A phantom for facilitating in vitro computed tomographic studies of gallstones.

A phantom was designed and constructed for in vitro studies of gallstones using a computed tomographic (CT) scanner. A primary objective of the design was to permit studies of multiple gallstones in a single CT slice. This was accomplished by incorporating in the phantom removable compartments that contain vertically oriented, cone-shaped voids for holding the gallstones and surrounding fluid medium. A slice through the center of the phantom passes through the apex of each cone-shaped holder, and hence through the center of each gallstone. The main body of the phantom is made of water-mimicking plastic, and each compartment can accommodate gallstones ranging up to 3 cm in diameter. Initial experience with the phantom has shown it to be a successful design.

Quantitative computed tomography scanning for measurement of bone and bone marrow fat content. A comparison of single- and dual-energy techniques using a solid synthetic phantom.

Quantitative CT (QCT) has become a popular method for estimating bone mineral content. In addition, QCT can be used to estimate the fat content of trabecular bone. Although the latter has received little attention, it may prove to be clinically significant. Using a set of custom-built, tissue-mimicking plastic inserts in an anthropomorphic phantom, we tested a variety of methods for estimating mineral and fat content. We also investigated the influence of patient size, reconstruction circle size, and reference phantom choice on the accuracy of the results. Best estimates were obtained when there was a match between patient and reconstruction circle size. Single-energy methods yielded the best estimates of mineral content for inserts that did not contain fat, and dual-energy methods yielded the best estimates for inserts that contained fat. A dual-energy method that we developed was best in estimating the mineral and fat content of the latter inserts. We found that an external calibration reference phantom containing aqueous solutions of K2HPO4 could be used satisfactorily to estimate the mineral content of trabecular bone mimicking inserts; however, more representative materials must be used for accurate estimates of fat content.

A DPA technique for simultaneously measuring bone, soft tissue, and fat content.

Although conventional dual photon absorptiometry (DPA) techniques can be used to estimate fat and soft tissue content, such estimates are not possible where bone is present. We propose a method that can make these estimates in the presence of bone in the extremities of the body. The combination of this method with the conventional method should yield fat and soft tissue composition from most points within the body. The proposed technique simultaneously measures thicknesses of bone, soft tissue, and fat. These thicknesses are determined of bone, soft tissue, and fat. These thicknesses are determined from a combination of gamma-ray transmission data at two energies and a measurement of total tissue thickness. To test the technique, a feasibility study was performed with known thicknesses of aluminum (simulating bone), lucite (simulating tissue), and polyethylene (simulating fat). A variety of thicknesses of each material were employed (Al: 0-1.3 cm, lucite: 0-5 cm, polyethylene: 0-5 cm). The accuracies (standard errors of the estimates) of the calculated versus true thicknesses of aluminum, lucite, and polyethylene were 0.6%, 2.6%, and 2.5%, respectively. The estimates of "bone" thickness were insensitive to the presence of varying thicknesses of "fat." (In contrast, application of the conventional DPA method to the same gamma-ray transmission data yielded underestimates in "bone" thickness due to "fat" by as much as 11%.) For a 60 minute (whole body) scan time, the reproducibility of the measurements of the thicknesses of aluminum, lucite, and polyethylene were 0.4%, 1.0%, and 1.3%, respectively. All of these values are in a clinically useful range.(ABSTRACT TRUNCATED AT 250 WORDS)

The composition of bone marrow for a dual-energy quantitative computed tomography technique. A cadaver and computer simulation study.

RATIONALE AND OBJECTIVES: The authors have been developing a dual-energy quantitative computed tomography (DE-QCT) technique that requires calibration standards that mimic the x-ray attenuation properties of bone, red marrow, and yellow marrow. To resolve questions regarding the compositions of red and yellow marrow that appear in the literature, the authors performed chemical analyses of bone marrow samples. The newly derived compositions were used in a simulation study to test the accuracy of the DEQCT technique. METHODS: Red marrow samples were extruded from the vertebrae of cadavers of young boys. Yellow marrow samples were removed directly from the femurs of cadavers of elderly women. The fat, protein, water, and mineral contents of these samples were determined. The compositions of 12 mixed marrow samples extruded from cadaver vertebrae also were measured. A computer simulation study was performed in which calibration standards with the new compositions were employed to estimate the fat and bone contents of spongiosas containing the 12 mixed marrows. RESULTS AND CONCLUSIONS: The red marrow samples contained 3% to 6% fat, 6% to 8% protein, 82% to 86% water, and 0.5% to 1% mineral. The yellow marrow samples contained 71% to 92% fat, 1% to 2% protein, 7% to 26% water, and 0.2% to 0.4% mineral. The simulation study yielded good results in three cases and mediocre to poor results in nine cases. An alternative approach was tried in which an average fat-free marrow was derived from the compositions of the 12 mixed marrows, and this substance, fat, and bone were used as the calibration standards. The DEQCT technique with these standards was applied to simulated spongiosas containing the 12 original mixed marrows plus nine additional mixed marrows. All of the estimates were in good agreement with the true compositions. The rms error of the mass fractions of fat was 0.03, and the rms error of the bone concentrations was 3.7 mg/mL.

Two postprocessing CT techniques for determining the composition of trabecular bone.

Two dual-energy CT techniques have been developed to analyze the mineral and fat content of trabecular bone. Both are postprocessing techniques that employ calibration standards. Experiments were performed to test these techniques against conventional single-energy techniques and two other dual-energy techniques. As expected, all of the dual-energy methods estimate the mineral content more accurately when fat is present. In contrast to the other dual-energy methods, the new methods described in this article are unique because they make a separate estimate of the fat content of the bone. The results of preliminary tests of these techniques in estimating fat content have been encouraging. Although not exact, the estimates show the correct trend in increasing proportionately as the fat content increases. Possible applications of the techniques in the study of osteoporosis and other bone diseases are described.

Evaluation of a new set of calibration standards for the measurement of fat content via DPA and DXA.

A simulation study was performed to evaluate a new set of calibration standards for estimating the fat content of the body via dual-photon absorptiometry (DPA) and dual-energy x-ray absorptiometry (DXA). The standards, proposed by Nord and Payne [presented at the 2nd meeting of The Bath Conference on Bone Mineral Measurement (1990)] consist of stearic acid (100% fat) and 0.6% NaCl in water (100% lean). They were compared with other standards consisting of average composition adipose/muscle tissues and fatty adipose/lean muscle tissues. Source and detector properties of a Gd-153 DPA system and three commercial DXA systems were modeled. For each system and calibration set, rms errors in the calculated fat contents of simulated tissues having fat mass percentages that ranged from about 4%-44% and thicknesses that ranged from 5-20 cm were determined. Beam hardening errors for the systems were evaluated as was a calibration technique employed by one of the manufacturers to correct for such errors. In general, the smallest rms errors (2% or less when the calibration standards and tissues were of equal thickness) were obtained with the average adipose/muscle standards. Equivalent results were obtained with standards consisting of stearic acid and 0.8% NaCl. The latter is a higher salt content than proposed by Nord and Payne and results from differences in the x-ray attenuation coefficients that were employed in calculating the fat equivalence of water. Other, more convenient standards, such as lucite and water may be employed by using appropriate fat equivalences (approximately 69% for lucite and approximately 10% for water). Beam hardening errors for the DXA systems are considerable, and the simulated correction technique was shown to be effective.

Beam hardening errors in post-processing dual energy quantitative computed tomography.

A computer simulation study was performed to assess the errors due to x-ray beam hardening in the fat and bone estimates of a post-processing dual-energy quantitative computed tomography technique. The "central" calibration method was employed in which calibration standards are inserted within a torso phantom of a size similar to that of the "patient." Although beam hardening errors are reduced with this method, they still occur as a result of mismatches between the torso phantom and patient body sizes. Two mismatch situations were investigated. In one, a single torso phantom was used for all subject sizes (i.e., one-size-fits-all). In the other, closest matches were made from a set of three different sized torso phantoms (small, medium, and large). In all cases, the compositions of the calibration standards that were inserted into the torso phantoms consisted of bone, fat (glycerol trioleate), and an average fat-free red marrow. Fifteen patient sizes were simulated ranging from 20 to 34 cm in diameter. There were 21 patients of each size. The vertebrae in these subjects contained known amounts of bone mixed in marrows of composition determined from chemical analyses of cadaver marrow samples. Vertebrae consisting of mixtures of the calibration standard materials were also studied. The computed effective x-ray beam energies at the vertebra location for the various subject sizes ranged from 54.3 to 56.4 keV at 80 kVp and from 74.4 to 78.8 keV at 140 kVp.(ABSTRACT TRUNCATED AT 250 WORDS)

Precision in quantitative CT: impact of x-ray dose and matrix size.

A study was performed to determine whether recommended technique factors for postprocessing dual-energy (DE) quantitative computed tomography are optimum in terms of precision and x-ray dose. In particular, possible dose reduction as a result of an upgrade of the CT scanner to more efficient detectors was explored. Series of images of an anthropomorphic phantom containing a human vertebra, a tissue-simulating lumbar simulator with various marrow inserts, and a polyethylene cylinder were generated. Recommended DE x-ray technique factors as well as factors resulting in about two times, one-half, and one-fourth the x-ray dose were employed. The effects of reconstruction with different matrix sizes was studied. Standard deviations of the CT numbers within regions of interest in individual images (noise) and standard deviations of mean CT numbers, single-energy (SE), and DE measurements for series of images (reproducibilities) were computed. It was found that the low-energy component of the DE technique was optimum, but the high-energy component could be reduced by a factor of 2 with negligible loss in precision. This translates into a dose reduction of 36% relative to the recommended DE technique. Vertebral inhomogeneities were responsible for more than 65% of the standard deviations in individual images of the vertebra even at the lowest doses. For all of the techniques, the noise in images of all objects decreased as the x-ray dose increased and as the matrix size decreased. Reproducibility of mean values, however, did not necessarily improve, and aberrant results such as improvement in reproducibility with a reduction in dose were sometimes observed. It is hypothesized that this may be due to variations in the actual kVp for each image in a series.

The effect of fat on the coherent-to-Compton scattering ratio in the calcaneus: a computational analysis.

The coherent-to-Compton scattering ratio (CCSR) is a technique that has been proposed for measuring trabecular bone mineral density (TBMD). This paper investigates the effect of fat on the CCSR and its correlation to the error in TBMD measurements. It is a computational study to determine the relationship between the magnitude of fat error and the momentum-transfer variable chi, which represents the incident photon energy and the scattering angle. Variation in fat content contributes significantly to the error in CCSR measurements. When employing a typical 241Am source (E gamma = 59.45 keV), the resulting error decreases with increasing momentum-transfer variable or angle. For example, the error ranges from +14 mg/cc at an angle of 45 degrees (chi = 18.3) to +3 mg/cc at an angle of 135 degrees (chi = 44.3) for an osteoporotic trabecular region (100 mg/cc mineral) of a calcaneus that contains 6% less fat than a calibration standard. The error is about 0.3-1.2 mg/cc less for regions containing 2-3X more bone mineral and is reduced and opposite in sign for regions containing about 7% more fat than the calibration standards (e.g., -9 mg/cc at 45 degrees and -1.5 mg/cc at 135 degrees). Others have shown that the intrinsic sensitivity of the CCSR method for measuring TBMD at a given photon energy generally increases with increasing detector angle. Thus large angles are advantageous both for reduced sensitivity to fat variation and increased sensitivity to bone mineral variation. The primary disadvantage is reduced count rates that degrade precision unless long counting lines are employed.(ABSTRACT TRUNCATED AT 250 WORDS)

Stereomammography: evaluation of depth perception using a virtual 3D cursor.

We are evaluating the usefulness of stereomammography in improving breast cancer diagnosis. One area that we are investigating is whether the improved depth perception associated with stereomammography might be significantly enhanced with the use of a virtual 3D cursor. A study was performed to evaluate the accuracy of absolute depth measurements made in stereomammograms with such a cursor. A biopsy unit was used to produce digital stereo images of a phantom containing 50 low contrast fibrils (0.5 mm diam monofilaments) at depths ranging from 1 to 11 mm, with a minimum spacing of 2 mm. Half of the fibrils were oriented perpendicular (vertical) and half parallel (horizontal) to the stereo shift direction. The depth and orientation of each fibril were randomized, and the horizontal and vertical fibrils crossed, simulating overlapping structures in a breast image. Left and right eye images were generated by shifting the x-ray tube from +2.5 degrees to -2.5 degrees relative to the image receptor. Three observers viewed these images on a computer display with stereo glasses and adjusted the position of a cross-shaped virtual cursor to best match the perceived location of each fibril. The x, y, and z positions of the cursor were indicated on the display. The z (depth) coordinate was separately calibrated using known positions of fibrils in the phantom. The observers analyzed images of two configurations of the phantom. Thus, each observer made 50 vertical filament depth measurements and 50 horizontal filament depth measurements. These measurements were compared with the true depths. The correlation coefficients between the measured and true depths of the vertically oriented fibrils for the three observers were 0.99, 0.97, and 0.89 with standard errors of the estimates of 0.39 mm, 0.83 mm, and 1.33 mm, respectively. Corresponding values for the horizontally oriented fibrils were 0.91, 0.28, and 0.08, and 1.87 mm, 4.19 mm, and 3.13 mm. All observers could estimate the absolute depths of vertically oriented objects fairly accurately in digital stereomammograms; however, only one observer was able to accurately estimate the depths of horizontally oriented objects. This may relate to different aptitudes for stereoscopic visualization. The orientations of most objects in actual mammograms are combinations of horizontal and vertical. Further studies are planned to evaluate absolute depth measurements of fibrils oriented at various intermediate angles and of objects of different shapes. The effects of the shape and contrast of the virtual cursor and the stereo shift angle on the accuracy of the depth measurements will also be investigated.

Investigation of the line-pair pattern method for evaluating mammographic focal spot performance.

The latest American College of Radiology (ACR) Mammography Quality Control Manual contains a new method for evaluating focal spot performance, which this paper refers to as the "line-pair pattern test." The ACR describes a variety of methods for performing this test, and does not advocate one method over another. The authors of this paper conducted an investigation to compare the optional ways for performing the test. Resolution measurements were obtained using a prototype line-pair resolution phantom imaged with a GE DMR mammography unit. Measurements were made with the line-pair pattern 4.5 cm above the breast support platforms in both conventional (contact) and magnification geometries. Both 4.5 cm of air and Lucite were tested as attenuators between the line-pair pattern and the breast support platform. Image receptors that were employed included film alone, screen-film, and screen-film that was not allowed to wait the recommended 15 min before exposure. kVp was varied as was the orientation of the line-pair pattern relative to the chest wall. For the air attenuator case, the screen degraded the measured resolution by 1-3 lp/mm when compared to the direct film. The Lucite attenuator reduced the resolution by an additional 1 1p/mm. Increasing kVp improved the resolution slightly for the conventional mode, but decreased it slightly for the magnification mode. Based upon the results of this study, recommendations are made for improving the test protocol. For a test of focal spot performance, one should use the no-attenuation with direct film detector setup. For a measure of the resolution of the entire imaging chain, one should use the Lucite attenuator with screen-film detector setup.

Design and evaluation of an external filter technique for exposure equalization in mammography.

We are developing an external filter method for equalizing x-ray exposure in the peripheral region of the breast. This method requires the use of only a limited number of custom-built filters for different breast shapes in a given view. This paper describes the design methodology for these external filters. The filter effectiveness was evaluated through a simulation study on 171 mediolateral and 196 craniocaudal view digitized mammograms and through imaging of a breast phantom. The degree of match between the simulated filter and the individual 3-D exposure profiles at the breast periphery was quantified. An analysis was performed to investigate the effect of filter misalignment. The simulation study indicates that the filter is effective in equalizing exposures for more than 80% of the breast images in our database. The tolerance in filter misalignment was estimated to be about +/- 2 mm for the CC view and +/- 1 mm for the MLO view at the image plane. Some misalignment artifacts were demonstrated with simulated filtered mammograms.

Computerized characterization of masses on mammograms: the rubber band straightening transform and texture analysis.

A new rubber band straightening transform (RBST) is introduced for characterization of mammographic masses as malignant or benign. The RBST transforms a band of pixels surrounding a segmented mass onto the Cartesian plane (the RBST image). The border of a mammographic mass appears approximately as a horizontal line, and possible speculations resemble vertical lines in the RBST image. In this study, the effectiveness of a set of directional textures extracted from the images before the RBST. A database of 168 mammograms containing biopsy-proven malignant and benign breast masses was digitized at a pixel size of 100 microns x 100 microns. Regions of interest (ROIs) containing the biopsied mass were extracted from each mammogram by an experienced radiologist. A clustering algorithm was employed for automated segmentation of each ROI into a mass object and background tissue. Texture features extracted from spatial gray-level dependence matrices and run-length statistics matrices were evaluated for three different regions and representations: (i) the entire ROI; (ii) a band of pixels surrounding the segmented mass object in the ROI; and (iii) the RBST image. Linear discriminant analysis was used for classification, and receiver operating characteristic (ROC) analysis was used to evaluate the classification accuracy. Using the ROC curves as the performance measure, features extracted from the RBST images were found to be significantly more effective than those extracted from the original images. Features extracted from the RBST images yielded an area (Az) of 0.94 under the ROC curve for classification of mammographic masses as malignant and benign.

Computerized image analysis: estimation of breast density on mammograms.

An automated image analysis tool is being developed for the estimation of mammographic breast density. This tool may be useful for risk estimation or for monitoring breast density change in prevention or intervention programs. In this preliminary study, a data set of 4-view mammograms from 65 patients was used to evaluate our approach. Breast density analysis was performed on the digitized mammograms in three stages. First, the breast region was segmented from the surrounding background by an automated breast boundary-tracking algorithm. Second, an adaptive dynamic range compression technique was applied to the breast image to reduce the range of the gray level distribution in the low frequency background and to enhance the differences in the characteristic features of the gray level histogram for breasts of different densities. Third, rule-based classification was used to classify the breast images into four classes according to the characteristic features of their gray level histogram. For each image, a gray level threshold was automatically determined to segment the dense tissue from the breast region. The area of segmented dense tissue as a percentage of the breast area was then estimated. To evaluate the performance of the algorithm, the computer segmentation results were compared to manual segmentation with interactive thresholding by five radiologists. A "true" percent dense area for each mammogram was obtained by averaging the manually segmented areas of the radiologists. We found that the histograms of 6% (8 CC and 8 MLO views) of the breast regions were misclassified by the computer, resulting in poor segmentation of the dense region. For the images with correct classification, the correlation between the computer-estimated percent dense area and the "truth" was 0.94 and 0.91, respectively, for CC and MLO views, with a mean bias of less than 2%. The mean biases of the five radiologists' visual estimates for the same images ranged from 0.1% to 11%. The results demonstrate the feasibility of estimating mammographic breast density using computer vision techniques and its potential to improve the accuracy and reproducibility of breast density estimation in comparison with the subjective visual assessment by radiologists.

False-positive reduction technique for detection of masses on digital mammograms: global and local multiresolution texture analysis.

We investigated the application of multiresolution global and local texture features to reduce false-positive detection in a computerized mass detection program. One hundred and sixty-eight digitized mammograms were randomly and equally divided into training and test groups. From these mammograms, two datasets were formed. The first dataset (manual) contained four regions of interest (ROIs) selected manually from each of the mammograms. One of the four ROIs contained a biopsy-proven mass and the other three contained normal parenchyma, including dense, mixed dense/fatty, and fatty tissues. The second dataset (hybrid) contained the manually extracted mass ROIs, along with normal tissue ROIs extracted by an automated Density-Weighted Contrast Enhancement (DWCE) algorithm as false-positive detections. A wavelet transform was used to decompose an ROI into several scales. Global texture features were derived from the low-pass coefficients in the wavelet transformed images. Local texture features were calculated from the suspicious object and the peripheral subregions. Linear discriminant models using effective features selected from the global, local, or combined feature spaces were established to maximize the separation between masses and normal tissue. Receiver Operating Characteristic (ROC) analysis was conducted to evaluate the classifier performance. The classification accuracy using global features were comparable to that using local features. With both global and local features, the average area, Az, under the test ROC curve, reached 0.92 for the manual dataset and 0.96 for the hybrid dataset, demonstrating statistically significant improvement over those obtained with global or local features alone. The results indicated the effectiveness of the combined global and local features in the classification of masses and normal tissue for false-positive reduction.

Image feature selection by a genetic algorithm: application to classification of mass and normal breast tissue.

We investigated a new approach to feature selection, and demonstrated its application in the task of differentiating regions of interest (ROIs) on mammograms as either mass or normal tissue. The classifier included a genetic algorithm (GA) for image feature selection, and a linear discriminant classifier or a backpropagation neural network (BPN) for formulation of the classifier outputs. The GA-based feature selection was guided by higher probabilities of survival for fitter combinations of features, where the fitness measure was the area Az under the receiver operating characteristic (ROC) curve. We studied the effect of different GA parameters on classification accuracy, and compared the results to those obtained with stepwise feature selection. The data set used in this study consisted of 168 ROIs containing biopsy-proven masses and 504 ROIs containing normal tissue. From each ROI, a total of 587 features were extracted, of which 572 were texture features and 15 were morphological features. The GA was trained and tested with several different partitionings of the ROIs into training and testing sets. With the best combination of the GA parameters, the average test Az value using a linear discriminant classifier reached 0.90, as compared to 0.89 for stepwise feature selection. Test Az values with a BPN classifier and a more limited feature pool were 0.90 with GA-based feature selection, and 0.89 for stepwise feature selection. The use of a GA in tailoring classifiers with specific design characteristics was also discussed. This study indicates that a GA can provide versatility in the design of linear or nonlinear classifiers without a trade-off in the effectiveness of the selected features.

Classification of mass and normal breast tissue on digital mammograms: multiresolution texture analysis.

We investigated the feasibility of using multiresolution texture analysis for differentiation of masses from normal breast tissue on mammograms. The wavelet transform was used to decompose regions of interest (ROIs) on digitized mammograms into several scales. Multiresolution texture features were calculated from the spatial gray level dependence matrices of (1) the original images at variable distances between the pixel pairs, (2) the wavelet coefficients at different scales, and (3) the wavelet coefficients up to certain scale and then at variable distances between the pixel pairs. In this study, 168 ROIs containing biopsy-proven masses and 504 ROIs containing normal parenchyma were used as the data set. The mass ROIs were randomly and equally divided into training and test groups along with corresponding normal ROIs from the same film. Stepwise linear discriminant analysis was used to select optimal features from the multiresolution texture feature space to maximize the separation of mass and normal tissue for all ROIs. We found that texture features at large pixel distances are important for the classification task. The wavelet transform can effectively condense the image information into its coefficients. With texture features based on the wavelet coefficients and variable distances, the area Az under the receiver operating characteristic curve reached 0.89 and 0.86 for the training and test groups, respectively. The results demonstrate that a linear discriminant classifier using the multiresolution texture features can effectively classify masses from normal tissue on mammograms.

Classification of compressed breast shapes for the design of equalization filters in x-ray mammography.

We are developing an external filter method for equalizing the x-ray exposure in mammography. Each filter is specially designed to match the shape of the compressed breast border and to preferentially attenuate the x-ray beam in the peripheral region of the breast. To be practical, this method should require the use of only a limited number of custom built filters. It is hypothesized that this would be possible if compressed breasts can be classified into a finite number of shapes. A study was performed to determine the number of shapes. Based on the parabolic appearance of the outer borders of compressed breasts in mammograms, the borders were fit with the polynomial equations y = ax2 + bx3 and y = ax2 + bx3 + cx4. The goodness-of-fit of these equations was compared. The a,b and a,b,c coefficients were employed in a K-Means clustering procedure to classify 470 CC-view and 484 MLO-view borders into 2-10 clusters. The mean coefficients of the borders within a given cluster defined the "filter" shape, and the individual borders were translated and rotated to best match that filter shape. The average rms differences between the individual borders and the "filter" were computed as were the standard deviations of those differences. The optimally shifted and rotated borders were refit with the above polynomial equations, and plotted for visual evaluation of clustering success. Both polynomial fits were adequate with rms errors of about 2 mm for the 2-coefficient equation, and about 1 mm for the 3-coefficient equation. Although the fits to the original borders were superior for the 3-coefficient equation, the matches to the "filter" borders determined by clustering were not significantly improved. A variety of modified clustering methods were developed and utilized, but none produced major improvements in clustering. Results indicate that 3 or 4 filter shapes may be adequate for each mammographic projection (CC- and MLO-view). To account for the wide variations in exposures observed at the peripheral regions of breasts classified to be of a particular shape, it may be necessary to employ different filters for thin, medium and thick breasts. Even with this added requirement, it should be possible to use a small number of filters as desired.

Computerized analysis of mammographic microcalcifications in morphological and texture feature spaces.

We are developing computerized feature extraction and classification methods to analyze malignant and benign microcalcifications on digitized mammograms. Morphological features that described the size, contrast, and shape of microcalcifications and their variations within a cluster were designed to characterize microcalcifications segmented from the mammographic background. Texture features were derived from the spatial gray-level dependence (SGLD) matrices constructed at multiple distances and directions from tissue regions containing microcalcifications. A genetic algorithm (GA) based feature selection technique was used to select the best feature subset from the multi-dimensional feature spaces. The GA-based method was compared to the commonly used feature selection method based on the stepwise linear discriminant analysis (LDA) procedure. Linear discriminant classifiers using the selected features as input predictor variables were formulated for the classification task. The discriminant scores output from the classifiers were analyzed by receiver operating characteristic (ROC) methodology and the classification accuracy was quantified by the area, Az, under the ROC curve. We analyzed a data set of 145 mammographic microcalcification clusters in this study. It was found that the feature subsets selected by the GA-based method are comparable to or slightly better than those selected by the stepwise LDA method. The texture features (Az = 0.84) were more effective than morphological features (Az = 0.79) in distinguishing malignant and benign microcalcifications. The highest classification accuracy (Az = 0.89) was obtained in the combined texture and morphological feature space. The improvement was statistically significant in comparison to classification in either the morphological (p = 0.002) or the texture (p = 0.04) feature space alone. The classifier using the best feature subset from the combined feature space and an appropriate decision threshold could correctly identify 35% of the benign clusters without missing a malignant cluster. When the average discriminant score from all views of the same cluster was used for classification, the Az value increased to 0.93 and the classifier could identify 50% of the benign clusters at 100% sensitivity for malignancy. Alternatively, if the minimum discriminant score from all views of the same cluster was used, the Az value would be 0.90 and a specificity of 32% would be obtained at 100% sensitivity. The results of this study indicate the potential of using combined morphological and texture features for computer-aided classification of microcalcifications.

Classification of mass and normal breast tissue: a convolution neural network classifier with spatial domain and texture images.

The authors investigated the classification of regions of interest (ROI's) on mammograms as either mass or normal tissue using a convolution neural network (CNN). A CNN is a backpropagation neural network with two-dimensional (2-D) weight kernels that operate on images. A generalized, fast and stable implementation of the CNN was developed. The input images to the CNN were obtained from the ROI's using two techniques. The first technique employed averaging and subsampling. The second technique employed texture feature extraction methods applied to small subregions inside the ROI. Features computed over different subregions were arranged as texture images, which were subsequently used as CNN inputs. The effects of CNN architecture and texture feature parameters on classification accuracy were studied. Receiver operating characteristic (ROC) methodology was used to evaluate the classification accuracy. A data set consisting of 168 ROIs containing biopsy-proven masses and 504 ROI's containing normal breast tissue was extracted from 168 mammograms by radiologists experienced in mammography. This data set was used for training and testing the CNN. With the best combination of CNN architecture and texture feature parameters, the area under the test ROC curve reached 0.87, which corresponded to a true-positive fraction of 90% at a false positive fraction of 31%. The authors' results demonstrate the feasibility of using a CNN for classification of masses and normal tissue on mammograms.