How Widely Is Computer-Aided Detection Used in Screening and Diagnostic Mammography?
ABSTRACT The aim of this study was to determine how widely computer-aided detection (CAD) is used in screening and diagnostic mammography and to see if there are differences between hospital facilities and private offices.
The nationwide Medicare Part B fee-for-service databases for 2004 to 2008 were used. The Current Procedural Terminology(®) codes for screening and diagnostic mammography (both digital and screen film) and the CAD add-on codes were selected. Procedure volume was compared for screening vs diagnostic mammography and for hospital facilities vs private offices.
From 2004 to 2008, Medicare screening mammography volume increased slightly from 5,728,419 to 5,827,326 (+2%), but the use of screening CAD increased from 2,257,434 to 4,305,595 (+91%). By 2008, CAD was used in 74% of all screening mammographic studies. During this same time period, the Medicare volume of diagnostic mammography declined slightly from 1,835,700 to 1,682,026 (-8%), but the use of diagnostic CAD increased from 360,483 to 845,461 (+135%). By 2008, CAD was used in 50% of all diagnostic mammographic studies. In hospital facilities in 2008, CAD was used in 70% of all screening mammographic studies, compared with 81% in private offices. For diagnostic mammography in 2008, CAD was used in 48% in hospitals, compared with 55% in private offices.
Despite some operational drawbacks to using CAD, radiologists have embraced it in an effort to improve cancer detection. Its use has grown rapidly, and in 2008, it was used in three-quarters of all screening mammographic studies and half of all diagnostic mammographic studies. Women undergoing either screening or diagnostic mammography are more likely to receive CAD if they go to a private office than if they go to a hospital facility, although the differences are not great.
- SourceAvailable from: Stephen H Taplin[Show abstract] [Hide abstract]
ABSTRACT: Computer-aided detection (CAD) is applied during screening mammography for millions of US women annually, although it is uncertain whether CAD improves breast cancer detection when used by community radiologists. We investigated the association between CAD use during film-screen screening mammography and specificity, sensitivity, positive predictive value, cancer detection rates, and prognostic characteristics of breast cancers (stage, size, and node involvement). Records from 684 956 women who received more than 1.6 million film-screen mammograms at Breast Cancer Surveillance Consortium facilities in seven states in the United States from 1998 to 2006 were analyzed. We used random-effects logistic regression to estimate associations between CAD and specificity (true-negative examinations among women without breast cancer), sensitivity (true-positive examinations among women with breast cancer diagnosed within 1 year of mammography), and positive predictive value (breast cancer diagnosed after positive mammograms) while adjusting for mammography registry, patient age, time since previous mammography, breast density, use of hormone replacement therapy, and year of examination (1998-2002 vs 2003-2006). All statistical tests were two-sided. Of 90 total facilities, 25 (27.8%) adopted CAD and used it for an average of 27.5 study months. In adjusted analyses, CAD use was associated with statistically significantly lower specificity (OR = 0.87, 95% confidence interval [CI] = 0.85 to 0.89, P < .001) and positive predictive value (OR = 0.89, 95% CI = 0.80 to 0.99, P = .03). A non-statistically significant increase in overall sensitivity with CAD (OR = 1.06, 95% CI = 0.84 to 1.33, P = .62) was attributed to increased sensitivity for ductal carcinoma in situ (OR = 1.55, 95% CI = 0.83 to 2.91; P = .17), although sensitivity for invasive cancer was similar with or without CAD (OR = 0.96, 95% CI = 0.75 to 1.24; P = .77). CAD was not associated with higher breast cancer detection rates or more favorable stage, size, or lymph node status of invasive breast cancer. CAD use during film-screen screening mammography in the United States is associated with decreased specificity but not with improvement in the detection rate or prognostic characteristics of invasive breast cancer.CancerSpectrum Knowledge Environment 08/2011; 103(15):1152-61. DOI:10.1093/jnci/djr206 · 15.16 Impact Factor
- CancerSpectrum Knowledge Environment 08/2011; 103(15):1139-41. DOI:10.1093/jnci/djr267 · 15.16 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: For many years, it has been recognized that even the best radiologists make errors when reading medical exams including perception failures and interpretation failures. To reduce these problems, computer aided detection and diagnosis systems have been designed to aid radiologists detecting and classifying abnormalities. The first part of this thesis concerns combining information from multiple mammographic projection views to improve detection performance of computer aided detection systems. Most computer-aided detection systems that are used in the clinic today are focussed on reducing perception errors. The research presented in the second part of this thesis investigates if presenting CAD results in a fundamentally different way to avoid interpretation errors is more effective than current computer aided detection methods that focus on preventing perceptual oversights in medical screening. In Chapter 2, two machine learning techniques, namely support vector machines and Bayesian networks, were evaluated for characterizing masses as either benign or malignant. In addition, the effectiveness of dimension reduction (principal component analysis) and normal distribution transformation (Manly transformation) were investigated. It was found that the area under the ROC curve (Az) of the naive Bayesian classifier increased significantly (p=0.0002) when the Manly transformation was used, from Az = 0:767 to Az = 0:795. The Manly transformation did not result in a significant change for support vector machines. The difference between the support vector machines and the naive Bayesian classifiers using the transformed data set was not statistically significant (p=0.78). Applying dimension reduction in the form of PCA improved the classification accuracy of both classifiers, but the difference between the two classifiers after applying PCA was not statistically significant. In a breast screening program, it is important to combine all available information from a patient for making a referral decision. In Chapter 3 a Bayesian network framework was proposed that exploits multi-view dependencies for the analysis of screening mammograms. Instead of focussing on improving the localized detection of breast cancer, a CAD system was built that discriminates between normal and cancerous patients. It was investigated whether a reliable likelihood measure for a patient being cancerous could be obtained by combining information available as detected regions from a single-view CAD system from both mammographic views. This approach was tested with screening mammograms for 1063 patients of whom 385 had breast cancer. The results show that the multi-view modeling lead to significantly better performance in discriminating between normal and cancerous patients compared to using a singleview CAD system. During screening, a medio-lateral oblique (MLO) and cranio caudal (CC) view are often obtained from both breast. To train CAD systems that use correspondence information, corresponding regions in those views have to be found. Therefore, in Chapter 4 a method was developed to classify region pairs into the four possible types (combinations between a TP region in both views, a combination between a TP and FP or FP and TP, and combinations between FP regions). For each combination between regions some similarity features are calculated such as the difference in distance to the nipple, grayscale correlation, histogram correlation and other features that indicate similarity. Using these features, a 4-class k-Nearest Neighbour classifier was trained, to determine the four likelihoods for each combination type. The method was tested on an annotated dataset with 412 cases. Results show that for 82.4% of the TP regions, a correct link could be established. The difference between the 4-class kNN classifier and the LDA classifier from previous research was not statistically significant. When choosing threshold such that the percentage of correct TP-TP combinations is 70%, the number of TP-FP combinations decreases significantly when using the 4-class kNN classifier (Fisher’s exact test, p � 0:0008). It is expected that the decrease in TP-TP combinations will have a less negative effect on the detection performance of the two-view classifier because the regions are independently analyzed by the single-view CAD system. Radiologists generally combine information from multiple views to detect suspicious regions in mammograms. However, most of the current CAD systems analyze each view independently. It was investigated if case-based detection performance could be improved by optimizing the learning process of the multi-view classifier. Based on the output of a correspondence classifier that classified region pairs into the four possible types, the selection of training patterns to train the multi-view CAD system was biased. In that way, the training could be focussed towards improvement of case-based detection performance. The method was tested on 454 mammograms consisting of 4 views with a malignant region visible in at least one of the views. Case based evaluation showed a mean sensitivity improvement of 4.7% in the range of 0.01-0.5 false positives per image. Mammographic CAD systems that are currently used in clinical practice focus only on the problem of perception errors; however, misinterpretation is a far more common cause of missing breast cancer in screening than perceptual oversights. In Chapter 6 it was investigated if a workstation that allows readers to probe image locations for the presence of CAD information while reading mammograms could improve detection performance. If a CAD finding was present on the queried location, it was displayed with the computer estimated malignancy score. The approach was evaluated using an observer study with nine readers (four screening radiologists and five non-radiologists). The participants read 120 cases of which 40 cases had a malignant mass that was missed at the original screening. The performance of the average reader significantly increased with interactive CAD at low false-positive rates from 25.1% to 34.8%, without affecting reading time. It was found that in addition to using CAD in the traditional way to avoid perception errors, there is a large potential for using CAD as a decision aid to reduce interpretation failures. In Chapter 7 it was investigated if the interactive computer-aided detection (CAD) system introduced in 6 increases mass detection performance in comparison to the regular CAD prompting systems currently used in clinical practice. An observer study was conducted in which six certified screening radiologists and three residents read 200 difficult cases. Results show that the reader sensitivity increased significantly (p < 0:01) when interactive CAD was used (58.5%) compared to both reading without CAD (51.2%) and reading with CAD prompts (51.1%). There was no significant difference found in the number of unreported abnormal cases when mammograms were read with interactive CAD compared to reading with prompting CAD or to reading without CAD. In the last chapter it was investigated if the interactive method of presenting CAD results could also improve the usefulness of CAD in another application, namely the detection of nodules in chest radiographs. The effect of prompts and interactive use of CAD for detecting chest nodules was compared. Six readers read 247 chest radiographs that were selected from the publicly available JSRT database. The CAD results were taken from the commercially available CAD system (Riverain OnGuard™5.0). It was shown that with CAD prompting, mean sensitivity of the readers increased significantly from 35.2% to 42.8%. When using interactive CAD, the performance of the average reader increased significantly to 49.5%. This showed that CAD as a decision aid can improve readers’ nodule detection performance compared to the traditional use of CAD prompts, in particular at low false positive rates.Department of Radiology, Radboud University Nijmegen, 12/2011, Degree: PhD, Supervisor: N. Karssemeijer