Subtraction of Temporally Sequential Digital Mammograms: Enhancing the Detection and Classification of Malignant Masses in Breast Imaging

Supplemental material is available for this article. Keywords: Mammography, Breast, Machine Learning © RSNA, 2023.

What effect does mammographic breast density have on lesion detection in digital mammography?

Contrast-enhanced spectral mammography (CESM) is a recently developed breast imaging technique. CESM relies on dual-energy acquisition following contrast agent injection to improve mammography sensitivity. CESM is comparable to contrast-enhanced MRI in terms of sensitivity, at a fraction of the cost. However, since lesion variability is large, even with the improved visibility provided by CESM, differentiation between benign and malignant enhancement is not accurate and a biopsy is usually performed for final assessment. Breast biopsies can be stressful to the patient and are expensive to healthcare systems. Moreover, as the biopsies results are most of the time benign, a specificity improvement in the radiologist diagnosis is required. This work presents a deep learning-based decision support system, which aims at improving the specificity of breast cancer diagnosis by CESM without affecting sensitivity. We compare two analysis approaches, fine-tuning a pretrained network and fully training a convolutional neural network, for classification of CESM breast mass as benign or malignant. Breast Imaging Reporting and Data Systems (BIRADS) is a radiological lexicon, used with breast images, to categorize lesions. We improve each classification network by incorporating BIRADS textual features as an additional input to the network. We evaluate two ways of BIRADS fusion as network input: feature fusion and decision fusion. This leads to multimodal network architectures. At classification, we also exploit information from apparently normal breast tissue in the CESM of the considered patient, leading to a patient-specific classification. We evaluate performance using fivefold cross-validation, on 129 randomly selected breast lesions annotated by an experienced radiologist. Each annotation includes a contour of the mass in the image, biopsy-proven label of benign or malignant lesion and BIRADS descriptors. At 100% sensitivity, specificity of 66% was achieved using a multimodal network, which combines inputs at feature level and patient-specific classification. The presented multimodal network may significantly reduce benign biopsies, without compromising sensitivity.

Mammography and Breast Localization for the Interventionalist

Mammography remains the most prevalent imaging tool for early breast cancer screening. The language used to describe abnormalities in mammographic reports is based on the Breast Imaging Reporting and Data System (BI-RADS). Assigning a correct BI-RADS category to each examined mammogram is a strenuous and challenging task for even experts. This paper proposes a new and effective computer-aided diagnosis (CAD) system to classify mammographic masses into four assessment categories in BI-RADS. The mass regions are first enhanced by means of histogram equalization and then semiautomatically segmented based on the region growing technique. A total of 130 handcrafted BI-RADS features are then extracted from the shape, margin, and density of each mass, together with the mass size and the patient's age, as mentioned in BI-RADS mammography. Then, a modified feature selection method based on the genetic algorithm (GA) is proposed to select the most clinically significant BI-RADS features. Finally, a back-propagation neural network (BPN) is employed for classification, and its accuracy is used as the fitness in GA. A set of 500 mammogram images from the digital database for screening mammography (DDSM) is used for evaluation. Our system achieves classification accuracy, positive predictive value, negative predictive value, and Matthews correlation coefficient of 84.5%, 84.4%, 94.8%, and 79.3%, respectively. To our best knowledge, this is the best current result for BI-RADS classification of breast masses in mammography, which makes the proposed system promising to support radiologists for deciding proper patient management based on the automatically assigned BI-RADS categories.

The Breast Imaging Reporting and Data System (BI-RADS) was initiated in the United States in the late 1980s to provide a standardised description of imaging features of breast lesions and to relate them to the underlying pathology and risk of malignancy. Features that were helpful in predicting benign or malignant pathology were chosen, originally for mammography and subsequently for ultrasound [1]. BI-RADS has been used as an education tool and to improve quality across the United States and it is now mandatory to include BI-RADS information in mammography reports. Given the diverse nature of practice in the United States with a spectrum of imaging availability and expertise including large dedicated breast centres and small practices where individual clinicians report relatively small numbers of cases, BI-RADS may have been the essential tool in achieving conformity. There are now 6 categories for each feature: 0, incomplete; 1, negative; 2, benign finding(s); 3, probably benign; 4, suspicious abnormality; 5, highly suggestive of malignancy; and 6, known biopsy-proven malignancy. The fourth category is subdivided into a, b and c. The risk of malignancy for each category is well established [2] and studies evaluating observer variability show fair concordance for features describing categories 1, 2, 3 and 5, although there is interobserver variability for the BI-RADS 4 subcategories [3]. BI-RADS is proven to be of value in the standardisation of reporting and is intended to provide clear guidance on further management; however, it has not been widely adopted in the UK. There are philosophical differences between accepted protocols for investigating breast lesions in the United States and the UK that limit the applications of the BI-RADS in the UK. In particular, lesions at a low risk of malignancy (BI-RADS 3 and 4a) [4] undergo short-term follow-up in the United States but, if palpable, would be subject to biopsy in the UK [5]. The reasons for this are unclear, although when BI-RADS was introduced in the United States some facilities only had mammographic services and limited access to image-guided biopsy (especially core biopsy), which is now more readily available. This may account for the different approach to breast diagnosis together with other considerations, such as reimbursement policies and avoidance of litigation. A 5-point scoring system for mammography, ultrasound and cytology was described in the UK in 1998 [6]. This differed from the BI-RADS as it included pathology results and supported the use of triple assessment by clinical examination, imaging and the needle test to improve sensitivity and specificity in the evaluation of breast lesions. Use of this system became fairly widespread, reinforced by the standards imposed by the National Health Service breast screening programme. The UK 5-point breast imaging scoring system has recently been formalised by Maxwell et al [7] on behalf of the Royal College of Radiologists (RCR) Breast Group. It promotes standardisation of reporting and is easily understood by all members of the multidisciplinary team. However, unlike BI-RADS, at present it does not give the likelihood of malignancy in each category. Therefore the article by Taylor et al [8] in this issue of the British Journal of Radiology is timely. The article describes mammography and ultrasound data from 23 741 assessment episodes to quantify the likelihood of cancer with each of the 5 points on the UK scoring system and compares them with points of equivalent cancer risk on the BI-RADS score. This will provide a benchmark for centres to audit the unit's performance. The positive impact that BI-RADS has had on raising standards and unifying research data cannot be underestimated. It has been extensively researched and validated and it is time now for the RCR Breast Group classification to undergo this rigorous testing. The UK 5-point breast imaging scoring system should be used for communication across the multidisciplinary team with analogous systems for clinical examination, MRI, cytology and histopathology reporting. Any discordance between the clinical scoring systems must be resolved before the management of each case is concluded. The careful development and application of a universally accepted scoring system for breast lesions has played an important role in the diagnosis of breast disease, especially in improving the sensitivity and specificity of diagnosticians and optimising the diagnosis of significant disease while minimising harm by overinvestigation. As with BI-RADS, the UK 5-point breast imaging scoring system will continue to evolve and has the capacity to incorporate emerging techniques and modalities. It provides an example of how the imaging community can collaborate to standardise practice and ultimately improve patient care.

This study evaluates the impact of integrating elastography into the Breast Imaging Reporting and Data System (BIRADS) categorization on breast cancer diagnostics in an African population. It explores the association and agreement between traditional BIRADS and those modified by elastography, as well as between quantitative and qualitative elastography methods. A total of 200 participants who underwent breast imaging as part of their diagnostic evaluation for breast lesions were included in the study. Participant characteristics, including age distribution and indicators for breast cancer diagnoses, were analyzed. Brightness mode (B-mode) findings without elastography were assessed using the BIRADS classification. Elastography was integrated into the BIRADS categorization to evaluate its impact on breast cancer diagnostics.The association and agreement between BIRADS with and without elastography were analyzed. Participants predominantly aged 40-49 showed significant staging differences with the integration of elastography. Traditional B-mode staging identified 29 (49%) of participants in BIRADS stage IV and 14 (23%) in stage V, whereas elastography adjusted these figures significantly, enhancing diagnostic refinement. There was a fair agreement between BIRADS with and without elastography (kappa = 0.322), while a substantial agreement was found between quantitative and qualitative elastography (kappa = 0.674). The results of the study provide evidence that the integration of elastography into BIRADS categorization can significantly improve the accuracy of breast cancer diagnosis in African women. Elastography enhanced lesion characterization, supporting more personalized and precise clinical management. Continued research is needed to fully integrate elastography into routine diagnostic workflows and understand its broader clinical implications in Africa.

Purpose: The Breast Imaging Reporting and Data System (BI-RADS) lexicon used in reporting breast imaging has several categories with specific positive predictive values for breast cancer. Among those, BI-RADS 4 is associated with a wider range of risk for breast cancer, which makes the decision for biopsy difficult. The study aim was to determine the malignancy rate and clinical outcomes of BI-RADS 4 lesions in Hospital Universiti Sains Malaysia (HUSM) for a period of five years.Methods: This was a retrospective study of patients diagnosed by mammographic or ultrasonographic findings with BI-RADS 4 breast lesions in HUSM, Kelantan from July 2015 to June 2020. Data were collected from the medical records and an electronic database. Patients with BI-RADS 4 lesions who underwent biopsy and had a known tissue diagnosis were included in this study. The data was used to calculate the malignancy rate and associated positive predictive factors for breast cancer associated with BI-RADS 4 lesions.Results: From the mammography and ultrasonography performed at HUSM from July 2015 to June 2020, a total of 256 lesions were categorized as BI-RADS 4. However, only 198 BI-RADS 4 lesions underwent biopsy and were included in the study. Of these 198 lesions, 26.8% were malignant on histopathological examination of the biopsy samples. Simple logistic regression analysis showed that age, diabetes mellitus, hypertension, number of parity, and certain mammogram findings were significantly associated with breast cancer. Invasive breast cancer was the most common type. Fibrocystic disease was the most common benign pathology, followed by fibroadenoma.Conclusion: The malignancy rate of BI-RADS 4 lesions in HUSM was similar to previously reported rates. A thorough evaluation of positive predictive factors and careful selection of patients for biopsy in BI-RADS 4 lesions will minimize unwanted biopsies and associated patient anxiety, in addition to reducing the health care burden.

Mammographic Breast Density Assessment Using Automated Volumetric Software and Breast Imaging Reporting and Data System (BIRADS) Categorization by Expert Radiologists

The diagnostic categorization of a lesion as Breast Imaging Reporting and Data System (BI-RADS) category 3 or 4A determines whether a patient needs a biopsy; however, interobserver variability exists among radiologists in mammographic interpretation. This variability may lead to underdiagnoses of BI-RADS 3 lesions and unnecessary biopsies of benign BI-RADS 4A lesions. Therefore, we assessed the diagnostic value of a mammography-based deep learning (DL) model for differentiating BI-RADS 3 and 4A lesions and its impact on radiologists' decision-making. This retrospective multicenter study analyzed 846 mammographically detected breast lesions (BI-RADS 3 and 4A) from 824 patients at Shenzhen People's Hospital and Shenzhen Luohu People's Hospital between January and December 2020. Six breast imaging specialists (three junior and three senior) independently reviewed all mammograms with and without DL model assistance. The follow-up or biopsy results were used as the reference standard. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves and the area under the curve (AUC), with the DeLong test being used to compare AUCs. The DL model yielded an AUC of 0.74 for distinguishing BI-RADS 3 and 4A lesions, outperforming junior radiologists' standalone performance (AUC =0.57, AUC =0.55, and AUC =0.58) but remaining inferior to senior radiologists (AUC =0.78, AUC =0.77, and AUC =0.76). With DL model assistance, all six radiologists had higher AUCs for diagnosing BI-RADS 3 and 4A lesions as compared to their unassisted performance. Importantly, DL integration significantly increased junior radiologists' AUCs to 0.74-0.77 (P<0.001), whereas the increase in the AUCs of the senior radiologists (to 0.79, 0.78, and 0.78) was not significant (P>0.05). The mammography-based DL model significantly improved the diagnostic performance of junior radiologists for BI-RADS 3 and 4A lesions, effectively reducing missed diagnoses and unnecessary biopsies.

To determine whether training in the Breast Imaging Reporting and Data System (BI-RADS) improves observer performance and agreement with the consensus of experienced breast imagers with regard to mammographic feature analysis and final assessment. A test set of mammograms was developed, with 54 proven lesions consisting of 28 masses (nine [32%] malignancies) and 26 microcalcifications (10 [38%] malignancies). Three experienced breast imagers reviewed cases independently and by means of consensus. Twenty-three practicing mammogram-interpreting physicians reviewed mammograms before and after a day's lectures on BI-RADS. Observer performance before and after training was measured by means of agreement (kappa) with consensus description and assessments, rate of biopsy of malignant and benign lesions, and areas under receiver operating characteristic (ROC) curves. Performance was also measured for 11 participants 2-3 months after training. Improved agreement with consensus feature analysis was found for mass margins and/or asymmetries, with a pretraining generalized kappa value of 0.36 and a posttraining generalized kappa value of 0.41. Similar improvement was seen for description of calcification morphology (pretraining kappa value of 0.36 improving to 0.44 after training). No improvement was seen in describing calcification distribution. Final assessments were more consistent after training, with a pretraining kappa value of 0.31, as compared with 0.45 after training. The mean biopsy rate for malignant lesions improved from 73% (range, 53%-89%) before training to 88% (range, 74%-100%) after training, with minimal increase in mean biopsy rate of benign lesions (43% [range, 26%-60%] before to 51% [range, 31%-63%] after training), and no net change in area under the ROC curve, as compared with histopathologic findings. For the subset of participants with delayed follow-up, no significant decline in posttraining results was seen. BI-RADS training resulted in improved agreement with the consensus of experienced breast imagers for feature analysis and final assessment. It is important that trainees showed improved rates of recommending biopsy for malignant lesions. This effect was maintained over 2-3 months.

BackgroundIt has always been difficult and challenging to quantify the breast imaging reporting and data system (BI‐RADS) criteria into several categories. Automatic quantitation can assist clinicians in the early diagnosis and treatment eventually reducing the mortality rate. As a result, in the recent years, early BC diagnosis methods based on pathological breast images have been in high demand.MethodWe propose a computer‐aided diagnosis (CAD) system that combines the transfer learning approach with meta‐heuristic optimization, and machine learning to classify BI‐RADS breast masses categories within levels 3 and 4. Transfer learning technique ResNet‐18 is used for high‐level feature extraction. The clinically important features are then chosen using a modified feature selection technique based on the Hyper Learning Binary DragonFly Algorithm (M‐HLBDA). Finally, a Fine K‐nearest neighbour (KNN) is employed for classification.ResultA series of mammography breast mass images from the curated breast imaging subset of DDSM (CBIS‐DDSM) are evaluated in order to categorize within BI‐RADS levels 3 and 4. Experimental findings demonstrated M‐HLBDA capability to identify the optimal feature subset, which minimizes the number of selected features and maximizes the classification. Our system attained classification accuracy of 87.5%, Sensitivity of 88.8%, Specificity of 86.5%, and AUC of 0.82 using KNN on selected features using M‐HLBDA.ConclusionOur model can annotate and classify BI‐RADS levels 3 and 4 with better classification accuracy, and it may be used as an automated system to help radiologists.

Objective To evaluate the inter-observer variability of static breast sonogram final assessment among observers with different breast imaging experience, using the first edition of the Breast Imaging Reporting and Data System(BI-RADS) for ultrasound. Methods Thirty patients with 30 breast lesions were included who underwent beast lesions resection operation. A pathological diagnosis was available for all 30 lesions:16 (53%) malignant and 14 (47%) benign. Twelve radiologists independently reviewed two sonograms of each lesion, and assigned a final BI-RADS assessment category. Inter-observer variability was measured using kappa statistic. Positive predictive value(PPV) and negative predictive value (NPV) for final assessment were also calculated. Results As for the experienced observers,kappa values of categories 3,4 and 5 were 0.72,0.28 and 0.60,NPV of category 3 was 93% ,PPV of category 5 was 97% ,all of which decreased as the breast imaging experience reduced. PPVs of categories 4a,4b and 4c were 56 % ,88% and 69%, respectively. Conclusions Using BI-RADS final assessment, radiologists with sufficient breast imaging experience can provide accurate and consistent assessment for breast ultrasonography,but the agreement of diagnosis decreased as the breast imaging experience reduced. The clinical feasibility of 4a,4b and 4c subcategories is uncertain. Key words: Ultrasonography; Breast diseases

The Breast Imaging Reporting and Data System (BI-RADS) was introduced in 1993 to standardize the interpretation of mammograms. Though many studies have assessed the validity of the system, fewer have examined its reliability. Our objective is to identify predictors of reliability as measured by the kappa statistic. We identified studies conducted between 1993 and 2009 which reported kappa values for interpreting mammograms using any edition of BI-RADS. Bivariate and multivariate multilevel analyses were used to examine associations between potential predictors and kappa values. We identified ten eligible studies, which yielded 88 kappa values for the analysis. Potential predictors of kappa included: whether or not the study included negative cases, whether single- or two-view mammograms were used, whether or not mammograms were digital versus screen-film, whether or not the fourth edition of BI-RADS was utilized, the BI-RADS category being evaluated, whether or not readers were trained, whether or not there was an overlap in readers' professional activities, the number of cases in the study and the country in which the study was conducted. Our best multivariate model identified training, use of two-view mammograms and BI-RADS categories (masses, calcifications, and final assessments) as predictors of kappa. Training, use of two-view mammograms and focusing on mass description may be useful in increasing reliability in mammogram interpretation. Calcification and final assessment descriptors are areas for potential improvement. These findings are important for implementing policies in BI-RADS use before introducing the system in different settings and improving current implementations.

Purpose To assess the ability of optoacoustic (OA) ultrasonography (US) to help correctly downgrade benign masses classified as Breast Imaging Reporting and Data System (BI-RADS) 4a and 4b to BI-RADS 3 or 2. Materials and Methods OA/US technology uses laser light to detect relative amounts of oxygenated and deoxygenated hemoglobin in and around suspicious breast masses. In this prospective, multicenter study, results of 209 patients with 215 breast masses classified as BI-RADS 4a or 4b at US are reported. Patients were enrolled between 2015 and 2016. Masses were first evaluated with US with knowledge of previous clinical information and imaging results, and from this information a US imaging-based probability of malignancy (POM) and BI-RADS category were assigned to each mass. The same masses were then re-evaluated at OA/US. During the OA/US evaluation, radiologists scored five OA/US features, and then reassigned an OA/US-based POM and BI-RADS category for each mass. BI-RADS downgrade and upgrade percentages at OA/US were assessed by using a weighted sum of the five OA feature scores. Results At OA/US, 47.9% (57 of 119; 95% CI: 0.39, 0.57) of benign masses classified as BI-RADS 4a and 11.1% (three of 27; 95% CI: 0.03, 0.28) of masses classified as BI-RADS 4b were correctly downgraded to BI-RADS 3 or 2. Two of seven malignant masses classified as BI-RADS 4a at US were incorrectly downgraded, and one of 60 malignant masses classified as BI-RADS 4b at US was incorrectly downgraded for a total of 4.5% (three of 67; 95% CI: 0.01, 0.13) false-negative findings. Conclusion At OA/US, benign masses classified as BI-RADS 4a could be downgraded in BI-RADS category, which would potentially decrease biopsies negative for cancer and short-interval follow-up examinations, with the limitation that a few masses may be inappropriately downgraded.