter study

The Effect of Breast Size and Density in Turkish Women on Radiation Dose in Full-Field Digital Mammography.

The purpose of this study was to look into the relationship between breast size and mammographic breast density in women and breast radiation dose on full-field digital mammography (FFDM), as well as the factors that influence radiation dose. The study included a total of 2,060 FFDM images from 515 consecutive participants. The participants were divided into two groups: those exposed to high doses (>3 mGy) and those exposed to low doses (<3 mGy). Moreover, the researchers analyzed the relationship between mean glandular dose (MGD) of the breast and patient age, compressed breast thickness, compression force, mammographic breast composition, and mammographic breast size. The mean mammographic breast volume was 936.2 ± 425.2 (114.5-3,018) mL, and the mean compressed breast tissue thickness was 56.75 ± 10.44 mm. Moreover, the mean MGD in the high-dose group was 3.51 ± 0.48 mGy and 1.92 ± 0.56 mGy in the low-dose group. The high-dose group had greater breast thickness, diameters, volume, compression pressure, and surgical rate. However, the high-dose group was younger and had less dense breasts. In multivariate logistic regression analysis, the most important predictors of dose determination were breast thickness [odds ratio (OR): 1.178, 95% confidence interval (CI): 1.156-1.200, p<0.001], history of previous surgery (OR: 2.210, 95% CI: 1.417-3.447, p<0.001), compression force (OR: 1.008, 95% CI: 1.004-1.013, p<0.001), and breast density (OR: 1.873, 95% CI: 1.359-2.580, p<0.001). Women with larger breast volumes are subjected to higher doses of radiation. Therefore, breast-screening programs can be individualized to young women with larger breast volumes and women who have had breast-conserving surgery.

Special issue on future hybrid artificial intelligence and machine learning for smart expert systems

Artificial Intelligence (AI) has grown extensively in recent times and already encroaches into our lives, for example, with smart cars and smart homes. AI current and future trends in areas such as smart cars, face recognition, robotic surgeries, or automatic disease classification, focus on hybridization to improve the performance of smart expert systems. Despite the huge growth of various AI and machine learning techniques, such as decision trees, data mining, or deep convergence learning, there are still many challenges and threats which limit the performance of these smart expert systems. This special issue focuses on new directions in hybrid AI and machine learning methodologies to address current research problems. It focuses on bringing together leading-edge articles on hybrid AI techniques for smart expert systems for various applications. From about 25 articles submitted, 4 papers were selected based on the review process. Each paper was reviewed by at least two independent reviewers and undertook at least one round of review. Their contributions are summarized below. Raghavendra et al. (2019) has proposed a research work on accurate and early detection of breast cancer using effective imaging modalities with medical image analysis. The research results have shown that two-dimensional synthesized mammogram imaging and conventional full-field digital mammography (FFDM) are two important imaging modalities which can be used for screening breast cancer. The developed method achieved an average performance of 92.9% accuracy using a probabilistic neural network classifier for FFDM images with tenfold cross validation. The obtained results can be used as a distinct system in rural hospitals. Gupta et al. (2020) has done their research studies on to detect infected leaves using machine learning techniques. An improved artificial plant optimization (IAPO) algorithm using machine learning has been introduced that identifies the plant diseases and categorize the leaves into healthy and infected on a private data set of 236 images. The degree of infection is eventually computed, and the leaves with infection greater than a certain calculated threshold are classified as infected leaves. The results show that IAPO can be used for the classification of infected and healthy leaves and this algorithm can be generalized to solve problems in other domains as well. Hammad et al. (2020) has found a new authentication approach based on biometric modalities such as heartbeat pattern obtained from electrocardiogram (ECG) signals. This work shows two novel deep neural network (DNN) models (convolutional neural network (CNN) and ResNet-Attention) using ECG signals for human authentication and the proposed CNN algorithm achieved an accuracy of 98.59% and 99.72% using PTB and CYBHi, respectively. The results obtained by the developed model proved that the performance is better than existing algorithms and can be used in real-time authentication systems after the validation with more diverse ECG data. Teles et al. (2020) has devised a decision support system on credit operations which can provide different estimations of expected recovery based on the same data sets. This research work classifies credit by the formulation of a rule that describes the values of a categorical variable according to some specified definition. Results reveal that while linear regression can be used for the prediction of a continuous variable in the credit process using collateral, it is not suitable for categorical data. Out of linear and logistic regressions, linear regression is more applicable in determining groups of variables that can significantly predict an outcome. We would like to take this opportunity to thank Dr. Jon G. Hall, Editor-in-Chief of Expert Systems, The Journal of Knowledge Engineering, for the privilege to edit this special issue. Gustavo Ramírez González is a professor at department of telematics engineering in Universidad del Cauca, Colombia. His research interests include image processing, secure communication, and machine learning. He has published several research papers in reputed journals and served as a Guest Editor for several Special Issues at many journals.

Abstract 2600: Deep-LIBRA: An artificial intelligence approach for fully-automated assessment of breast density in digital mammography

Abstract Accurate estimation of mammographic breast density could aid in augmenting breast cancer risk assessment for women undergoing breast screening with full-field digital mammography (FFDM). Breast density can be estimated from FFDM and is most commonly assessed in the clinic by visual grading into one of the four categories defined by the American College of Radiology BI-RADS. However, BI-RADS density assessment is highly subjective and does not provide a quantitative, continuous measure of percent density (PD), which would allow for more refined risk stratification and assessment of density changes. Here, we introduce Deep-LIBRA, an artificial intelligence (AI) tool for fully-automated assessment of breast PD from FFDM images. Two key modules form the core of Deep-LIBRA: 1) an implementation of a modified U-Net architecture for breast segmentation and 2) a radiomic machine learning module that performs PD estimation within the segmented breast region. To develop and validate Deep-LIBRA, raw (i.e., "For Processing") FFDM images (Selenia Dimensions, Hologic Inc.) acquired at two breast cancer screening practices were retrospectively analyzed. For the breast segmentation module, we used a total of 12,100 FFDM studies from 2,200 individual women and a 90%-10% split-sample training-validation approach, using the Dice coefficient to evaluate the accuracy of Deep-LIBRA versus ground-truth manual breast segmentation. For the PD estimation module we used a total of 3,304 FFDM images from 1,652 individual women; manual PD scores obtained with the widely used Cumulus software were used as the "gold standard" in a three-fold cross-validation setting to assess the accuracy of Deep-LIBRA in PD estimation. PD estimates from Deep-LIBRA were also compared with breast density estimates from the commercially available Volpara software. Breast segmentation had a Dice coefficient of 95.31% when compared to ground-truth manual breast segmentation in the validation set. Deep-LIBRA average differences from ground-truth PD scores in the three cross-validation folds were 4.91%, 4.65%, and 4.22%, while Volpara had corresponding average differences of 6.20%, 6.01%, and 5.94%. Deep-LIBRA PD scores were also significantly different from Volpara PD (t-test p-value &amp;lt; 0.001 in all three folds). Preliminary evaluation results show that Deep-LIBRA is a promising AI approach for accurately assessing PD from FFDM images. Citation Format: Omid Haji Maghsoudi, Scott Christopher, Aimilia Gastounioti, Lauren Pantalone, Fang-Fang Wu, Eric A. Cohen, Winham Stacey, Emily F. Conant, Celine Vachon, Despina Kontos. Deep-LIBRA: An artificial intelligence approach for fully-automated assessment of breast density in digital mammography [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 2600.

Comparison of two-dimensional synthesized mammograms versus original digital mammograms: a quantitative assessment.

This study objectively evaluates the similarity between standard full-field digital mammograms and two-dimensional synthesized digital mammograms (2DSM) in a cohort of women undergoing mammography. Under an institutional review board-approved data collection protocol, we retrospectively analyzed 407 women with digital breast tomosynthesis (DBT) and full-field digital mammography (FFDM) examinations performed from September 1, 2014, through February 29, 2016. Both FFDM and 2DSM images were used for the analysis, and 3216 available craniocaudal (CC) and mediolateral oblique (MLO) view mammograms altogether were included in the dataset. We analyzed the mammograms using a fully automated algorithm that computes 152 structural similarity, texture, and mammographic density-based features. We trained and developed two different global mammographic image feature analysis-based breast cancer detection schemes for 2DSM and FFDM images, respectively. The highest structural similarity features were obtained on the coarse Weber Local Descriptor differential excitation texture feature component computed on the CC view images (0.8770) and MLO view images (0.8889). Although the coarse structures are similar, the global mammographic image feature-based cancer detection scheme trained on 2DSM images outperformed the corresponding scheme trained on FFDM images, with area under a receiver operating characteristic curve (AUC) = 0.878 ± 0.034 and 0.756 ± 0.052, respectively. Consequently, further investigation is required to examine whether DBT can replace FFDM as a standalone technique, especially for the development of automated objective-based methods.

Association of the Differences in Average Glandular Dose with Breast Cancer Risk

Objectives. To compare the differences in normalized average glandular dose (NAGD) between the breasts of healthy subjects and those of cancer patients and to determine if the NAGD difference is associated with breast cancer risk and improves breast cancer classification. Materials and Methods. Craniocaudal view and mediolateral view full-field digital mammography (FFDM) images were obtained from 1682 healthy subjects whose breasts were categorized as Breast Imaging-Reporting and Data System (BI-RADS) I or II and from 811 biopsy-confirmed unilateral breast cancer patients whose breasts on the contralateral side were category I or II. Both populations were randomized into training and test sets. Multivariate logistic regression analysis was used to build the breast cancer risk assessment model, and the area under the receiver operating characteristic curve ( A z ) was used to evaluate the model. Twenty-two breast cancer patients who were originally categorized as BI-RADS I or II for both breasts, but were diagnosed with unilateral biopsy-confirmed breast cancer subsequently, were included to validate the model. Results. The NAGD differences in both FFDM images between tumor-bearing breasts and the healthy breasts of patients were significantly higher than those in healthy subjects ( P &lt; 0.001 ). The model with NAGD differences had a higher A z value than the model without NAGD differences. While there was no NAGD differences between originally healthy breasts of breast cancer patients, significant NAGD differences between now tumor-bearing breasts and the then previously healthy breasts were found in both FFDM images. Conclusions. NAGD differences between both breasts can be included in the breast cancer risk assessment model to evaluate breast cancer risk.

Comparison between Recorded and Measured Radiation Doses in Diagnostic Full-field Digital Mammography: A Phantom Study.

Objectives:There are concerns regarding the difference between directly recorded and measured entrance skin dose (ESD) and average glandular dose (AGD) in full-field digital mammography (FFDM). The objective of the study was to evaluate the effect of different exposure parameters on ESD and AGD recorded directly and measured from an FFDM unit using a phantom.Material and Methods:The ESD and AGD of 27 FFDM (craniocaudal [CC] projection) images of tissue-equivalent phantoms were acquired using a general electric (GE Senographe Essential) FFDM unit. The phantoms were used to simulate three different breast thicknesses and compositions. Tube potential, tube load, and target/ filter combinations also were recorded directly from the FFDM unit.Results:The mean differences between the directly recorded and measured ESD and AGD were 0.23 and 0.080, respectively. The 95% confidence intervals for ESD and AGD were 0.1–0.36 and 0.04–0.10, respectively. Results of paired t-test showed statistically significant difference between the directly recorded and measured ESD (P = .001) and AGD (P < .001). A positive and significant correlation was noted between the directly recorded and measured ESD (r = 0.85, P < .001) and AGD (r = 0.91, P < .001).Conclusion:This observation confirms that we can use the directly recorded doses obtained from an FFDM for quality control program.

Additive value of exposure parameters for breast cancer diagnosis in digital mammography.

To investigate the effect of different breast lesions on exposure parameters in digital mammography and to determine whether the exposure parameters can additively improve diagnostic efficiency. Craniocaudal view and mediolateral view full-field digital mammography images from 982 women with unilateral lesions (341 with malignant lesions, 189 with benign lesions, and 452 healthy women) obtained at Nanfang Hospital were reviewed. Differences in exposure parameters (tube voltage and load, breast thickness (BT), and average glandular dose (AGD)) between breasts were calculated. The relationships between parameter differences and lesion size were explored. A logistic regression model was used based on the AGD and BT differences, and the area under the receiver operating characteristic curve (AUC) was used to assess the performance of these parameters in differentiating malignant from benign and healthy subjects. Independently, data from 129 women (82 with malignant and 47 with benign lesions) treated at Sun Yat-sen Memorial Hospital were collected to validate the model. Differences in tube voltage and load, BT, and AGD between breasts were significantly greater in the malignant subjects than benign (p < 0.05) and healthy subjects (p < 0.05). The AUCs for the comparisons of malignant vs. healthy subjects, malignant vs. benign subjects, and benign vs. healthy subjects were 0.77 ± 0.02, 0.72 ± 0.02, and 0.57 ± 0.02, respectively. The model combining the exposure parameters with the BI-RADS category resulted in a higher AUC (0.910 ± 0.03) compared with physician diagnosis alone (0.820 ± 0.04) for differentiating between malignant and benign lesions. Exposure parameters additively improved diagnostic accuracy for breast cancer and yielded more reliable results. • Differences in kVp, mAs, BT, and AGD between breasts were significantly greater in the malignant subjects than benign and healthy subjects. • The model combining exposure parameters with the BI-RADS category resulted in a higher AUC compared with the physician's diagnosis for differentiating between malignant and benign lesions. • Exposure parameters additively improved diagnostic accuracy for breast cancer.

Clinical performance of contrast-enhanced spectral mammography in pre-surgical evaluation of breast malignant lesions in dense breasts: a single center study

PurposeTo compare the efficacy of contrast-enhanced spectral mammography, with ultrasound, full field digital mammography and magnetic resonance imaging in detection and size estimation of histologically proven breast tumors.MethodsThis open-label, single center, prospective study, included 160 dense breast women with at least one suspicious mammary lesion evaluated by ultrasound, full field digital mammography and magnetic resonance imaging in whom a mammary tumor was histologically proven after surgery performed at the European Institute of Oncology between January 2013 and December 2015. Following the complete diagnostic procedure, the patients were further investigated by contrast-enhanced spectral mammography prior to surgery.ResultsOverall, the detection rate of malignant breast lesions (in situ and invasive) was 93.8% (165/176) for contrast-enhanced spectral mammography, 94.4% (168/178) for ultrasound, 85.5 (147/172) for full field digital mammography and 97.7% (173/177) for magnetic resonance imaging. Radiological measurements were concordant with the post-surgical pathological measurements of the invasive tumor (i.e., within 5 mm) in: 64.6% for contrast-enhanced spectral mammography, 62.0% for ultrasound, 45.2% for full field digital mammography (p < 0.0001) and 69.9% for magnetic resonance imaging (p = 0.28); underestimated in: 17.4% for contrast-enhanced spectral mammography, 19.6% for ultrasound, 24.2% for full field digital mammography (p = 0.03) and 6.7% for magnetic resonance imaging (p = 0.0005); and overestimated in: 16.2% for contrast-enhanced spectral mammography, 16.6% for ultrasound, 16.6% for full field digital mammography and 22.7% for magnetic resonance imaging (p = 0.02).ConclusionsOur data suggest that contrast-enhanced spectral mammography improves on full field digital mammography and is comparable to ultrasound and magnetic resonance imaging in terms of detection sensitivity and size estimation of malignant lesions in dense breasts.

Evaluation of an AI system for breast cancer screening in mammograms of young women.

e14068 Background: The interpretation of the mammography is challenging, especially in young women, who have dense breasts. Artificial intelligence (AI) promises to improve breast cancer detection; however these systems should be tested on different datasets. Our aim is to evaluate the performance of a publicly available deep convolutional neural network, developed by Wu et al. (IEEE Trans. Med. Imaging, 2019), applied to mammograms of young women. Methods: The test dataset consisted of mammograms obtained on a single occasion from 135 young women (up to 40 years old) on a Siemens mammography system. Each exam consisted of 4 full-field digital mammography images and had two labels (left malignant and right malignant). Mammograms were analyzed by a single mammography trained radiologist, using BI-RADS reporting tool. Among 270 labels, 170 were malignant and 100 were non-malignant. We used the program developed by Wu et al. that, according to the authors, presents AUC of 0.895 for the general population. As a preliminary test, we ran this program in a publicly available dataset named INbreast and obtained AUC of 0.8708, very close to the result reported by the authors. Results: We applied the program to our dataset of young women and obtained AUC of 0.876. We computed its standard error, obtaining 0.0290. At equal error rate point of the ROC curve, specificity and sensitivity are both 0.774. With this result we conclude that, at least for our dataset, cancer detection in young women is not substantially more difficult than in general population for an AI system. We fine-tuned the weights of the original network to the population of young women using transfer learning and obtained a slight improvement in AUC: 0.9018±0.0528, where the mean and the standard error were obtained using 5-fold cross validation. As the improvement was small and the standard errors are large, we would have to test on a larger test set to ensure that the observed improvement is real. Conclusions: We conclude, based on the experimental data, that there is no substantial degradation in accuracy when a mammogram screening program for general population is used for young women. We also conclude that it seems to be possible to obtain a slight improvement in accuracy by fine-tuning the network for the population of young women.