Knowledge-based computer-aided detection of masses on digitized mammograms: a preliminary assessment.

The purpose of this work was to develop and evaluate a computer-aided detection (CAD) scheme for the improvement of mass identification on digitized mammograms using a knowledge-based approach. Three hundred pathologically verified masses and 300 negative, but suspicious, regions, as initially identified by a rule-based CAD scheme, were randomly selected from a large clinical database for development purposes. In addition, 500 different positive and 500 negative regions were used to test the scheme. This suspicious region pruning scheme includes a learning process to establish a knowledge base that is then used to determine whether a previously identified suspicious region is likely to depict a true mass. This is accomplished by quantitatively characterizing the set of known masses, measuring "similarity" between a suspicious region and a "known" mass, then deriving a composite "likelihood" measure based on all "known" masses to determine the state of the suspicious region. To assess the performance of this method, receiver-operating characteristic (ROC) analyses were employed. Using a leave-one-out validation method with the development set of 600 regions, the knowledge-based CAD scheme achieved an area under the ROC curve of 0.83. Fifty-one percent of the previously identified false-positive regions were eliminated, while maintaining 90% sensitivity. During testing of the 1,000 independent regions, an area under the ROC curve as high as 0.80 was achieved. Knowledge-based approaches can yield a significant reduction in false-positive detections while maintaining reasonable sensitivity. This approach has the potential of improving the performance of other rule-based CAD schemes.

Performance gain in computer-assisted detection schemes by averaging scores generated from artificial neural networks with adaptive filtering.

The authors investigated a new method to optimize artificial neural networks (ANNs) with adaptive filtering used in computer-assisted detection schemes in digitized mammograms and to assess performance changes when averaging classification scores from three sets of optimized schemes. Two independent training and testing image databases involving 978 and 830 digitized mammograms, respectively, were used in this study. In the training data set, initial filtering and subtraction resulted in the identification of 592 mass regions and 3790 suspicious, but actually negative regions. These regions (including both true-positive and negative regions) were segmented into three subsets three times based on the calculation of the values of three features as segmentation indices. The indices were "mass" size multiplied by their digital value contrast, conspicuity, and circularity. Nine ANN-based classifiers were separately optimized using a genetic algorithm for each subset of regions. Each region was assigned three classification scores after applying the three adaptive ANNs. The performance gain of the CAD scheme after averaging the three scores for each suspicious region was tested using an independent data set and a ROC methodology. The experimental results showed that the areas under ROC curves (Az) for the testing database using three sets of optimized ANNs individually were 0.84+/-0.01, 0.83+/-0.01, and 0.84+/-0.01, respectively. The between-index correlations of three A values were 0.013, -0.007, and 0.086. Similar to averaging diagnostic ratings from independent observers, by averaging three ANN-generated scores for each testing region, the performance of the CAD scheme was significantly improved (p<0.001) with Az value of 0.95+/-0.01.

Detection of masses and clustered microcalcifications on data compressed mammograms: an observer performance study.

OBJECTIVE: To evaluate observers' ability to detect breast masses and clustered microcalcifications depicted on data compressed mammograms, an observer performance study was performed. MATERIALS AND METHODS: Eight observers assessed 60 mammographic images obtained in six modes, ranging from noncompressed to a maximum data compression level of 101:1. Observers were asked to rate the images on a scale of 0 to 100 for the likelihood of the presence of a mass and also independently for the likelihood of the presence of clustered microcalcifications. In addition, observers were asked to rate their subjective assessment of the quality of each image for the detection of a mass and separately for the detection of microcalcifications. Receiver operating characteristic analyses were performed. RESULTS: The average area under the receiver operating characteristic curve, A(z), for the detection of clustered microcalcifications decreases significantly at the highest data compression level when compared with the noncompressed and two lowest levels of data compression (p < 0.01), and a trend test of the average area under the receiver operating characteristic curve for all observers is statistically significant (p < 0.05). No statistically significant differences among or between any of the data compression level modes for the detection of masses were detected. CONCLUSION: At a high level of mammogram data compression, observer performance was degraded for the detection of clustered microcalcifications. Detection of masses was not affected by the data compression methods and levels used in this study.

Observer variation and the performance accuracy gained by averaging ratings of abnormality.

Six radiologists used continuous scales to rate 529 chest-film cases for likelihood of five different types of abnormalities (interstitial disease, nodule, pneumothorax, alveolar infiltrate, and rib fracture) in each of six replicated readings, yielding 36 separate ratings of each case for the five abnormalities. Separate data analyses of all cases and subsets of the difficult/subtle cases for each abnormality estimated the relative gains in accuracy (linear-scaled area below the ROC curve) obtained by averaging the case-ratings across (a) six independent replications by each reader (25% gain), (b) six different readers within each replication (34% gain), or (c) all 36 readings (48% gain). Although accuracy differed among both readers and abnormalities, ROC curves for the median ratings showed similar relative gains in accuracy, somewhat greater than those predicted from the measured rating correlations. A model for variance components in the observer's latent decision variable could predict these gains from measured correlations in the single ratings of cases. Depending on whether the model's estimates were based on realized accuracy gains or on rating correlations, about 48% or 39% of each reader's total decision variance (summed variance for positive and negative cases) consisted of random (within-reader) error that was uncorrelated between replications, another 10% or 14% came from idiosyncratic responses to individual cases, and about 43% or 47% was systematic variation that all readers found in the sampled cases.

Applying computer-assisted detection schemes to digitized mammograms after JPEG data compression: an assessment.

RATIONALE AND OBJECTIVES: The authors' purpose was to assess the effects of Joint Photographic Experts Group (JPEG) image data compression on the performance of computer-assisted detection (CAD) schemes for the detection of masses and microcalcification clusters on digitized mammograms. MATERIALS AND METHODS: This study included 952 mammograms that were digitized and compressed with a JPEG-compatible image-compression scheme. A CAD scheme, previously developed in the authors' laboratory and optimized for noncompressed images, was applied to reconstructed images after compression at five levels. The performance was compared with that obtained with the original noncompressed digitized images. RESULTS: For mass detection, there were no significant differences in performance between noncompressed and compressed images for true-positive regions (P = .25) or false-positive regions (P = .40). In all six modes the scheme identified 80% of masses with less than one false-positive region per image. For the detection of microcalcification clusters, there was significant performance degradation (P < .001) at all compression levels. Detection sensitivity was reduced by 4%-10% as compression ratios increased from 17:1 to 62:1. At the same time, the false-positive detection rate was increased by 91%-140%. CONCLUSION: The JPEG algorithm did not adversely affect the performance of the CAD scheme for detecting masses, but it did significantly affect the detection of microcalcification clusters.

Computerized localization of breast lesions from two views. An experimental comparison of two methods.

RATIONALE AND OBJECTIVES: The authors compared two computerized methods, the arc and cartesian straight-line, for the localization of breast lesions in two mammographic views. METHODS: A total of 571 craniocaudal and 571 mediolateral oblique matched mammographic image pairs (or 1142 individual images) depicting 290 pathology-verified masses on both views were selected from our image database. Using a previously developed computer-aided detection scheme, all 290 masses and 3992 suspicious but negative regions were identified. After pairing all identified regions from both views, all masses (true-positive-true-positive matched pairs) and a total of 10330 false-positive pairs (including false-positive-false-positive, true-positive-false-positive, and false-positive-true positive pairs) were assessed as to their position in relation to the nipple using both the arc and the cartesian straight-line methods. Receiver operating characteristic methodology was used to evaluate the performance levels for each method in determining, based solely on location, whether a pair of suspicious regions represented a true mass or a false-positive combination. RESULTS: The areas under the receiver operating characteristic curves (Az) were 0.79 and 0.78 for the arc and cartesian straight-line methods, respectively. The difference between the two techniques (as measured by Az) was not statistically significant (P > 0.99). CONCLUSIONS: These preliminary results demonstrated that the two methods are comparable in identifying true masses from triangulated observations on two views. However, the arc method is somewhat favorable because only the nipple location is required for localization.

Observer sensitivity to small differences: a multipoint rank-order experiment.

OBJECTIVE: To evaluate observer sensitivity to small differences in image presentation, a multipoint rank-order experiment was used to identify small differences or trends in observations. MATERIALS AND METHODS: Ten observers were presented with 50 sets of breast images that had been compressed at five different levels. Each set contained six images ranging from noncompressed to approximately 101:1 compression. Observers were asked to review all images of a case side by side and rank order the quality of each to enable determination of the presence or absence of masses and clustered microcalcifications. RESULTS: As a group, observers were able to detect small differences among the images, even at the lower compression levels (p < .001). As compression levels and image degradation increased, the ability to identify differences between different modes also increased. Large observer variability in discrimination ability was observed. CONCLUSION: Multipoint rank ordering of images viewed side by side can be an efficient method to identify small differences in image presentation. This approach to image ranking could be used to rule out or confirm the need for objective observer performance-type studies.

Feature selection for computerized mass detection in digitized mammograms by using a genetic algorithm.

RATIONALE AND OBJECTIVES: To investigate optimization of feature selection for computerized mass detection in digitized mammograms, and to compare the effectiveness of a genetic algorithm (GA) in such optimization with that of an "exhaustive" search of all feature permutations. MATERIALS AND METHODS: A Bayesian belief network (BBN) was used to classify positive and negative regions for masses depicted in digitized mammograms; 20 features were computed for each of 592 positive and 3,790 negative regions in two databases. Conditional probabilities for the BBN were computed by using a "training" database of 288 positive and 2,204 negative regions. Performance was measured by the area under the receiver operating characteristic curve (A) by using the remainder database (304 positive and 1,586 negative regions). The optimal set was first found by using an "exhaustive" (complete permutation) searching method. A GA-based search for the optimal set then was applied, and the results of the two approaches were compared. RESULTS: As the number of features in the classifier increased, the A value increased until it reached a maximum performance for 11 features of 0.876 +/- 0.008. The A value then decreased monotonically as the number of features increased from 11 to 20. Using 100 random chromosomes (seeds) in the first generation, the GA identified the same optimal set of features but reduced the total computation time by a factor of 65. CONCLUSION: A GA-based search might be an efficient and effective approach to selecting an optimal feature set.

Computer-assisted diagnosis of breast cancer using a data-driven Bayesian belief network.

This study investigates a simple Bayesian belief network for the diagnosis of breast cancer, and specifically addresses the question of whether integrating image and non-image based features into a single network can yield better performance than hybrid combinations of independent networks. From a dataset of 419 cases, including 92 malignancies, 13 features relating to mammographic findings, physical examinations and patients' clinical histories, were extracted to build three Bayesian belief networks. The scenarios tested included a network incorporating all features and two hybrids which combined the outputs of sub-networks corresponding to the image or non-image features. Average areas (Az) under the corresponding ROC curves were used as measures of performance. The network incorporating only image based features performed better (Az =0.81) than that using nonimage features (Az = 0.71). Both hybrid classifiers yielded better performance (Az =0.85 for averaging and Az = 0.87 for logistic regression), but neither hybrid was as accurate as the network incorporating all features (Az = 0.89). This preliminary study suggests that, like human observers who concurrently consider different types of information, a single classifier that simultaneously evaluates both image and non-image information can achieve better diagnostic performance than the hybrid combinations considered here.

Identification of clustered microcalcifications on digitized mammograms using morphology and topography-based computer-aided detection schemes. A preliminary experiment.

RATIONALE AND OBJECTIVES: A mathematical morphology-based computer-aided detection (CAD) scheme for the identification of clustered microcalcifications was developed and tested. The potential for improving either sensitivity or specificity by combining the results with those previously reported was investigated. METHODS: The CAD scheme presented here is based on mathematical morphology and a series of simple rule-based criteria for the identification of clustered microcalcifications. A database of 105 digitized mammograms was used for training and rule setting of the scheme. A test set of 191 digitized mammograms was used to evaluate its performance. The same test set had been used to evaluate a multilayer, topography-based scheme. The results obtained by the two schemes were then combined using logical OR and AND operations. RESULTS: The morphology-based and topography-based CAD schemes performed at sensitivities of 82.9% and 89.5%, with false-positive detection rates of 1.3 and 0.4 per image, respectively. A logical OR operation resulted in 95.4% sensitivity. An AND operation achieved 76.2% sensitivity, with no false identifications on 93% of images. CONCLUSIONS: By combining the results of the morphology-based and the topography-based schemes, either sensitivity or specificity can be improved.

Adequacy testing of training set sample sizes in the development of a computer-assisted diagnosis scheme.

RATIONALE AND OBJECTIVES: The authors assessed the performance changes of a computer-assisted diagnosis (CAD) scheme as a function of the number of regions used for training (rule-setting). MATERIALS AND METHODS: One hundred twenty regions depicting actual masses and 400 suspicious but actually negative regions were selected as a testing data set from a database of 2,146 regions identified as suspicious on 618 mammograms. An artificial neural network using 24 and 16 region-based features as input neurons was applied to classify the regions as positive or negative for the presence of a mass. CAD scheme performance was evaluated on the testing data set as the number of regions used for training increased from 60 to 496. RESULTS: As the number of regions in the training sets increased, the results decreased and plateaued beyond a sample size of approximately 200 regions. Performance with the testing data set continued to improve as the training data set increased in size. CONCLUSION: A trend in a system's performance as a function of training set size can be used to assess adequacy of the training data set in the development of a CAD scheme.

Subjective quality assessment of computed radiography hand images.

To evaluate the sensitivity of a non-receiver-operating characteristic (ROC) study in assessing small differences of perceived image quality of hand images acquired by computed radiography (CR) and conventional screen-film systems, hand images were acquired on 12 patients with both conventional screen-film and CR. Each CR image was then processed with three different edge-enhancement algorithms. One conventional film and four CR images were then viewed side by side by five radiologists. Observers rated perceived image quality of each radiograph using a 10-category discrete scale. The study was repeated after 6 weeks using a different block randomization scheme. Despite the small sample size, significant differences (P < .05) in assigned image quality were detected among CR images acquired at low, medium, and high resolutions. Image processing routines did not fully compensate for differences in quality between conventional film and CR-acquired images. The quality rating of the reference conventional image was found to be dependent on the quality of images with which it was compared. Small, highly sensitive study designs can be used to identify radiologists' perceived differences in image quality. "Reference" or "gold standard" quality are important in such studies. Edge-enhancement schemes cannot fully compensate for perceived image quality degradations because of reduced image resolution.

Joint photographic experts group (JPEG) compatible data compression of mammograms.

We have developed a Joint Photographic Experts Group (JPEG) compatible image compression scheme tailored to the compression of digitized mammographic images. This includes a preprocessing step that segments the tissue area from the background, replaces the background pixels with a constant value, and applies a noise-removal filter to the tissue area. The process was tested by performing a just-noticeable difference (JND) study to determine the relationship between compression ratio and a reader's ability to discriminate between compressed and noncompressed versions of digitized mammograms. We found that at compression ratios of 15:1 and below, image-processing experts are unable to detect a difference, whereas at ratios of 60:1 and above they can identify the compressed image nearly 100% of the time. The performance of less specialized viewers was significantly lower because these viewers seemed to have difficulty in differentiating between artifact and real information at the lower and middle compression ratios. This preliminary study suggests that digitized mammograms are very amenable to compression by techniques compatible with the JPEG standard. However, this study was not designed to address the efficacy of image compression process for mammography, but is a necessary first step in optimizing the compression in anticipation of more elaborate reader performance (ROC) studies.

Subjective and objective assessment of image quality--a comparison.

Forced-choice just noticeable difference (JND) studies are extremely sensitive to image quality variations that are below the threshold at which the differences are apparent to or definable by the observer. Paired comparisons of 4K and 2K laser-printed posteroanterior chest images consistently demonstrated that although images are viewed as comparable by radiologists, when forced to choose the better ("sharper") image, they actually select the higher-resolution images in 83% of the paired observations. We conclude that small differences in image quality may be detectable even in image sets which are considered to be comparable by subjective assessments.

Sequential viewing of abdominal CT images at varying rates.

PURPOSE: To evaluate radiologists' ability to detect abdominal masses during sequential viewing of series of computed tomographic (CT) scans at varying rates. MATERIALS AND METHODS: Receiver operating characteristic (ROC) analysis was used to assess the ability of five experienced radiologists to determine the presence or absence of subtle abdominal masses in 29 cases (15 positive, 14 negative) while viewing CT scans sequentially at different rates (0.5, 1, 2, 4, 7, and 21 images per second) and also at reader-selectable rates. RESULTS: Even at extremely fast viewing rates (21 images per second), radiologists performed significantly better (P < .05) than would be expected by chance alone (average area Az under the ROC curve = 0.73 vs 0.5). As the viewing rate decreased, their performance increased. The reader-selectable mode was better than any fixed-rate cine mode (average Az = 0.93). CONCLUSION: Fixed-rate sequential viewing of CT images for the primary diagnosis of subtle abdominal masses should be restricted to no more than one or two images per second, but the reader-selectable viewing mode is preferable to any fixed-rate cine mode.

Diagnostic reading session: temporal patterns and case-order effects.

The authors analyzed receiver-operating-characteristic studies to determine temporal patterns and performance as a function of the elapsed time in a reading session. Nineteen radiologists each read as many as 300 chest images with use of seven different display modalities, including conventional and laser-printed film and high-resolution soft display. With a computerized reporting system, the ratio of observers' interpretation rates (time to diagnosis) were recorded for the last five and 10 compared with the first five and 10 of 30-40 cases seen in sessions lasting 45-110 minutes. Observers tended to accelerate their interpretation as the sessions progressed by an average of 15% (P < .001). The acceleration was consistent for all readers (both fast and slow) with a variety of display modes under the nonrestricted time environment.

Stable xenon CT cerebral blood flow imaging: rationale for and role in clinical decision making.

The stable xenon CT method of measuring cerebral blood flow has been investigated in research studies for over 10 years. Recently, it has been gaining clinical acceptance, primarily owing to a combination of several unique advantages it holds over other cerebral blood flow measurement techniques. The accuracy of this technique in quantifying low cerebral blood flow gives it a unique application in cases of brain death and acute stroke and it can be repeated after an interval of 20 min. making it possible to evaluate autoregulation and cerebrovascular reserve. Furthermore, cerebral blood flow information is directly coupled to CT anatomy. Although it is more difficult to administer than a standard CT scan, careful monitoring can ensure patient safety during the examination. In this article we review the physiologic and technical bases for the clinical application of xenon CT-derived quantitative cerebral blood flow information and discuss the advantages and disadvantages of the technique. We also describe its current clinical applications, including its usefulness in the evaluation of acute stroke, occlusive vascular disease, carotid occlusion testing, vasospasm, arteriovenous malformations, and head trauma management.

Primary CT diagnosis of abdominal masses in a PACS environment.

Whether the display medium--film versus cathode ray tube (CRT)--affects observer performance during interpretation of computed tomographic (CT) images is an important research issue in these times of implementation and growth of picture archiving and communications systems in radiology. The authors performed a multiobserver receiver operating characteristic (ROC) study to determine the performance of radiologists who read abdominal CT studies displayed on film, as well as on a high-resolution workstation (video monitor) that made use of three different display modes. A total of 166 examinations were evaluated by eight radiologists, who recorded their ordinal confidence ratings of the demonstration of presence or absence of abdominal masses. ROC analysis showed small differences in the confidence ratings assigned by individual readers for the detection and interpretation tasks. Results for the group as a whole showed no significant reduction or improvement in observer performance when ratings for any one of the workstation display modes were analyzed. The results of this study demonstrate that current CRT display technology is adequate for enabling the primary detection of abdominal masses with CT examinations.

Xenon-enhanced CT of the brain: effect of flow activation on derived cerebral blood flow measurements.

The errors associated with derivation of cerebral blood flow values by the xenon-enhanced CT method have been evaluated through computer simulations as a function of flow-activation patterns and different scanning protocols. The results of this study indicate that actual flow activation during inhalation increases the derived flow values in a systematic way. Compared with the errors introduced by CT noise and/or variations in scanning protocols, flow activation introduces relatively small errors in the derived flow value when the washin technique is used.

The effect of image processing on chest radiograph interpretations in a PACS environment.

The question of whether image processing affects a radiologist's diagnostic performance is becoming more important as the digital modalities proliferate. In the multi-observer study reported, the performance of radiologists who interpret a series of posteroanterior digitized chest images displayed on a high-resolution workstation, with and without a set of image processing options, is determined. These include brightness, contrast, reverse look-up tables (black-bone), and two edge enhancement options. Three hundred images were evaluated twice (once in each mode) by each of seven board-certified radiologists, who recorded their confidence ratings for the presence or absence of one or more of the following abnormalities: interstitial disease, nodule, and pneumothorax. The original, unprocessed digital image was available for reference for those sessions in which the processing options were available. With the exception of one reader, receiver operating characteristic (ROC) analysis showed no statistically significant difference between the two modes (with and without processing) for the detection of any of the different abnormalities by individual readers. Likewise, the group as a whole showed no significant difference (P less than .05) for detection of any of the three abnormalities between the two reading modes.