Alexandra Edwards

Relevance vector machine for automatic detection of clustered microcalcifications

Clustered microcalcifications (MC) in mammograms can be an important early sign of breast cancer in women. Their accurate detection is important in computer-aided detection (CADe). In this paper, we propose the use of a recently developed machine-learning technique - relevance vector machine (RVM) - for detection of MCs in digital mammograms. RVM is based on Bayesian estimation theory, of which a distinctive feature is that it can yield a sparse decision function that is defined by only a very small number of so-called relevance vectors. By exploiting this sparse property of the RVM, we develop computerized detection algorithms that are not only accurate but also computationally efficient for MC detection in mammograms. We formulate MC detection as a supervised-learning problem, and apply RVM as a classifier to determine at each location in the mammogram if an MC object is present or not. To increase the computation speed further, we develop a two-stage classification network, in which a computationally much simpler linear RVM classifier is applied first to quickly eliminate the overwhelming majority, non-MC pixels in a mammogram from any further consideration. The proposed method is evaluated using a database of 141 clinical mammograms (all containing MCs), and compared with a well-tested support vector machine (SVM) classifier. The detection performance is evaluated using free-response receiver operating characteristic (FROC) curves. It is demonstrated in our experiments that the RVM classifier could greatly reduce the computational complexity of the SVM while maintaining its best detection accuracy. In particular, the two-stage RVM approach could reduce the detection time from 250 s for SVM to 7.26 s for a mammogram (nearly 35-fold reduction). Thus, the proposed RVM classifier is more advantageous for real-time processing of MC clusters in mammograms.

Automated detection of microcalcification clusters for digital breast tomosynthesis using projection data only: A preliminary study

Digital breast tomosynthesis (DBT) is a promising modality for breast imaging in which an anisotropic volume image of the breast is obtained. We present an algorithm for computerized detection of microcalcification clusters (MCCs) for DBT. This algorithm operates on the projection views only. Therefore it does not depend on reconstruction, and is computationally efficient. The algorithm was developed using a database of 30 image sets with microcalcifications, and a control group of 30 image sets without visible findings. The patient data were acquired on the first DBT prototype at Massachusetts General Hospital. Algorithm sensitivity was estimated to be 0.86 at 1.3 false positive clusters, which is below that of current MCC detection algorithms for full-field digital mammography. Because of the small number of patient cases, algorithm parameters were not optimized and one linear classifier was used. An actual limitation of our approach may be that the signal-to-noise ratio in the projection images is too low for microcalcification detection. Furthermore, the database consisted of predominantly small MCC. This may be related to the image quality obtained with this first prototype.

Clinically Missed Cancer: How Effectively Can Radiologists Use Computer-Aided Detection?

OBJECTIVE The purpose of this study was to determine the effectiveness with which radiologists can use computer-aided detection (CADe) to detect cancer missed at screening. MATERIALS AND METHODS An observer study was performed to measure the ability of radiologists to detect breast cancer on mammograms with and without CADe. The images in the study were from 300 analog mammographic examinations. In 234 cases the mammograms were read clinically as normal and free of cancer for at least 2 subsequent years. In the other 66 cases, cancers were missed clinically. In 256 cases, current and previous mammograms were available. Eight radiologists read the dataset and recorded a BI-RADS assessment, the location of the lesion, and their level of confidence that the patient should be recalled for diagnostic workup for each suspicious lesion. Jackknife alternative free-response receiver operating characteristic analysis was used. RESULTS The jackknife alternative free-response receiver operating characteristic figure of merit was 0.641 without aid and 0.659 with aid (p = 0.06; 95% CI, -0.001 to 0.036). The sensitivity increased 9.9% (95% CI, 3.4-19%) and the callback rate 12.1% (95% CI, 7.3-20%) with CADe. Both increases were statistically significant (p < 0.001). Radiologists on average ignored 71% of correct computer prompts. CONCLUSION Use of CADe can increase radiologist sensitivity 10% with a comparable increase in recall rate. There is potential for CADe to have a bigger clinical impact because radiologists failed to recognize a correct computer prompt in 71% of missed cancer cases [corrected].

Level Set Segmentation of Breast Masses in Contrast-Enhanced Dedicated Breast CT and Evaluation of Stopping Criteria

Dedicated breast CT (bCT) produces high-resolution 3D tomographic images of the breast, fully resolving fibroglandular tissue structures within the breast and allowing for breast lesion detection and assessment in 3D. In order to enable quantitative analysis, such as volumetrics, automated lesion segmentation on bCT is highly desirable. In addition, accurate output from CAD (computer-aided detection/diagnosis) methods depends on sufficient segmentation of lesions. Thus, in this study, we present a 3D lesion segmentation method for breast masses in contrast-enhanced bCT images. The segmentation algorithm follows a two-step approach. First, 3D radial-gradient index segmentation is used to obtain a crude initial contour, which is then refined by a 3D level set-based active contour algorithm. The data set included contrast-enhanced bCT images from 33 patients containing 38 masses (25 malignant, 13 benign). The mass centers served as input to the algorithm. In this study, three criteria for stopping the contour evolution were compared, based on (1) the change of region volume, (2) the average intensity in the segmented region increase at each iteration, and (3) the rate of change of the average intensity inside and outside the segmented region. Lesion segmentation was evaluated by computing the overlap ratio between computer segmentations and manually drawn lesion outlines. For each lesion, the overlap ratio was averaged across coronal, sagittal, and axial planes. The average overlap ratios for the three stopping criteria ranged from 0.66 to 0.68 (dice coefficient of 0.80 to 0.81), indicating that the proposed segmentation procedure is promising for use in quantitative dedicated bCT analyses.

Segmentation of breast masses on dedicated breast computed tomography and three-dimensional breast ultrasound images

Abstract. We present and evaluate a method for the three-dimensional (3-D) segmentation of breast masses on dedicated breast computed tomography (bCT) and automated 3-D breast ultrasound images. The segmentation method, refined from our previous segmentation method for masses on contrast-enhanced bCT, includes two steps: (1) initial contour estimation and (2) active contour-based segmentation to further evolve and refine the initial contour by adding a local energy term to the level-set equation. Segmentation performance was assessed in terms of Dice coefficients (DICE) for 129 lesions on noncontrast bCT, 38 lesions on contrast-enhanced bCT, and 98 lesions on 3-D breast ultrasound (US) images. For bCT, DICE values of 0.82 and 0.80 were obtained on contrast-enhanced and noncontrast images, respectively. The improvement in segmentation performance with respect to that of our previous method was statistically significant (p=0.002). Moreover, segmentation appeared robust with respect to the presence of glandular tissue. For 3-D breast US, the DICE value was 0.71. Hence, our method obtained promising results for both 3-D imaging modalities, laying a solid foundation for further quantitative image analysis and potential future expansion to other 3-D imaging modalities.

Evaluation of a 3D lesion segmentation algorithm on DBT and breast CT images

Recently, tomosynthesis (DBT) and CT (BCT) have been developed for breast imaging. Since each modality produces a fundamentally different representation of the breast volume, our goal was to investigate whether a 3D segmentation algorithm for breast masses could be applied to both DBT and breast BCT images. A secondary goal of this study was to investigate a simplified method for comparing manual outlines to a computer segmentation. The seeded mass lesion segmentation algorithm is based on maximizing the radial gradient index (RGI) along a constrained region contour. In DBT, the constraint function was a prolate spherical Gaussian, with a larger FWHM along the depth direction where the resolution is low, while it was a spherical Gaussian for BCT. For DBT, manual lesion outlines were obtained in the in-focus plane of the lesion, which was used to compute the overlap ratio with the computer segmentation. For BCT, lesions were manually outlined in three orthogonal planes, and the average overlap ratio from the three planes was computed. In DBT, 81% of all lesions were segmented at an overlap ratio of 0.4 or higher, based on manual outlines in one slice through the lesion center. In BCT, 93% of all segmentations achieved an average overlap ratio of 0.4, based on the manual outlines in three orthogonal planes. Our results indicate mass lesions in both BCT and DBT images can be segmented with the proposed 3D segmentation algorithm, by selecting an appropriate set of parameters and after images have undergone specific pre-processing.

Can radiologists recognize that a computer has identified cancers that they have overlooked

For computer-aided detection (CADe) to be effective, the computer must be able to identify cancers that a radiologist misses clinically and the radiologist must be able to recognize that a cancer was missed when he or she reviews the computer output. There are several papers indicating CADe can detected clinically missed cancers. The purpose of this study is to examine whether radiologists can use the CADe output effectively to detect more cancers. Three-hundred mammographic cases, which included current and previous exams, were collected: 66 cases containing a missed cancer that was recognized in retrospect and 234 were normal cases. These were analyzed by a commercial CADe system. An observer study with eight MQSA-qualified radiologists was conducted using a sequential reading method. That is, the radiologist viewed the mammograms and scored the case. Then they reviewed the CADe output and rescored the case. The computer had a sensitivity of 55% with an average of 0.59 false detections per image. For all cancers (n=69), the radiologists had a sensitivity of 58% with no aid and 64% with aid (p=0.002). In cases where the computer detected the cancer in all views that the cancer was visible (n=17), the radiologists had a sensitivity of 74% unaided and increased to 85% aided (p=0.02). In cases where the computer missed the cancer in one view (n=21), the radiologists had a sensitivity of 65% unaided and 72% aided (p<0.001). The radiologists, on average, ignored 20% of all correct computer prompts.

Validation of a power-law noise model for simulating small-scale breast tissue

We have validated a small-scale breast tissue model based on power-law noise. A set of 110 patient images served as truth. The statistical model parameters were determined by matching the radially averaged power-spectrum of the projected simulated tissue with that of the central tomosynthesis patient breast projections. Observer performance in a signal-known exactly detection task in simulated and actual breast backgrounds was compared. Observers included human readers, a pre-whitening observer model and a channelized Hotelling observer model. For all observers, good agreement between performance in the simulated and actual backgrounds was found, both in the tomosynthesis central projections and the reconstructed images. This tissue model can be used for breast x-ray imaging system optimization. The complete statistical description of the model is provided.

Impact of lesion segmentation metrics on computer-aided diagnosis/detection in breast computed tomography

Abstract. Evaluation of segmentation algorithms usually involves comparisons of segmentations to gold-standard delineations without regard to the ultimate medical decision-making task. We compare two segmentation evaluations methods—a Dice similarity coefficient (DSC) evaluation and a diagnostic classification task–based evaluation method using lesions from breast computed tomography. In our investigation, we use results from two previously developed lesion-segmentation algorithms [a global active contour model (GAC) and a global with local aspects active contour model]. Although similar DSC values were obtained (0.80 versus 0.77), we show that the global + local active contour (GLAC) model, as compared with the GAC model, is able to yield significantly improved classification performance in terms of area under the receivers operating characteristic (ROC) curve in the task of distinguishing malignant from benign lesions. [Area under the ROC curve (AUC)=0.78 compared to 0.63, p≪0.001]. This is mainly because the GLAC model yields better detailed information required in the calculation of morphological features. Based on our findings, we conclude that the DSC metric alone is not sufficient for evaluating segmentation lesions in computer-aided diagnosis tasks.

Automatic 3D lesion segmentation on breast ultrasound images

Automatically acquired and reconstructed 3D breast ultrasound images allow radiologists to detect and evaluate breast lesions in 3D. However, assessing potential cancers in 3D ultrasound can be difficult and time consuming. In this study, we evaluate a 3D lesion segmentation method, which we had previously developed for breast CT, and investigate its robustness on lesions on 3D breast ultrasound images. Our dataset includes 98 3D breast ultrasound images obtained on an ABUS system from 55 patients containing 64 cancers. Cancers depicted on 54 US images had been clinically interpreted as negative on screening mammography and 44 had been clinically visible on mammography. All were from women with breast density BI-RADS 3 or 4. Tumor centers and margins were indicated and outlined by radiologists. Initial RGI-eroded contours were automatically calculated and served as input to the active contour segmentation algorithm yielding the final lesion contour. Tumor segmentation was evaluated by determining the overlap ratio (OR) between computer-determined and manually-drawn outlines. Resulting average overlap ratios on coronal, transverse, and sagittal views were 0.60 ± 0.17, 0.57 ± 0.18, and 0.58 ± 0.17, respectively. All OR values were significantly higher the 0.4, which is deemed “acceptable”. Within the groups of mammogram-negative and mammogram-positive cancers, the overlap ratios were 0.63 ± 0.17 and 0.56 ± 0.16, respectively, on the coronal views; with similar results on the other views. The segmentation performance was not found to be correlated to tumor size. Results indicate robustness of the 3D lesion segmentation technique in multi-modality 3D breast imaging.