Carl J. Vyborny

Image feature analysis and computer‐aided diagnosis in digital radiography. I. Automated detection of microcalcifications in mammography

We have investigated the application of computer-based methods to the detection of microcalcifications in digital mammograms. The computer detection system is based on a difference-image technique in which a signal-suppressed image is subtracted from a signal-enhanced image to remove the structured background in a mammogram. Signal-extraction techniques adapted to the known physical characteristics of microcalcifications are then used to isolate microcalcifications from the remaining noise background. We employ Monte Carlo methods to generate simulated clusters of microcalcifications that are superimposed on normal mammographic backgrounds. This allows quantitative evaluation of detection accuracy of the computer method and the dependence of this accuracy on the physical characteristics of the microcalcifications. Our present computer method can achieve a true-positive cluster detection rate of approximately 80% at a false-positive detection rate of one cluster per image. The potential application of such a computer-aided system to mammographic interpretation is demonstrated by its ability to detect microcalcifications in clinical mammograms.

Computerized detection of masses in digital mammograms: Analysis of bilateral subtraction images

A computerized scheme is being developed for the detection of masses in digital mammograms. Based on the deviation from the normal architectural symmetry of the right and left breasts, a bilateral subtraction technique is used to enhance the conspicuity of possible masses. The scheme employs two pairs of conventional screen-film mammograms (the right and left mediolateral oblique views and craniocaudal views), which are digitized by a TV camera/Gould digitizer. The right and left breast images in each pair are aligned manually during digitization. A nonlinear bilateral subtraction technique that involves linking multiple subtracted images has been investigated and compared to a simple linear subtraction method. Various feature-extraction techniques are used to reduce false-positive detections resulting from the bilateral subtraction. The scheme has been evaluated using 46 pairs of clinical mammograms and was found to yield a 95% true-positive rate at an average of three false-positive detections per image. This preliminary study indicates that the scheme is potentially useful as an aid to radiologists in the interpretation of screening mammograms.

Analysis of spiculation in the computerized classification of mammographic masses

Spiculation is a primary sign of malignancy for masses detected by mammography. In this study, we developed a technique that analyzes patterns and quantifies the degree of spiculation present. Our current approach involves (1) automatic lesion extraction using region growing and (2) feature extraction using radial edge-gradient analysis. Two spiculation measures are obtained from an analysis of radial edge gradients. These measures are evaluated in four different neighborhoods about the extracted mammographic mass. The performance of each of the two measures of spiculation was tested on a database of 95 mammographic masses using ROC analysis that evaluates their individual ability to determine the likelihood of malignancy of a mass. The dependence of the performance of these measures on the choice of neighborhood was analyzed. We have found that it is only necessary to accurately extract an approximate outline of a mass lesion for the purposes of this analysis since the choice of a neighborhood that accommodates the thin spicules at the margin allows for the assessment of margin spiculation with the radial edge-gradient analysis technique. The two measures performed at their highest level when the surrounding periphery of the extracted region is used for feature extraction, yielding Az values of 0.83 and 0.85, respectively, for the determination of malignancy. These are similar to that achieved when a radiologists ratings of spiculation (Az = 0.85) are used alone. The maximum value of one of the two spiculation measures (FWHM) from the four neighborhoods yielded an Az of 0.88 in the classification of mammographic mass lesions.

Computerized diagnosis of breast lesions on ultrasound

We present a computer-aided diagnosis (CAD) method for breast lesions on ultrasound that is based on the automatic segmentation of lesions and the automatic extraction of four features related to the lesion shape, margin, texture, and posterior acoustic behavior. Using a database of 400 cases (94 malignant lesions, 124 complex cysts, and 182 benign solid lesions), we investigate the marginal benefit of each feature in our CAD method and the performance of our CAD method in distinguishing malignant lesions from various classes of benign lesions. Finally, independent validation is performed on our CAD method. Eleven independent trials yielded an average Az value of 0.87 in the task of distinguishing malignant from benign lesions.

Automated computerized classification of malignant and benign masses on digitized mammograms

RATIONALE AND OBJECTIVES To develop a method for differentiating malignant from benign masses in which a computer automatically extracts lesion features and merges them into an estimated likelihood of malignancy. MATERIALS AND METHODS Ninety-five mammograms depicting masses in 65 patients were digitized. Various features related to the margin and density of each mass were extracted automatically from the neighborhoods of the computer-identified mass regions. Selected features were merged into an estimated likelihood of malignancy by, using three different automated classifiers. The performance of the three classifiers in distinguishing between benign and malignant masses was evaluated by receiver operating characteristic analysis and compared with the performance of an experienced mammographer and that of five less experienced mammographers. RESULTS Our computer classification scheme yielded an area under the receiver operating characteristic curve (Az) value of 0.94, which was similar to that for an experienced mammographer (Az = 0.91) and was statistically significantly higher than the average performance of the radiologists with less mammographic experience (Az = 0.81) (P = .013). With the database used, the computer scheme achieved, at 100% sensitivity, a positive predictive value of 83%, which was 12% higher than that for the performance of the experienced mammographer and 21% higher than that for the average performance of the less experienced mammographers (P < .0001). CONCLUSION Automated computerized classification schemes may be useful in helping radiologists distinguish between benign and malignant masses and thus reducing the number of unnecessary biopsies.

Computerized lesion detection on breast ultrasound

We investigated the use of a radial gradient index (RGI) filtering technique to automatically detect lesions on breast ultrasound. After initial RGI filtering, a sensitivity of 87% at 0.76 false-positive detections per image was obtained on a database of 400 patients (757 images). Next, lesion candidates were segmented from the background by maximizing an average radial gradient (ARD) index for regions grown from the detected points. At an overlap of 0.4 with a radiologist lesion outline, 75% of the lesions were correctly detected. Subsequently, round robin analysis was used to assess the quality of the classification of lesion candidates into actual lesions and false-positives by a Bayesian neural network. The round robin analysis yielded an Az value of 0.84, and an overall performance by case of 94% sensitivity at 0.48 false-positives per image. Use of computerized analysis of breast sonograms may ultimately facilitate the use of sonography in breast cancer screening programs.

Effect of case selection on the performance of computer‐aided detection schemes

The choice of clinical cases used to train and test a computer-aided diagnosis (CAD) scheme can affect the test results (i.e., error rate). In this study, we deliberately modified the components of our testing database to study the effects of this modification on measured performance. Using a computerized scheme for the automated detection of breast masses from mammograms, it was found that the sensitivity of the scheme ranged between 26% and 100% (at a false positive rate of 1.0 per image) depending on the cases used to test the scheme. Even a 20% change in the cases comprising the database can reduce the measured sensitivity by 15%-25%. Because of the strong dependence of measured performance on the testing database, it is difficult to estimate reliably the accuracy of a CAD scheme. Furthermore, it is questionable to compare different CAD schemes when different cases are used for testing. Sharing databases, creating a common database, or using a quantitative measure to characterize databases are possible solutions to this problem. However, none of these solutions exists or is practiced at present. Therefore, as a short-term solution, it is recommended that the method used for selecting cases, and histograms or mean and standard deviations of relevant image features be reported whenever performance data are presented.

Automatic segmentation of breast lesions on ultrasound

In this paper we present a computationally efficient segmentation algorithm for breast masses on sonography that is based on maximizing a utility function over partition margins defined through gray-value thresholding of a preprocessed image. The performance of the segmentation algorithm is evaluated on a database of 400 cases in two ways. Of the 400 cases, 124 were complex cysts, 182 were benign solid lesions, and 94 were malignant lesions. In the first evaluation, the computer-delineated margins were compared to manually delineated margins. At an overlap threshold of 0.40, the segmentation algorithm correctly delineated 94% of the lesions. In the second evaluation, the performance of our computer-aided diagnosis method on the computer-delineated margins was compared to the performance of our method on the manually delineated margins. Round robin evaluation yielded Az values of 0.90 and 0.87 on the manually delineated margins and the computer-delineated margins, respectively, in the task of distinguishing between malignant and nonmalignant lesions.

COMPUTERIZED DETECTION OF MASSES IN DIGITAL MAMMOGRAMS : AUTOMATED ALIGNMENT OF BREAST IMAGES AND ITS EFFECT ON BILATERAL-SUBTRACTION TECHNIQUE

An automated technique for the alignment of right and left breast images has been developed for use in the computerized analysis of bilateral breast images. In this technique, the breast region is first identified in each digital mammogram by use of histogram analysis and morphological filtering operations. The anterior portions of the tracked breast border and computer-identified nipple positions are selected as landmarks for use in image registration. The paired right and left breast images, either from mediolateral oblique or craniocaudal views, are then registered relative to each other by use of a least-squares matching method. This automated alignment technique has been applied to our computerized detection scheme that employs a nonlinear bilateral-subtraction method for the initial identification of possible masses. The effectiveness of using bilateral subtraction in identifying asymmetries between corresponding right and left breast images is examined by comparing detection performances obtained with various computer-simulated misalignments of 40 pairs of clinical mammograms. Based on free-response receiver operating characteristic and regression analyses, the detection performance obtained with the automated alignment technique was found to be higher than that obtained with simulated misalignments. Detection performance decreased gradually as the amount of simulated misalignment increased. These results indicate that automatic alignment of breast images is possible and that mass-detection performance appears to improve with the inclusion of asymmetric anatomic information but is not sensitive to slight misalignment.

Digital mammography: ROC studies of the effects of pixel size and unsharp-mask filtering on the detection of subtle microcalcifications

We investigated the spatial resolution requirement and the effect of unsharp-mask filtering on the detectability of subtle microcalcifications in digital mammography. Digital images were obtained by digitizing conventional screen-film mammograms with a 0.1 X 0.1 mm2 pixel size, processed with unsharp masking, and then reconstituted on film with a Fuji image processing/simulation system (Fuji Photo Film Co., Tokyo, Japan). Twenty normal cases and 12 cases with subtle microcalcifications were included. Observer performance experiments were conducted to assess the detectability of subtle microcalcifications in the conventional, the unprocessed digital, and the unsharp-masked mammograms. The observer response data were evaluated using receiver operating characteristic (ROC) and LROC (ROC with localization) analyses. Our results indicate that digital mammograms obtained with 0.1 X 0.1 mm2 pixels provide lower detectability than the conventional screen-film mammograms. The detectability of microcalcifications in the digital mammograms is improved by unsharp-mask filtering; the processed mammograms still provide lower accuracy than the conventional mammograms, however, chiefly because of increased false-positive detection rates for the processed images at each subjective confidence level. Viewing unprocessed digital and unsharp-masked images in pairs resulted in approximately the same detectability as that obtained with the unsharp-masked images alone. However, this result may be influenced by the fact that the same limited viewing time was necessarily divided between the two images.