Alejandro Rodriguez-Ruiz

New reconstruction algorithm for digital breast tomosynthesis: better image quality for humans and computers:

Background The image quality of digital breast tomosynthesis (DBT) volumes depends greatly on the reconstruction algorithm. Purpose To compare two DBT reconstruction algorithms used by the Siemens Mammomat Inspiration system, filtered back projection (FBP), and FBP with iterative optimizations (EMPIRE), using qualitative analysis by human readers and detection performance of machine learning algorithms. Material and Methods Visual grading analysis was performed by four readers specialized in breast imaging who scored 100 cases reconstructed with both algorithms (70 lesions). Scoring (5-point scale: 1 = poor to 5 = excellent quality) was performed on presence of noise and artifacts, visualization of skin-line and Cooper’s ligaments, contrast, and image quality, and, when present, lesion visibility. In parallel, a three-dimensional deep-learning convolutional neural network (3D-CNN) was trained (n = 259 patients, 51 positives with BI-RADS 3, 4, or 5 calcifications) and tested (n = 46 patients, nine positives), separately with FBP and EMPIRE volumes, to discriminate between samples with and without calcifications. The partial area under the receiver operating characteristic curve (pAUC) of each 3D-CNN was used for comparison. Results EMPIRE reconstructions showed better contrast (3.23 vs. 3.10, P = 0.010), image quality (3.22 vs. 3.03, P < 0.001), visibility of calcifications (3.53 vs. 3.37, P = 0.053, significant for one reader), and fewer artifacts (3.26 vs. 2.97, P < 0.001). The 3D-CNN-EMPIRE had better performance than 3D-CNN-FBP (pAUC-EMPIRE = 0.880 vs. pAUC-FBP = 0.857; P < 0.001). Conclusion The new algorithm provides DBT volumes with better contrast and image quality, fewer artifacts, and improved visibility of calcifications for human observers, as well as improved detection performance with deep-learning algorithms.

Improvements of an objective model of compressed breasts undergoing mammography: Generation and characterization of breast shapes

Purpose To develop a set of accurate 2D models of compressed breasts undergoing mammography or breast tomosynthesis, based on objective analysis, to accurately characterize mammograms with few linearly independent parameters, and to generate novel clinically realistic paired cranio‐caudal (CC) and medio‐lateral oblique (MLO) views of the breast. Methods We seek to improve on an existing model of compressed breasts by overcoming detector size bias, removing the nipple and non‐mammary tissue, pairing the CC and MLO views from a single breast, and incorporating the pectoralis major muscle contour into the model. The outer breast shapes in 931 paired CC and MLO mammograms were automatically detected with an in‐house developed segmentation algorithm. From these shapes three generic models (CC‐only, MLO‐only, and joint CC/MLO) with linearly independent components were constructed via principal component analysis (PCA). The ability of the models to represent mammograms not used for PCA was tested via leave‐one‐out cross‐validation, by measuring the average distance error (ADE). Results The individual models based on six components were found to depict breast shapes with accuracy (mean ADE‐CC = 0.81 mm, ADE‐MLO = 1.64 mm, ADE‐Pectoralis = 1.61 mm), outperforming the joint CC/MLO model (P ≤ 0.001). The joint model based on 12 principal components contains 99.5% of the total variance of the data, and can be used to generate new clinically realistic paired CC and MLO breast shapes. This is achieved by generating random sets of 12 principal components, following the Gaussian distributions of the histograms of each component, which were obtained from the component values determined from the images in the mammography database used. Conclusion Our joint CC/MLO model can successfully generate paired CC and MLO view shapes of the same simulated breast, while the individual models can be used to represent with high accuracy clinical acquired mammograms with a small set of parameters. This is the first step toward objective 3D compressed breast models, useful for dosimetry and scatter correction research, among other applications.

Analysis of volume overestimation artifacts in the breast outline segmentation in tomosynthesis

In digital breast tomosynthesis (DBT), the reconstruction is calculated from x-ray projection images acquired over a small range of angles. One step in the reconstruction process is to identify the pixels that fall outside the shadow of the breast, to segment the breast from the background (air). In each projection, rays are back-projected from these pixels to the focal spot. All voxels along these rays are identified as air. By combining these results over all projections, a breast outline can be determined for the reconstruction. This paper quantifies the accuracy of this breast segmentation strategy in DBT. In this study, a physical phantom modeling a breast under compression was analyzed with a prototype next-generation tomosynthesis (NGT) system described in previous work. Multiple wires were wrapped around the phantom. Since the wires are thin and high contrast, their exact location can be determined from the reconstruction. Breast parenchyma was portrayed outside the outline defined by the wires. Specifically, the size of the phantom was overestimated along the posteroanterior (PA) direction; i.e., perpendicular to the plane of conventional source motion. To analyze how the acquisition geometry affects the accuracy of the breast outline segmentation, a computational phantom was also simulated. The simulation identified two ways to improve the segmentation accuracy; either by increasing the angular range of source motion laterally or by increasing the range in the PA direction. The latter approach is a unique feature of the NGT design; the advantage of this approach was validated with our prototype system.

Pectoral muscle segmentation in breast tomosynthesis with deep learning

Digital breast tomosynthesis (DBT) has superior detection performance than mammography (DM) for population-based breast cancer screening, but the higher number of images that must be reviewed poses a challenge for its implementation. This may be ameliorated by creating a twodimensional synthetic mammographic image (SM) from the DBT volume, containing the most relevant information. When creating a SM, it is of utmost importance to have an accurate lesion localization detection algorithm, while segmenting fibroglandular tissue could also be beneficial. These tasks encounter an extra challenge when working with images in the medio-lateral oblique view, due to the presence of the pectoral muscle, which has similar radiographic density. In this work, we present an automatic pectoral muscle segmentation model based on a u-net deep learning architecture, trained with 136 DBT images acquired with a single system (different BIRADS ® densities and pathological findings). The model was tested on 36 DBT images from that same system resulting in a dice similarity coefficient (DSC) of 0.977 (0.967-0.984). In addition, the model was tested on 125 images from two different systems and three different modalities (DBT, SM, DM), obtaining DSCs between 0.947 and 0.970, a range determined visually to provide adequate segmentations. For reference, a resident radiologist independently annotated a mix of 25 cases obtaining a DSC of 0.971. The results suggest the possibility of using this model for inter-manufacturer DBT, DM and SM tasks that benefit from the segmentation of the pectoral muscle, such as SM generation, computer aided detection systems, or patient dosimetry algorithms.

One-view digital breast tomosynthesis as a stand-alone modality for breast cancer detection : do we need more?

PurposeTo compare the performance of one-view digital breast tomosynthesis (1v-DBT) to that of three other protocols combining DBT and mammography (DM) for breast cancer detection.Materials and methodsSix radiologists, three experienced with 1v-DBT in screening, retrospectively reviewed 181 cases (76 malignant, 50 benign, 55 normal) in two sessions. First, they scored sequentially: 1v-DBT (medio-lateral oblique, MLO), 1v-DBT (MLO) + 1v-DM (cranio-caudal, CC) and two-view DM + DBT (2v-DM+2v-DBT). The second session involved only 2v-DM. Lesions were scored using BI-RADS® and level of suspiciousness (1–10). Sensitivity, specificity, receiver operating characteristic (ROC) and jack-knife alternative free-response ROC (JAFROC) were computed.ResultsOn average, 1v-DBT was non-inferior to any of the other protocols in terms of JAFROC figure-of-merit, area under ROC curve, sensitivity or specificity (p>0.391). While readers inexperienced with 1v-DBT screening improved their sensitivity when adding more images (69–79 %, p=0.019), experienced readers showed similar sensitivity (76 %) and specificity (70 %) between 1v-DBT and 2v-DM+2v-DBT (p=0.482). Subanalysis by lesion type and breast density showed no difference among modalities.ConclusionDetection performance with 1v-DBT is not statistically inferior to 2v-DM or to 2v-DM+2v-DBT; its use as a stand-alone modality might be sufficient for readers experienced with this protocol.Key points• One-view breast tomosynthesis is not inferior to two-view digital mammography.• One-view DBT is not inferior to 2-view DM plus 2-view DBT.• Training may lead to 1v-DBT being sufficient for screening.

Comparison of breast cancer detection and depiction between planar and rotating synthetic mammography generated from breast tomosynthesis

PURPOSE To compare breast cancer detection and depiction between planar synthetic mammography (SM) and rotating synthetic mammography (RM) generated from digital breast tomosynthesis (DBT). MATERIALS AND METHODS In a fully-crossed multi-reader multi-case (MRMC) study, three radiologists retrospectively reviewed 190 cases (27 malignant, 31 benign, 132 normal), once with SM alone and once with RM alone, the DBT stack of slices was not reviewed. Lesions were scored using BI-RADS® and level of suspiciousness (1-10). Sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were computed using MRMC Analysis of Variance using the open-access software iMRMC. Additionally, readers were asked to make a visual grading analysis (VGA) on visibility of calcifications and soft tissue lesions (1-5 scale with 5 = Excellent visualization). The VGA scores were analyzed using the visual grading characteristics (VGC) method. RESULTS On average, the AUC was similar between SM and RM (0.66 versus 0.67, P = 0.818). The sensitivity was equivalent (0.62 versus 0.60, P = 0.794), while specificity was significantly lower in SM than in RM (0.66 versus 0.72, P = 0.028). Radiologists significantly (P < 0.05) preferred the display of all types of lesions in RM over SM. The average reading time per case was higher for RM than for SM (30 s versus 23 s, P < 0.05). CONCLUSION Radiologists achieve similar cancer detection with RM as with SM. They prefer the 3D-like rotating representation of soft tissue lesions and calcifications in comparison to the 2D visualization, which might improve their specificity, but at the expense of longer reading time.

Phantom-based comparison of microcalcification visibility between digital and synthetic mammography using humans and a deep neural network as observers.

The 2D synthetic image (SM) generated from digital breast tomosynthesis (DBT) has the potential to replace conventional digital mammography (DM), therefore reducing patient dose without affecting the cancer detection performance. In this work, we analysed the image quality of SMs from three different manufacturers for the specific task of detecting microcalcifications (MC), in comparison to DM. A phantom with MC clusters on a uniform background was employed, thus also allowing to explore its feasibility to be used for quality control (QC). A 4-Alternative Forced Choice (4AFC) experiment was performed by four human observers, for detection of MC clusters on a region-of-interest level. We also explored the possibility to replace human observers with a virtual observer. For this, we developed a deep learning convolutional neural network (CNN) for the task of classifying the same images from the 4AFC study, and then compare the results to the human-based study. The results showed that for the four readers and all the systems, the percentage of correct answers (PC) was 100% and the visibility was 3 for the largest MC clusters. However, SM yielded worse detectability than DM for MC with sizes between 180 and 100 μm (PC was around 18% inferior in average). The CNN yielded the same relative results across modalities and systems than the 4AFC study, but in terms of the area under the receiver operating characteristic curve. This might encourage the possibility to develop QC procedures based on artificial intelligence image reading, improving reproducibility and reducing costs.

Can radiologists improve their breast cancer detection in mammography when using a deep learning based computer system as decision support

For more than a decade, radiologists have used traditional computer aided detection systems to read mammograms, but mainly because of a low computer specificity may not improve their screening performance, according to several studies. The breakthrough in deep learning techniques has boosted the performance of machine learning algorithms, also for breast cancer detection in mammography. The objective of this study was to determine whether radiologists improve their breast cancer detection performance when they concurrently use a deep learningbased computer system for decision support, compared to when they read mammography unaided. A retrospective, fully-crossed, multi-reader multi-case (MRMC) study was designed to compare this. The employed decision support system was Transpara™ (Screenpoint Medical, Nijmegen, the Netherlands). Radiologists interact by clicking an area on the mammogram, for which the computer system displays its cancer likelihood score (1-100). In total, 240 cases (100 cancers, 40 false positive recalls, 100 normals) acquired with two different mammography systems were retrospectively collected. Seven radiologists scored each case once with, and once without the use of decision support, providing a forced BI-RADS® score and a level of suspiciousness (1-100). MRMC analysis of variance of the area under the receiver operating characteristic curves (AUC), and specificity and sensitivity were computed. When using decision support, the AUC increased from 0.87 to 0.89 (P=0.043) and specificity increased from 73% to 78% (P=0.030), while sensitivity did not significantly increment (84% to 87%, P=0.180). In conclusion, radiologists significantly improved their performance when using a deep learningbased computer system as decision support.

How does wide-angle breast tomosynthesis depict calcifications in comparison to digital mammography? A retrospective observer study.

Digital breast tomosynthesis (DBT) provides superior breast cancer detection performance compared to digital mammography (DM), but it is unclear whether DBT alone is sufficient to accurately visualize lesions with calcifications, or supplemental DM is needed. In this work, we performed a retrospective observer study to assess and compare the depiction of calcifications on DM, DBT, and synthetic mammography (SM). Eighty views from 40 lesions with calcifications in 39 patients acquired with a wide-angle DBT system were included (two views per case - cranio-caudal and medio-lateral oblique). Four experienced researchers (3, 10, 11, 21 years) in breast imaging scored the images. For each case, the regions-of-interest containing calcifications in DM, DBT and SM were shown simultaneously. The readers ranked (ties allowed) the three modalities for the depiction of calcifications and assessed if more blurring was present in DM or DBT. DM was ranked as the best modality to depict calcification lesions in 84% of the cases, DBT in 22%, and SM in 7% (P<0.001). Similarly, for 86% of the views, DBT had more blurring of the calcifications than DM. In some cases, DBT showed higher contrast of calcifications providing better visualization, but worse characterization due to signal blurring. For cases with subtle calcifications, the higher noise of DBT images deteriorated their visualization. SM was preferred over DBT for large clusters, while it failed in some cases to display any calcifications. In conclusion, our results show the current limitations of DBT and its derived SM to depict calcifications in comparison to DM.

Automated lesion detection and segmentation in digital mammography using a u-net deep learning network.

Computer-aided detection or decision support systems aim to improve breast cancer screening programs by helping radiologists to evaluate digital mammography (DM) exams. Commonly such methods proceed in two steps: selection of candidate regions for malignancy, and later classification as either malignant or not. In this study, we present a candidate detection method based on deep learning to automatically detect and additionally segment soft tissue lesions in DM. A database of DM exams (mostly bilateral and two views) was collected from our institutional archive. In total, 7196 DM exams (28294 DM images) acquired with systems from three different vendors (General Electric, Siemens, Hologic) were collected, of which 2883 contained malignant lesions verified with histopathology. Data was randomly split on an exam level into training (50%), validation (10%) and testing (40%) of deep neural network with u-net architecture. The u-net classifies the image but also provides lesion segmentation. Free receiver operating characteristic (FROC) analysis was used to evaluate the model, on an image and on an exam level. On an image level, a maximum sensitivity of 0.94 at 7.93 false positives (FP) per image was achieved. Similarly, per exam a maximum sensitivity of 0.98 at 7.81 FP per image was achieved. In conclusion, the method could be used as a candidate selection model with high accuracy and with the additional information of lesion segmentation.