Showing papers on "Digital mammography published in 2018"
TL;DR: In this article, the Faster R-CNN-based CAD system was proposed to detect malignant or benign lesions on a mammogram without any human intervention, which achieved 2nd place in the Digital Mammography DREAM Challenge with AUC = 0.95.
Abstract: In the last two decades, Computer Aided Detection (CAD) systems were developed to help radiologists analyse screening mammograms, however benefits of current CAD technologies appear to be contradictory, therefore they should be improved to be ultimately considered useful. Since 2012, deep convolutional neural networks (CNN) have been a tremendous success in image recognition, reaching human performance. These methods have greatly surpassed the traditional approaches, which are similar to currently used CAD solutions. Deep CNN-s have the potential to revolutionize medical image analysis. We propose a CAD system based on one of the most successful object detection frameworks, Faster R-CNN. The system detects and classifies malignant or benign lesions on a mammogram without any human intervention. The proposed method sets the state of the art classification performance on the public INbreast database, AUC = 0.95. The approach described here has achieved 2nd place in the Digital Mammography DREAM Challenge with AUC = 0.85. When used as a detector, the system reaches high sensitivity with very few false positive marks per image on the INbreast dataset. Source code, the trained model and an OsiriX plugin are published online at https://github.com/riblidezso/frcnn_cad .
421 citations
TL;DR: The aim of this work was to investigate a deep learning-based breast density classifier to consistently distinguish these two categories, aiming at providing a potential computerized tool to assist radiologists in assigning a BI-RADS category in current clinical workflow.
Abstract: Purpose
Mammographic breast density is an established risk marker for breast cancer and is visually assessed by radiologists in routine mammogram image reading, using four qualitative Breast Imaging and Reporting Data System (BI-RADS) breast density categories. It is particularly difficult for radiologists to consistently distinguish the two most common and most variably assigned BI-RADS categories, i.e., “scattered density” and “heterogeneously dense”. The aim of this work was to investigate a deep learning-based breast density classifier to consistently distinguish these two categories, aiming at providing a potential computerized tool to assist radiologists in assigning a BI-RADS category in current clinical workflow.
Methods
In this study, we constructed a convolutional neural network (CNN)-based model coupled with a large (i.e., 22,000 images) digital mammogram imaging dataset to evaluate the classification performance between the two aforementioned breast density categories. All images were collected from a cohort of 1,427 women who underwent standard digital mammography screening from 2005 to 2016 at our institution. The truths of the density categories were based on standard clinical assessment made by board-certified breast imaging radiologists. Effects of direct training from scratch solely using digital mammogram images and transfer learning of a pretrained model on a large nonmedical imaging dataset were evaluated for the specific task of breast density classification. In order to measure the classification performance, the CNN classifier was also tested on a refined version of the mammogram image dataset by removing some potentially inaccurately labeled images. Receiver operating characteristic (ROC) curves and the area under the curve (AUC) were used to measure the accuracy of the classifier.
Results
The AUC was 0.9421 when the CNN-model was trained from scratch on our own mammogram images, and the accuracy increased gradually along with an increased size of training samples. Using the pretrained model followed by a fine-tuning process with as few as 500 mammogram images led to an AUC of 0.9265. After removing the potentially inaccurately labeled images, AUC was increased to 0.9882 and 0.9857 for without and with the pretrained model, respectively, both significantly higher (P < 0.001) than when using the full imaging dataset.
Conclusions
Our study demonstrated high classification accuracies between two difficult to distinguish breast density categories that are routinely assessed by radiologists. We anticipate that our approach will help enhance current clinical assessment of breast density and better support consistent density notification to patients in breast cancer screening.
191 citations
02 Nov 2018
TL;DR: An improved lesion detection performance favoring tomosynthesis for all breast sizes and lesion types is found, suggesting that in silico imaging trials and imaging system computer simulation tools can in some cases be considered viable sources of evidence for the regulatory evaluation of imaging devices.
Abstract: Importance Expensive and lengthy clinical trials can delay regulatory evaluation of innovative technologies, affecting patient access to high-quality medical products. Simulation is increasingly being used in product development but rarely in regulatory applications. Objectives To conduct a computer-simulated imaging trial evaluating digital breast tomosynthesis (DBT) as a replacement for digital mammography (DM) and to compare the results with a comparative clinical trial. Design, Setting, and Participants The simulated Virtual Imaging Clinical Trial for Regulatory Evaluation (VICTRE) trial was designed to replicate a clinical trial that used human patients and radiologists. Images obtained with in silico versions of DM and DBT systems via fast Monte Carlo x-ray transport were interpreted by a computational reader detecting the presence of lesions. A total of 2986 synthetic image–based virtual patients with breast sizes and radiographic densities representative of a screening population and compressed thicknesses from 3.5 to 6 cm were generated using an analytic approach in which anatomical structures are randomly created within a predefined breast volume and compressed in the craniocaudal orientation. A positive cohort contained a digitally inserted microcalcification cluster or spiculated mass. Main Outcomes and Measures The trial end point was the difference in area under the receiver operating characteristic curve between modalities for lesion detection. The trial was sized for an SE of 0.01 in the change in area under the curve (AUC), half the uncertainty in the comparative clinical trial. Results In this trial, computational readers analyzed 31 055 DM and 27 960 DBT cases from 2986 virtual patients with the following Breast Imaging Reporting and Data System densities: 286 (9.6%) extremely dense, 1200 (40.2%) heterogeneously dense, 1200 (40.2%) scattered fibroglandular densities, and 300 (10.0%) almost entirely fat. The mean (SE) change in AUC was 0.0587 (0.0062) (P Conclusions and Relevance The results of the simulated VICTRE trial are consistent with the performance seen in the comparative trial. While further research is needed to assess the generalizability of these findings, in silico imaging trials represent a viable source of regulatory evidence for imaging devices.
140 citations
TL;DR: The main aim of the Malmö Breast Tomosynthesis Screening Trial (MBTST) was to investigate the accuracy of one-view digital breast tomosynthesis in population screening compared with standard two- view digital mammography.
Abstract: Summary Background Digital breast tomosynthesis is an advancement of the mammographic technique, with the potential to increase detection of lesions during breast cancer screening. The main aim of the Malmo Breast Tomosynthesis Screening Trial (MBTST) was to investigate the accuracy of one-view digital breast tomosynthesis in population screening compared with standard two-view digital mammography. Methods In this prospective, population-based screening study, of women aged 40–74 years invited to attend national breast cancer screening at Skane University Hospital, Malmo, Sweden, a random sample was asked to participate in the trial (every third woman who was invited to attend regular screening was invited to participate). Participants had to be able to speak English or Swedish and were excluded from the study if they were pregnant. Participants underwent screening with two-view digital mammography (ie, craniocaudal and mediolateral oblique views) followed by one-view digital breast tomosynthesis with reduced compression in the mediolateral oblique view (with a wide tomosynthesis angle of 50°) at one screening visit. Images were read with masked double reading and scoring by two separate reading groups, one for each method, made up of seven radiologists. Any cancer detected with a malignancy probability score of three or higher by any reader in either group was discussed in a consensus meeting of at least two readers, from which the decision of whether or not to recall the woman for further investigation was made. The primary outcome measures were sensitivity and specificity of breast cancer detection. Secondary outcome measures were screening performance measures of cancer detection, recall, and interval cancers (cancers clinically detected between screenings), and positive predictive value for screen recalls and negative predictive value of each method. Outcomes were analysed in the per-protocol population. Follow-up of the participants for at least 2 years allowed for identification of interval cancers. This trial is registered with ClinicalTrials.gov , number NCT01091545 . Findings Between Jan 27, 2010, and Feb 13, 2015, of 21 691 women invited, 14 851 (68%) agreed to participate. Three women withdrew consent during follow-up and were excluded from the analyses. 139 breast cancers were detected in 137 ( Interpretation Breast cancer screening by use of one-view digital breast tomosynthesis with a reduced compression force has higher sensitivity at a slightly lower specificity for breast cancer detection compared with two-view digital mammography and has the potential to reduce the radiation dose and screen-reading burden required by two-view digital breast tomosynthesis with two-view digital mammography. Funding The Swedish Cancer Society, The Swedish Research Council, The Breast Cancer Foundation, The Swedish Medical Society, The Crafoord Foundation, The Gunnar Nilsson Cancer Foundation, The Skane University Hospital Foundation, Governmental funding for clinical research, The South Swedish Health Care Region, The Malmo Hospital Cancer Foundation and The Cancer Foundation at the Department of Oncology, Skane University Hospital.
124 citations
TL;DR: Wang et al. as discussed by the authors proposed a Shallow-Deep Convolutional Neural Network (SD-CNN) where a shallow CNN is developed to derive "virtual" recombined images from low energy (LE) images, and a deep CNN is employed to extract novel features from LE, recombined or virtual" recombine images for ensemble models to classify the cases as benign vs cancer.
Abstract: Breast cancer is the second leading cause of cancer death among women worldwide. Nevertheless, it is also one of the most treatable malignances if detected early. Screening for breast cancer with digital mammography (DM) has been widely used. However it demonstrates limited sensitivity for women with dense breasts. An emerging technology in the field is contrast-enhanced digital mammography (CEDM), which includes a low energy (LE) image similar to DM, and a recombined image leveraging tumor neoangiogenesis similar to breast magnetic resonance imaging (MRI). CEDM has shown better diagnostic accuracy than DM. While promising, CEDM is not yet widely available across medical centers. In this research, we propose a Shallow-Deep Convolutional Neural Network (SD-CNN) where a shallow CNN is developed to derive "virtual" recombined images from LE images, and a deep CNN is employed to extract novel features from LE, recombined or "virtual" recombined images for ensemble models to classify the cases as benign vs. cancer. To evaluate the validity of our approach, we first develop a deep-CNN using 49 CEDM cases collected from Mayo Clinic to prove the contributions from recombined images for improved breast cancer diagnosis (0.86 in accuracy using LE imaging vs. 0.90 in accuracy using both LE and recombined imaging). We then develop a shallow-CNN using the same 49 CEDM cases to learn the nonlinear mapping from LE to recombined images. Next, we use 69 DM cases collected from the hospital located at Zhejiang University, China to generate "virtual" recombined images. Using DM alone provides 0.91 in accuracy, whereas SD-CNN improves the diagnostic accuracy to 0.95.
111 citations
TL;DR: DBT and SM screening increased the detection rate of histologically favorable tumors compared with that attained with DM screening, and rates of lymph node involvement and tumor subtypes did not differ.
Abstract: Our study showed the rate of screen-detected breast cancer was substantially higher in women who underwent digital breast tomosynthesis and synthetic mammography screening than in women who underwe...
99 citations
TL;DR: The presented work shows the feasibility of a DBN-based CAD system for use as in the field of breast cancer diagnosis, and demonstrates that the proposed DBN outperforms the conventional classifiers.
Abstract: Computer-aided diagnosis (CAD) offers assistance to radiologists in the interpretation of medical images. A CAD system learns the nature of different tissues and uses this information to diagnose abnormalities. In this paper, we propose a CAD system for breast cancer diagnosis via deep belief network (DBN) that automatically detects breast mass regions and recognizes them as normal, benign, or malignant. In this study, we utilize a standard digital database of mammography to evaluate our proposed DBN-based CAD system for breast cancer diagnosis. We utilize two techniques of ROI extraction: multiple mass regions of interest (ROIs) and whole mass ROIs. In the former technique, we randomly extract four ROIs with a size of 32 × 32 pixels from a detected mass. In the latter technique, the whole detected breast mass is utilized. A total of 347 statistical features are extracted for both techniques to train and test our proposed CAD system. For classification, we utilized linear discriminant analysis, quadratic discriminant analysis, and neural network classifiers as the conventional techniques. Finally, we employed DBN and compared the results. Our results demonstrate that the proposed DBN outperforms the conventional classifiers. The overall accuracies of a DBN are 92.86% and 90.84% for the two ROI techniques, respectively. The presented work shows the feasibility of a DBN-based CAD system for use as in the field of breast cancer diagnosis.
96 citations
TL;DR: CESM was significantly more sensitive than standard digital mammography for detecting breast cancer in this screening population and may be a valuable supplemental screening modality for women at intermediate risk who have dense breasts.
Abstract: OBJECTIVE. The purpose of this study was to compare the diagnostic performance of contrast-enhanced spectral mammography (CESM) and ultrasound with that of standard digital mammography for breast c...
95 citations
TL;DR: Reappraisals of Swedish mammography trials demonstrate that the design and statistical analysis of these trials were different from those of all trials on screening for cancers other than breast cancer, and compelling indications that these trials overestimated reductions in breast cancer mortality associated with screening are found.
94 citations
TL;DR: This study demonstrates that automatic deep learning CNN methods can identify nuanced mammographic imaging features to distinguish recalled-benign images from malignant and negative cases, which may lead to a computerized clinical toolkit to help reduce false recalls.
Abstract: Purpose: False positives in digital mammography screening lead to high recall rates, resulting in unnecessary medical procedures to patients and health care costs. This study aimed to investigate the revolutionary deep learning methods to distinguish recalled but benign mammography images from negative exams and those with malignancy. Experimental Design: Deep learning convolutional neural network (CNN) models were constructed to classify mammography images into malignant (breast cancer), negative (breast cancer free), and recalled-benign categories. A total of 14,860 images of 3,715 patients from two independent mammography datasets: Full-Field Digital Mammography Dataset (FFDM) and a digitized film dataset, Digital Dataset of Screening Mammography (DDSM), were used in various settings for training and testing the CNN models. The ROC curve was generated and the AUC was calculated as a metric of the classification accuracy. Results: Training and testing using only the FFDM dataset resulted in AUC ranging from 0.70 to 0.81. When the DDSM dataset was used, AUC ranged from 0.77 to 0.96. When datasets were combined for training and testing, AUC ranged from 0.76 to 0.91. When pretrained on a large nonmedical dataset and DDSM, the models showed consistent improvements in AUC ranging from 0.02 to 0.05 (all P > 0.05), compared with pretraining only on the nonmedical dataset. Conclusions: This study demonstrates that automatic deep learning CNN methods can identify nuanced mammographic imaging features to distinguish recalled-benign images from malignant and negative cases, which may lead to a computerized clinical toolkit to help reduce false recalls. Clin Cancer Res; 1–8. ©2018 AACR.
93 citations
TL;DR: Contrast-enhanced spectral mammography provides low-energy 2D mammographic images comparable to standard digital mammography and a post-contrast recombined image to assess tumor neovascularity similar to magnetic resonance imaging (MRI).
Abstract: Contrast-enhanced spectral mammography (CESM) provides low-energy 2D mammographic images comparable to standard digital mammography and a post-contrast recombined image to assess tumor neovascularity similar to magnetic resonance imaging (MRI). The utilization of CESM in the United States is currently low but could increase rapidly given many potential indications for clinical use. This article discusses historical background and literature review of indications and diagnostic accuracy of CESM to date. CESM is a growing technique for breast cancer detection and diagnosis that has levels of sensitivity and specificity on par with contrast-enhanced breast MRI. Because of its similar performance and ease of implementation, CESM is being adopted for multiple indications previously reserved for MRI, such as problem-solving, disease extent in newly diagnosed patients, and evaluating the treatment response of neoadjuvant chemotherapy.
TL;DR: ABUS seemed to outperform HHUS in the detection of architectural distortion on the coronal plane and can supplement mammography in the Detection of non-calcified carcinomas in women with dense breasts.
Abstract: This study aimed to evaluate automated breast ultrasound (ABUS) compared to hand-held traditional ultrasound (HHUS) in the visualisation and BIRADS characterisation of breast lesions. From January 2016 to January 2017, 1,886 women with breast density category C or D (aged 48.6±10.8 years) were recruited. All participants underwent ABUS and HHUS examination; a subcohort of 1,665 women also underwent a mammography. The overall agreement between HHUS and ABUS was 99.8 %; kappa=0.994, p<0.0001. Two cases were graded as BI-RADS 1 in HHUS, but were graded as BIRADS 4 in ABUS; biopsy revealed a radial scar. Three carcinomas were graded as BI-RADS 2 in mammography but BI-RADS 4 in ABUS; two additional carcinomas were graded as BI-RADS 2 in mammography but BI-RADS 5 in ABUS. Two carcinomas, appearing as a well-circumscribed mass or developing asymmetry in mammography, were graded as BI-RADS 4 in mammography but BI-RADS 5 in ABUS. ABUS could be successfully used in the visualisation and characterisation of breast lesions. ABUS seemed to outperform HHUS in the detection of architectural distortion on the coronal plane and can supplement mammography in the detection of non-calcified carcinomas in women with dense breasts. • The new generation of ABUS yields comparable results to HHUS.
• ABUS seems superior to HHUS in detecting architectural distortions.
• In dense breasts, supplemental ABUS to mammography detects additional cancers.
TL;DR: DBT+DM depicts 90% more cancers in a population previously screened with DM, with similar recall rates, compared to DM alone, which was notable for small and medium invasive cancers, but not for large ones.
Abstract: Tomosynthesis plus digital mammography detects 90% more cancers than digital mammography alone, with similar recall rate. This higher detection may have a beneficial impact on cancer prognosis.
TL;DR: This work deals with mammography images and presents a novel supervised deep learning-based framework for region classification into semantically coherent tissues using Convolutional Neural Network to learn discriminative features automatically.
Abstract: Automatic tissue classification from medical images is an important step in pathology detection and diagnosis. Here, we deal with mammography images and present a novel supervised deep learning-based framework for region classification into semantically coherent tissues. The proposed method uses Convolutional Neural Network (CNN) to learn discriminative features automatically. We overcome the difficulty involved in a medium-size database by training the CNN in an overlapping patch-wise manner. In order to accelerate the pixel-wise automatic class prediction, we use convolutional layers instead of the classical fully connected layers. This approach results in significantly faster computation, while preserving the classification accuracy. The proposed method was tested on annotated mammography images and demonstrates promising image segmentation and tissue classification results.
TL;DR: These findings show a modest increase of radiation dose to the breast by tomosynthesis compared to FFDM, and its use in conjunction with synthetic 2D images should not be deterred by concerns regarding radiation burden, and should draw on evidence of potential clinical benefit.
Abstract: To compare radiation dose delivered by digital mammography (FFDM) and breast tomosynthesis (DBT) for a single view. 4,780 FFDM and 4,798 DBT images from 1,208 women enrolled in a screening trial were used to ground dose comparison. Raw images were processed by an automatic software to determine volumetric breast density (VBD) and were used together with exposure data to compute the mean glandular dose (MGD) according to Dance’s model. DBT and FFDM were compared in terms of operation of the automatic exposure control (AEC) and MGD level. Statistically significant differences were found between FFDM and DBT MGDs for all views (CC: MGDFFDM=1.366 mGy, MGDDBT=1.858 mGy; p<0.0001; MLO: MGDFFDM=1.374 mGy, MGDDBT=1.877 mGy; p<0.0001). Considering the 4,768 paired views, Bland-Altman analysis showed that the average increase of DBT dose compared to FFDM is 38 %, and a range between 0 % and 75 %. Our findings show a modest increase of radiation dose to the breast by tomosynthesis compared to FFDM. Given the emerging role of DBT, its use in conjunction with synthetic 2D images should not be deterred by concerns regarding radiation burden, and should draw on evidence of potential clinical benefit. • Most studies compared tomosynthesis in combination with mammography vs. mammography alone.
• There is some concern about the dose increase with tomosynthesis.
• Clinical data show a small increase in radiation dose with tomosynthesis.
• Synthetic 2D images from tomosynthesis at zero dose reduce potential harm.
• The small dose increase should not be a barrier to use of tomosynthesis.
TL;DR: A fully automated pipeline of mammogram image processing is proposed, which estimates skin-air boundary using gradient weight map, detects pectoral-breast boundary by unsupervised pixel-wise labeling with no pre-labeled areas needed, and finally detects calcifications inside the breast region with a novel texture filter.
TL;DR: A Shallow-Deep Convolutional Neural Network (SD-CNN) where a shallow CNN is developed to derive "virtual" recombined images from LE images, and a deep CNN is employed to extract novel features from LE, recombined or " virtual" recombine images for ensemble models to classify the cases as benign vs. cancer.
TL;DR: A computer-aided detection (CAD) framework for mass detection in DBT has been developed and is described in this paper, demonstrating that this proposed CAD framework achieves much better performance than CAD systems that use hand-crafted features and deep cardinality-restricted Bolzmann machines to detect masses in D BTs.
TL;DR: This work proposes a novel approach to remove the pectoral muscle in terms of the mediolateral-oblique observation of a mammogram using a discrete differentiation operator that tolerates an extensive variety of the pECToral muscle geometries with minimum risk of bias in breast profile than existing techniques.
Abstract: In digital mammography, finding accurate breast profile segmentation of women’s mammogram is considered a challenging task. The existence of the pectoral muscle may mislead the diagnosis of cancer due to its high-level similarity to breast body. In addition, some other challenges due to manifestation of the breast body pectoral muscle in the mammogram data include inaccurate estimation of the density level and assessment of the cancer cell. The discrete differentiation operator has been proven to eliminate the pectoral muscle before the analysis processing. We propose a novel approach to remove the pectoral muscle in terms of the mediolateral-oblique observation of a mammogram using a discrete differentiation operator. This is used to detect the edges boundaries and to approximate the gradient value of the intensity function. Further refinement is achieved using a convex hull technique. This method is implemented on dataset provided by MIAS and 20 contrast enhanced digital mammographic images. To assess the performance of the proposed method, visual inspections by radiologist as well as calculation based on well-known metrics are observed. For calculation of performance metrics, the given pixels in pectoral muscle region of the input scans are calculated as ground truth. Our approach tolerates an extensive variety of the pectoral muscle geometries with minimum risk of bias in breast profile than existing techniques.
14 May 2018
TL;DR: The performance of the proposed CAD method for breast cancer screening using convolutional neural network (CNN) and follow-up scans is very promising, considering the early-stage cancerous status (1-year ago was normal).
Abstract: We propose a computer-aided detection (CAD) method for breast cancer screening using convolutional neural network (CNN) and follow-up scans. First, mammographic images are examined by three cascading object detectors to detect suspicious cancerous regions. Then all regional images are fed to a trained CNN (based on the pre-trained VGG-19 model) to filter out false positives. Three cascading detectors are trained with Haar features, local binary pattern (LBP) and histograms of oriented gradient (HOG) separately via an AdaBoost approach. The bounding boxes (BBs) from three featured detectors are merged to generate a region proposal. Each regional image, consisting of three channels, current scan (red channel), registered prior scan (green channel) and their difference (blue channel), is scaled to 224×224×3 for CNN classification. We tested the proposed method using our digital mammographic database including 69 cancerous subjects of mass, architecture distortion, and 27 healthy subjects, each of which includes two scans, current (cancerous or healthy), prior scan (healthy 1 year before). On average 165 BBs are created by three cascading classifiers on each mammogram, but only 3 BBs remained per image after the CNN classification. The overall performance is described as follows: sensitivity = 0.928, specificity = 0.991, FNR = 0.072, and FPI (false positives per image) = 0.004. Considering the early-stage cancerous status (1-year ago was normal), the performance of the proposed CAD method is very promising.
TL;DR: It is demonstrated that DES images eliminated overlapping effect of dense breast tissue, which helps improve mass segmentation accuracy, and applying a novel approach to optimally map mass region segmented from DES images to LE images enabled CAD to yield significantly improved performance.
Abstract: Contrast-enhanced digital mammography (CEDM) is a promising imaging modality in breast cancer diagnosis. This study aims to investigate how to optimally develop a computer-aided diagnosis (CAD) scheme of CEDM images to classify breast masses. A CEDM dataset of 111 patients was assembled, which includes 33 benign and 78 malignant cases. Each CEDM includes two types of images namely, low energy (LE) and dual-energy subtracted (DES) images. A CAD scheme was applied to segment mass regions depicting on LE and DES images separately. Optimal segmentation results generated from DES images were also mapped to LE images or vice versa. After computing image features, multilayer perceptron based machine learning classifiers that integrate with a correlation-based feature subset evaluator and leave-one-case-out cross-validation method were built to classify mass regions. When applying CAD to DES and LE images with original segmentation, areas under ROC curves (AUC) were 0.759 ± 0.053 and 0.753 ± 0.047, respectively. After mapping the mass regions optimally segmented on DES images to LE images, AUC significantly increased to 0.848 ± 0.038 (p < 0.01). Study demonstrated that DES images eliminated overlapping effect of dense breast tissue, which helps improve mass segmentation accuracy. The study demonstrated that applying a novel approach to optimally map mass region segmented from DES images to LE images enabled CAD to yield significantly improved performance.
TL;DR: Automated absolute breast density is a predictor of breast cancer risk in populations at enhanced risk due to either positive mammographic findings or family history, and could be a trigger for more intensive imaging in the screening context.
09 Mar 2018
TL;DR: The authors' VCTs showed an average AUC improvement for DBT vs DM of 0.027 for microcalcifications and 0.103 for masses, in close agreement (within 1%) of clinical data reported in the literature.
Abstract: We have designed and conducted 35 virtual clinical trials (VCTs) of breast lesion detection in digital mammography (DM) and digital breast tomosynthesis (DBT) using a novel open-source simulation pipeline, OpenVCT. The goal of the VCTs is to test in-silico reports that DBT provides substantial improvements in the detectability of masses, while the detectability of microcalcifications remains comparable to DM. For this test, we generated 12 software breast phantoms (volume 700ml, compressed thickness 6.33cm), varying the number of simulated tissue compartments and their shape. Into each phantom, we inserted multiple lesions located 2cm apart in the plane parallel to detector at the level of the nipple. Simulated ellipsoidal masses (oblate spheroids 7mm in diameter and of various thicknesses) and single calcifications of various size and composition were inserted; a total of 17,640 lesions were simulated for this project. DM and DBT projections of phantoms with and without lesions were synthesized assuming a clinical acquisition geometry. Exposure parameters (mAs and kVp) were selected to match AEC settings. Processed DM images and reconstructed DBT slices were obtained using a commercially available software library. Lesion detection was simulated by channelized Hotelling observers, with 15 LG channels and a spread of 22, using independent sets of 480 image samples (150×150 pixel ROIs) for training and 480 samples for testing. Our VCTs showed an average AUC improvement for DBT vs DM of 0.027 for microcalcifications and 0.103 for masses, in close agreement (within 1%) of clinical data reported in the literature.
TL;DR: It is recommended that digital mammography should replace screen-film mammography as a basic tool to detect breast cancer for both screening and diagnostic purposes.
Abstract: Introduction
Breast cancer has a high prevalence in the community and places very high demands on resources. Digital mammography provides a good quality image with reduced radiation dose and can detect breast carcinoma in its earlier stages, resulting in good prognosis and improved patient survival.
Objective
To calculate the diagnostic accuracy of digital mammography in the detection of breast cancer, using histopathology as a gold standard in women aged over 30 years, who are undergoing mammography for screening and diagnostic purposes.
Materials and methods
This was a cross-sectional analytical study, conducted in the department of radiology, for a total duration of 10 months. A total of 122 patients of age above 30 years, referred for digital mammography for the evaluation of different symptoms related to breast diseases, followed by biopsy/surgery and histopathology, were included in the study.
Result
Our data confirmed that digital mammography is a highly accurate tool for breast cancer detection having a sensitivity of 97%, a specificity of 64.5%, a positive predictive value of 89%, and a negative predictive value of 90.9%, with a diagnostic accuracy of 89.3%.
Conclusion
Considering our results, we recommend that digital mammography should replace screen-film mammography as a basic tool to detect breast cancer for both screening and diagnostic purposes.
TL;DR: Interval breast cancer rate amongst screening participants in the STORM trial was marginally lower (and screening sensitivity higher) than estimates amongst 2D-screened women; these findings should be interpreted with caution given the small number of interval cases and the sample size of the trial.
TL;DR: It was found that CAD use at US digital screening mammography facilities was stable from 2008 to 2016, and this persistent utilization is relevant to the debate on the value of targeting ductal carcinoma in situ in screening.
Abstract: Purpose Computer-aided detection (CAD) for screening mammography is a software technology designed to improve radiologists' reading performance. Since 2007, multiple Breast Cancer Surveillance Consortium research papers have shown that CAD decreases performance by increasing recalls and decreasing the detection of invasive cancer while increasing the detection of ductal carcinoma in situ. The aim of this study was to test the hypothesis that CAD use by digital mammography facilities would decrease over time. Methods In August 2007, August 2011, and March 2016, the FDA database of certified mammography facilities was accessed, and a random sample of 400 of approximately 8,500 total facilities was generated. In 2008 and 2011, a telephone survey was conducted of the facilities regarding digital mammography and CAD use. In 2016, facility websites were reviewed before calling the facilities. Bonferroni-corrected P values were used to assess statistical differences in the proportion of CAD at digital facilities for the three surveys. Results The mean proportion of digital facilities using CAD was 91.4%, including 91.4% (128 of 140) in 2008, 90.2% (238 of 264) in 2011, and 92.3% (358 of 388) in 2016. The difference for 2008 versus 2011 was 1.3% (95% confidence interval [CI], −0.5% to 7.7%), for 2011 versus 2016 was −2.1% (95% CI, −6.9% to 2.7%), and for 2008 versus 2016 was −0.8% (95% CI, −6.7% to 5.0%). Conclusions In three national surveys, it was found that CAD use at US digital screening mammography facilities was stable from 2008 to 2016. This persistent utilization is relevant to the debate on the value of targeting ductal carcinoma in situ in screening.
TL;DR: A fully automatic computer-aided detection and diagnosis (CAD) system to detect and classify breast cancer as malignant or benign, by using mammography and building on the multiple-instance learning (MIL) algorithms, which has been confirmed beneficial for radiologist decision sustenance.
Abstract: Cancer tissues in mammography images exhibit abnormal regions; it is of great clinical importance to label a mammography image as having cancerous regions or not, perform the corresponding image segmentation. However, the detailed annotation of the cancer region is often an ambiguous and challenging task. The authors describe a fully automatic computer-aided detection and diagnosis (CAD) system to detect and classify breast cancer as malignant or benign, by using mammography and building on the multiple-instance learning (MIL) algorithms, which has been confirmed beneficial for radiologist decision sustenance. Traditional learning methods require great effort to annotate the training data by costly manual labelling and specialised computational models to detect these annotations during the test. The proposed CAD system simultaneously performs pixel-level segmentation (suspicious versus normal tissue) and image-level classification (benign versus malignant image). The set-up of the proposed system is in order: automatically segmented regions of interest (ROIs). Then, features derived from ROIs detected such as textural features and shape features are selected and extracted from each region and combined them to classify ROIs as `benign' or `malignant', by implementing MIL algorithms. Experimental results demonstrate the efficiency and robustness of the proposed CAD system compared with previous work in the literature.
TL;DR: Digital breast tomosynthesis increases reader confidence in the detection of architectural distortion, and DBT decreases IOV, increases confidence, and improves sensitivity while maintaining high specificity in detecting AD.
Abstract: To compare interobserver variability (IOV), reader confidence, and sensitivity/specificity in detecting architectural distortion (AD) on digital mammography (DM) versus digital breast tomosynthesis (DBT). This IRB-approved, HIPAA-compliant reader study used a counterbalanced experimental design. We searched radiology reports for AD on screening mammograms from 5 March 2012–27 November 2013. Cases were consensus-reviewed. Controls were selected from demographically matched non-AD examinations. Two radiologists and two fellows blinded to outcomes independently reviewed images from two patient groups in two sessions. Readers recorded presence/absence of AD and confidence level. Agreement and differences in confidence and sensitivity/specificity between DBT versus DM and attendings versus fellows were examined using weighted Kappa and generalised mixed modeling, respectively. There were 59 AD patients and 59 controls for 1,888 observations (59 × 2 (cases and controls) × 2 breasts × 2 imaging techniques × 4 readers). For all readers, agreement improved with DBT versus DM (0.61 vs. 0.37). Confidence was higher with DBT, p = .001. DBT achieved higher sensitivity (.59 vs. .32), p .90). DBT achieved higher positive likelihood ratio values, smaller negative likelihood ratio values, and larger ROC values. DBT decreases IOV, increases confidence, and improves sensitivity while maintaining high specificity in detecting AD. • Digital breast tomosynthesis decreases interobserver variability in the detection of architectural distortion. • Digital breast tomosynthesis increases reader confidence in the detection of architectural distortion. • Digital breast tomosynthesis improves sensitivity in the detection of architectural distortion.
TL;DR: This study proposed to leverage the newly emerged deep learning approach to investigate how the MLO and CC view images of a mammogram examination may have been clinically used by radiologists in coming up with a BI-RADS density category, and implemented a convolutional neural network (CNN)-based deep learning model aimed at distinguishing the breast density categories.
Abstract: Mammographic breast density has been established as an independent risk marker for developing breast cancer. Breast density assessment is a routine clinical need in breast cancer screening and current standard is using the Breast Imaging and Reporting Data System (BI-RADS) criteria including four qualitative categories (i.e., fatty, scattered density, heterogeneously dense, or extremely dense). In each mammogram examination, a breast is typically imaged with two different views, i.e., the mediolateral oblique (MLO) view and cranial caudal (CC) view. The BI-RADS-based breast density assessment is a qualitative process made by visual observation of both the MLO and CC views by radiologists, where there is a notable inter- and intra-reader variability. In order to maintain consistency and accuracy in BI-RADS-based breast density assessment, gaining understanding on radiologists' reading behaviors will be educational. In this study, we proposed to leverage the newly emerged deep learning approach to investigate how the MLO and CC view images of a mammogram examination may have been clinically used by radiologists in coming up with a BI-RADS density category. We implemented a convolutional neural network (CNN)-based deep learning model, aimed at distinguishing the breast density categories using a large (15,415 images) set of real-world clinical mammogram images. Our results showed that the classification of density categories (in terms of area under the receiver operating characteristic curve) using MLO view images is significantly higher than that using the CC view. This indicates that most likely it is the MLO view that the radiologists have predominately used to determine the breast density BI-RADS categories. Our study holds a potential to further interpret radiologists' reading characteristics, enhance personalized clinical training to radiologists, and ultimately reduce reader variations in breast density assessment.
TL;DR: Although the data provide some reassurance that DBT does not increase the proportion of screen-detected DCIS, they highlight mixed findings on comparative tumour characteristics, suggesting a potential for enhancing screening benefit and possibly also over-diagnosis from DBT screening.
Abstract: The Verona population-based breast cancer (BC) screening program provides biennial mammography to women aged 50–69 years. Based on emerging evidence of enhanced detection, the program transitioned to digital breast tomosynthesis (DBT) screening. This is a prospective pilot evaluation of DBT with synthesised 2D mammography screening implemented during April 2015–March 2017; the rate and characteristics of cancers detected at DBT screening were compared with those detected at the preceding digital mammography (DM) screening round (April 2013–March 2015) in the same screening program. Distribution of imaging and tumour characteristics were compared. Amongst 34,071 women screened in the Verona DBT pilot, 315 BCs were detected; 153 BCs were detected amongst 29,360 women in the DM screening round. Estimated CDRs were 9.2/1000 (95% CI 8.3–10.3) DBT screens versus 5.2/1000 (95% CI 4.4–6.1) DM screens, P < 0.001. Statistically significant differences were found in the distribution of whether recall by one/both screen readers (more BCs recalled by both readers at DBT than DM); whether detected on one/two views (higher proportion detected on only one view at DBT than DM); type of radiological lesions; tumour stage, pT and histological categories (lower proportion of DCIS/pTis, higher proportions of pT1a and pT1b, and higher proportion of invasive cancers of special types, at DBT than DM); and tumour grade (higher proportion of grade I at DBT than DM). There were no differences in distributions of nodal and hormone receptor (ER/PR) status. Our findings provide early insights into the extent that transitioning to DBT screening may modify the characteristics of screen-detected breast cancer to inform discussion regarding pros and cons of DBT screening; although our data provide some reassurance that DBT does not increase the proportion of screen-detected DCIS, they highlight mixed findings on comparative tumour characteristics, suggesting a potential for enhancing screening benefit and possibly also over-diagnosis from DBT screening.