Philippe Bamberger

Improved mammographic interpretation of masses using computer-aided diagnosis

Abstract. The aim of this study was to evaluate the effectiveness of computerized image enhancement, to investigate criteria for discriminating benign from malignant mammographic findings by computer-aided diagnosis (CAD), and to test the role of quantitative analysis in improving the accuracy of interpretation of mass lesions. Forty sequential mammographically detected mass lesions referred for biopsy were digitized at high resolution for computerized evaluation. A prototype CAD system which included image enhancement algorithms was used for a better visualization of the lesions. Quantitative features which characterize the spiculation were automatically extracted by the CAD system for a user-defined region of interest (ROI). Reference ranges for malignant and benign cases were acquired from data generated by 214 known retrospective cases. The extracted parameters together with the reference ranges were presented to the radiologist for the analysis of 40 prospective cases. A pattern recognition scheme based on discriminant analysis was trained on the 214 retrospective cases, and applied to the prospective cases. Accuracy of interpretation with and without the CAD system, as well as the performance of the pattern recognition scheme, were analyzed using receiver operating characteristics (ROC) curves. A significant difference (p < 0.005) was found between features extracted by the CAD system for benign and malignant cases. Specificity of the CAD-assisted diagnosis improved significantly (p < 0.02) from 14 % for the conventional assessment to 50 %, and the positive predictive value increased from 0.47 to 0.62 (p < 0.04). The area under the ROC curve (Az) increased significantly (p < 0.001) from 0.66 for the conventional assessment to 0.81 for the CAD-assisted analysis. The Az for the results of the pattern recognition scheme was higher (0.95). The results indicate that there is an improved accuracy of diagnosis with the use of the mammographic CAD system above that of the unassisted radiologist. Our findings suggest that objective quantitative features extracted from digitized mammographic findings may help in differentiating between benign and malignant masses, and can assist the radiologist in the interpretation of mass lesions.

Can the size of microcalcifications predict malignancy of clusters at mammography

RATIONALE AND OBJECTIVES The purpose of this study was to determine whether the size of mammographically detected microcalcifications is predictive of malignancy. MATERIALS AND METHODS Two hundred sixty mammograms showing clustered microcalcifications with proven diagnoses (160 malignant, 100 benign) were respectively reviewed by experienced mammographers. Lesions that were obviously benign in appearance were excluded from the study. A computer-aided diagnosis system digitized the lesions at 600 dpi, and the microcalcifications on the digital image were interactively defined by mammographers. Subsequently, three quantitative features that reflected the size of the microcalcifications-length, area, and brightness-were automatically extracted by the system. For each feature, the standard average of values obtained for individual calcifications within the cluster and the average with emphasis on extreme values (E) obtained in a single cluster were analyzed and matched with pathologic results. RESULTS In the malignant group of cases, the mean values of the standard average length and area were significantly higher (P < .0001) than the mean values in the benign group. Distribution analysis demonstrated that an average length of more than 0.41 mm was associated with malignant lesions 77% of the time, while an average length of less than 0.41 mm was associated with benign lesions 71% of the time. The mean of the average length (E) and area (E) of microcalcifications within the cluster demonstrated an even higher discriminative power when compared with the standard average length and area. The average brightness, on the other hand, showed only a low discriminative power. CONCLUSION Digital computerized analysis of mammographically detected calcifications demonstrated that the average length and area of the calcifications in benign clusters were significantly smaller than those in malignant clusters.

Optimizing parameters for computer-aided diagnosis of microcalcifications at mammography.

RATIONALE AND OBJECTIVES The purpose of this study was to optimize selection of the mammographic features most useful in discriminating benign from malignant clustered microcalcifications. MATERIALS AND METHODS The computer-aided diagnosis (CAD) system automatically extracted from digitized mammograms 13 quantitative features characterizing microcalcification clusters. Archival cases (n = 134; patient age range, 31-77 years; mean age, 56.8 years) with known histopathologic results (79 malignant, 55 benign) were selected. Three radiologists at three facilities independently analyzed the microcalcifications by using the CAD system. Stepwise discriminant analysis selected the features best discriminating benign from malignant microcalcifications. A classification scheme was constructed on the basis of these optimized features, and its performance was evaluated by using receiver operating characteristic (ROC) analysis. RESULTS Six of the 13 variables extracted by the CAD system were selected by stepwise determinant analysis for generating the classification scheme, which yielded an ROC curve with an area (Az) of 0.98, specificity of 83.64%, positive predictive value of 89.53%, and accuracy of 91.79% for 98% sensitivity. When patient age was an additional variable, the schemes performance improved, but this was not statistically significant (Az = 0.98). The ROC curve of the classifier (without age as an additional variable) yielded a high Az of 0.96 for patients younger than 50 years and an even higher (P < .02) Az of 0.99 for those 50 years or older. CONCLUSION Stepwise discriminant analysis optimized performance of a classification scheme for microcalcifications by selecting six optimized features. Scheme performance was significantly (P < .02) higher for women 50 years or older, but the addition of patient age as a variable did not produce a statistically significant increase in performance.

Analysis of clustered microcalcifications by using a single numeric classifier extracted from mammographic digital images

RATIONALE AND OBJECTIVES The authors prospectively tested the performance of a single numeric classifier constructed from a discriminative analysis classification system based on automatic computer-extracted quantitative features of clustered microcalcifications. MATERIALS AND METHODS Mammographically detected clustered microcalcifications in patients who had been referred for biopsy were digitized at 600 dpi with an 8-bit gray scale. A software program was developed to extract features automatically from digitized images to describe the clustered microcalcifications quantitatively. The significance of these features was evaluated by using the Wilcoxon test, the Welch modified two-sample t test, and the two-sample Kolmogorov-Smirnov test. A discriminant analysis pattern recognition system was constructed to generate a single numeric classifier for each case, based on the extracted features. This system was trained on 137 archival known reference cases and its performance tested on 24 unknown prospective cases. The results were evaluated by using receiver operating characteristic analysis. RESULTS Thirty-seven extracted parameters demonstrated a statistically significant difference between the values for the benign and for the malignant lesions. Seven independent factors were selected to construct the classifier and to evaluate the unknown prospective cases. The area under the receiver operating characteristic curve for the prospective cases was 0.88. CONCLUSION A pattern recognition classifier based on quantitative features for clustered microcalcifications at screen-film mammography was found to perform satisfactorily. The software may be of value in the interpretation of mammographically detected microcalcifications.

The use of an interactive software program for quantitative characterization of microcalcifications on digitized film-screen mammograms.

RATIONALE AND OBJECTIVES Mammography is relatively nonspecific for the early detection of breast cancer. This study evaluates the accuracy of mammographic interpretation using quantitative features characterizing microcalcifications, which are extracted by a computerized system. METHODS A computer-aided diagnosis (CAD) system enabling digitization of film-screen mammograms and automatic feature extraction was developed. A classification scheme (discriminant analysis) based on these features was constructed and trained on 217 cases with known pathology. The diagnostic performance of the classification scheme was tested against the radiologists conventional interpretation on 45 additional cases of microcalcifications, each analyzed independently by four radiologists. RESULTS The sensitivity of the CAD system analysis (95.7%) was significantly better than that of conventional interpretation (84.8%). The positive predictive value of interpretation increased significantly, as did the area under the receiver operating characteristic curve. CONCLUSIONS This classification scheme for microcalcifications, based on quantitative features characterizing the lesion, significantly improved the accuracy of mammographic interpretation.

Multiple-Instance Learning Improves CAD Detection of Masses in Digital Mammography

We propose a novel multiple-instance learning(MIL) algorithm for designing classifiers for use in computer aided detection(CAD). The proposed algorithm has 3 advantages over classical methods. First, unlike traditional learning algorithms that minimize the candidate level misclassification error, the proposed algorithm directly optimizes the patient-wise sensitivity. Second, this algorithm automatically selects a small subset of statistically useful features. Third, this algorithm is very fast, utilizes all of the available training data (without the need for cross-validation etc.), and requires no human hand tuning or intervention. Experimentally the algorithm is more accurate than state of the art support vector machine (SVM) classifier, and substantially reduces the number of features that have to be computed.

Addressing image variability while learning classifiers for detecting clusters of micro-calcifications

Computer aided detection systems for mammography typically use standard classification algorithms from machine learning for detecting lesions. However, these general purpose learning algorithms make implicit assumptions that are commonly violated in CAD problems. We propose a new ensemble algorithm that explicitly accounts for the small fraction of outlier images which tend to produce a large number of false positives. A bootstrapping procedure is used to ensure that the candidates from these outlier images do not skew the statistical properties of the training samples. Experimental studies on the detection of clusters of micro-calcifications indicate that the proposed method significantly outperforms a state-of-the-art general purpose method for designing classifiers (SVM), in terms of FROC curves on a hold out test set.

Optimizing the CAD Process for Detecting Mammographic Lesions by a New Generation Algorithm Using Linear Classifiers and a Gradient Based Approach

This study evaluates the performance of a new generation algorithm designed to both increase detection sensitivity of cancers and to markedly reduce the false mark rate. In the advanced algorithm, several improvements were implemented. The algorithm for the initial detection of potential mass candidates was upgraded to ignore dense areas that do not represent masses. For the initial detection of potential clusters candidates, the advanced algorithm considers interdependence between various stages of the parametric clusterization process and implements automatic performance optimization. Moreover, the advanced algorithm includes a one-step global classification model, which assigns a score to each candidate lesion, instead of sequential multi-step filtration at various steps of the algorithm. Both the advanced and the previous algorithm were run on 83 malignant cases, with proven pathology, and on 523 normal screening cases that were consecutively culled from 4 clinical sites. The overall sensitivity of the advanced algorithm was 86%, compared to a sensitivity of 84% for the previous one. The false mark (FM) rate per case, decreased from 3.20 for the previous algorithm, to 1.39 for the advanced one. The advanced algorithm reduced both mass FMs and cluster FMs. In conclusion, the new algorithm outperforms the old one with a slight increase in sensitivity and with a substantial reduction in false mark rate for both masses and clusters.

Improved mammographic accuracy with CAD assisted classification of lesions

We evaluate a CAD device with detection and classification capabilities and compare conventional to computerized analysis. 243 cases (126 malignant, 117 benign) were analysed using BI-RADS and digitized (600 DPI, 12 bits). Lesions were detected, classified by likelihood of malignancy, and stratified into BI-RADS categories 2–5 by the CAD device. The falsely detected findings scored by CAD as low probability of malignancy were discarded to evaluate the true false positive rate. The CAD device sensitivity was 96% for masses and 95% for clusters of MCs. Malignancies were correctly classified by CAD in 95%. 67% of the false positive detected masses and 76% of the false positive clusters were classified benign by the CAD device, reducing the false positive rate per view from 0.59 to 0.20 for masses and from 0.30 to 0.07 for clusters. Conventional interpretation yielded a ROC Az of 0.76. CAD improved the Az to 0.88 (pO.OOl).

Interactive Quantitative Characterization of Micro-Calcifications on Digitized Film-Screen Mammograms

Early detection of breast cancer through the use of mammography has been shown to be effective in reducing mortality from the disease [1],[2]. The presence of clustered micro-calcifications on a mammogram is a sensitive but non-specific indicator of malignancy. In order to find the relatively small number of malignancies present among clustered micro-calcifications, many women are referred for biopsies. The increased number of biopsies increases the overall cost of screening for breast cancer, and raises a barrier to its implementation [3]. In addition, the morbidity, both psychological and physical, is not inconsequential.