Evaluation of the effect of computer-aided classification of benign and malignant lesions on reader performance in automated three-dimensional breast ultrasound.

RATIONALE AND OBJECTIVES: To investigate the effect of a newly developed computer-aided diagnosis (CAD) system on reader interpretation of breast lesions in automated three-dimensional (3D) breast ultrasound. MATERIALS AND METHODS: A CAD system was developed to differentiate malignant lesions from benign lesions including automated lesion segmentation in three dimensions; extraction of lesion features such as spiculation, margin contrast, and posterior acoustic behavior; and a classification stage. Eighty-eight patients with breast lesions were included for an observer study: 47 lesions were malignant and 41 were benign. Eleven readers (seven radiologists and four residents) read the cases with and without CAD. We compared the performance of readers with and without CAD using receiver operating characteristic (ROC) analysis. RESULTS: The CAD system had an area under the ROC curve (AUC) of 0.92 for discriminating benign and malignant lesions, whereas the unaided reader AUC ranged from 0.77 to 0.92. Mean performance of inexperienced readers improved when CAD was used (AUC = 0.85 versus 0.90; P = .007), whereas mean performance of experienced readers did not change with CAD (AUC = 0.89). CONCLUSIONS: By using the CAD system for classification of lesions in automated 3D breast ultrasound, which on its own performed as good as the best readers, the performance of inexperienced readers improved while that of experienced readers remained unaffected.

Generating Synthetic Mammograms From Reconstructed Tomosynthesis Volumes.

Digital breast tomosynthesis (DBT) is a promising 3-D modality that may replace mammography in the future. However, lesion search is likely to require more time in DBT volumes, while comparisons between views from different projections and prior exams might be harder to make. This may make screening with DBT cumbersome. A solution may be provided by synthesizing 2-D mammograms from DBT, which may then be used to guide the search for abnormalities. In this work we focus on synthesizing mammograms in which masses and architectural distortions are optimally visualized. Our approach first determines relevant points in a DBT volume with a computer-aided detection system and then renders a mammogram from the intersection of a surface fitted through these points and the DBT volume. The method was evaluated in a pilot observer study where three readers reported mass findings in 87 patients (25 malignant, 62 normal) for which both DBT and digital mammograms were available. We found that on average, diagnostic accuracy in the synthetic mammograms was higher (Az=0.85) than in conventional mammograms (Az=0.81), although the difference was not statistically significant. Preliminary results suggest that the synthesized mammograms are an acceptable alternative for real mammograms regarding the detection of mass lesions.

Improved location features for linkage of regions across ipsilateral mammograms.

Improved performance has been reported for computer aided detection (CADe) methods using information from multiple mammographic views over single-view CADe approaches. Linkage across the views is based on assuming that location and image features from the same lesion depicted in both views will be similar. In this study we investigate if the location features can be improved and what effect such an improvement has on the linkage of lesions across ipsilateral views. Performance of different methods to define the location features was first assessed with respect to the location of 137 manually annotated and linked masses. Taking the median result from five complementary methods (based on pectoral muscle boundary, breast shape and intensity signature) increased the mean accuracy compared to the current standard (7.1 vs. 6.3 mm). Thereafter the impact of this best method on the automatic linkage of detected regions across views was assessed for a second, independent dataset of 131 mammogram pairs. Linkage was based on the combination of location and single-view image features by a linear discriminate analysis classifier trained to differentiate between links of corresponding true-positive (TP) regions versus links including TP and false-positive (FP) regions. Nested cross-validation results showed that using the improved location features significantly increased the classification performance and the percentage of correctly linked regions.

Mass detection in reconstructed digital breast tomosynthesis volumes with a computer-aided detection system trained on 2D mammograms.

PURPOSE: To develop a computer-aided detection (CAD) system for masses in digital breast tomosynthesis (DBT) which can make use of an existing CAD system for detection of breast masses in full-field digital mammography (FFDM). This approach has the advantage that large digital screening databases that are becoming available can be used for training. DBT is currently not used for screening which makes it hard to obtain sufficient data for training. METHODS: The proposed CAD system is applied to reconstructed DBT volumes and consists of two stages. In the first stage, an existing 2D CAD system is applied to slabs composed of multiple DBT slices, after processing the slabs to a representation similar to that of the FFDM training data. In the second stage, the authors group detections obtained in the slabs that detect the same object and determine the 3D location of the grouped findings using one of three different approaches, including one that uses a set of features extracted from the DBT slabs. Experiments were conducted to determine performance of the CAD system, the optimal slab thickness for this approach and the best method to establish the 3D location. Experiments were performed using a database of 192 patients (752 DBT volumes). In 49 patients, one or more malignancies were present which were described as a mass, architectural distortion, or asymmetry. Free response receiver operating characteristic analysis and bootstrapping were used for statistical evaluation. RESULTS: Best performance was obtained when slab thickness was in the range of 1-2 cm. Using the feature based 3D localization procedure developed in the study, accurate 3D localization could be obtained in most cases. Case sensitivities of 80% and 90% were achieved at 0.35 and 0.99 false positives per volume, respectively. CONCLUSIONS: This study indicates that there may be a large benefit in using 2D mammograms for the development of CAD for DBT and that there is no need to exclusively limit development to DBT data.

Correlating locations in ipsilateral breast tomosynthesis views using an analytical hemispherical compression model.

To improve cancer detection in mammography, breast examinations usually consist of two views per breast. In order to combine information from both views, corresponding regions in the views need to be matched. In 3D digital breast tomosynthesis (DBT), this may be a difficult and time-consuming task for radiologists, because many slices have to be inspected individually. For multiview computer-aided detection (CAD) systems, matching corresponding regions is an essential step that needs to be automated. In this study, we developed an automatic method to quickly estimate corresponding locations in ipsilateral tomosynthesis views by applying a spatial transformation. First we match a model of a compressed breast to the tomosynthesis view containing a point of interest. Then we estimate the location of the corresponding point in the ipsilateral view by assuming that this model was decompressed, rotated and compressed again. In this study, we use a relatively simple, elastically deformable sphere model to obtain an analytical solution for the transformation in a given DBT case. We investigate three different methods to match the compression model to the data by using automatic segmentation of the pectoral muscle, breast tissue and nipple. For validation, we annotated 208 landmarks in both views of a total of 146 imaged breasts of 109 different patients and applied our method to each location. The best results are obtained by using the centre of gravity of the breast to define the central axis of the model, around which the breast is assumed to rotate between views. Results show a median 3D distance between the actual location and the estimated location of 14.6 mm, a good starting point for a registration method or a feature-based local search method to link suspicious regions in a multiview CAD system. Approximately half of the estimated locations are at most one slice away from the actual location, which makes the method useful as a mammographic workstation tool for radiologists to interactively find corresponding locations in ipsilateral tomosynthesis views.