| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Commentary |
1 Radiology Department, George Washington University, 2150 Pennsylvania Ave., NW, Washington, DC 21117.
Received November 1, 2006; accepted after revision November 3, 2006.
Address correspondence to R. F. Brem.
Keywords: breast • breast cancer • CAD • mammography
In the December 2006 issue of AJR, two studies are presented that evaluate computer-assisted detection (CAD) for the diagnosis of breast cancer [1, 2]. Although there are some similarities in their approach, there are more differences and, not surprisingly, their results and conclusions differ.
The study by Taplin et al. [1] is a sophisticated study by an established group of investigators that shows no improvement in the sensitivity of cancer detection with CAD although the authors do report an increase in specificity. This study was performed in the research setting with a representative selection of cases that were refereed for overall breast density and difficulty of cancer visualization by one expert mammographer and used a number of assumptions in its data evaluation. It is novel in its approach in that it evaluates mammograms 0 to 3 years before cancer detection and evaluates a patient population in which there was no cancer in one third, cancer was detected in one third within 12 months of the screening mammogram, and cancer was detected in one third 13 to 24 months after the screening examination. It is an important study in the research setting with a very challenging evaluation of cancer detection—before the time the patient was diagnosed with cancer. However, at this point it is critical to evaluate the impact of CAD in the clinical setting, where it is to be used.
The study by Ko et al. [2] is exemplary. It was their laudable goal to prospectively examine the true impact of integrating CAD into clinical practice in a well-designed, prospective study showing the incremental improvement of breast cancer detection as a consequence of the interaction of the CAD system with the radiologist. Although numerous studies evaluating the impact of CAD are now available, most are retrospective studies evaluating the performance of CAD. Retrospective studies can achieve statistical power to support findings in a research setting, however, there are intrinsic limitations and with the increasing use of CAD in clinical practice, it is critical to understand its impact in its "natural," clinical environment. The study by Ko and colleagues shows a 4.7% increase in cancer detection with CAD. This incremental increase in cancer detection is somewhat lower than previously reported in other prospective clinical studies, which showed a 7.4-19.5% improvement in both academic and private practice sectors [3-6]. Nevertheless, the findings of all these studies consistently show that the use of CAD increases breast cancer detection by the radiologist and that the cancers detected are smaller and at an earlier stage than cancers detected without CAD. The positive impact of clinically implementing CAD without a downside is further confirmed by the small increase in recall rate (not statistically significant in most studies) and without a decrease in the positive predictive value of biopsy. Ko's study is further evidence that the use of CAD should be the state-of-the-art for the improved diagnosis of earlier and smaller breast cancers.
There are a number of differences in the studies beyond the fundamental study design, which may further explain the differences in findings. The study by Ko et al. [2] uses a current population of screening patients whereas the study by Taplin et al. [1] evaluated mammograms between 0 and 3 years before cancer detection. Furthermore, the radiologists in the study by Ko et al. had at least 2 months of experience using the CAD system in clinical practice and presumably were more familiar with its use than the radiologists in the study by Taplin et al., who had a 2-hour explanatory session with 10 cases before beginning the study. Regardless, the research versus the clinical setting and the vast difference in study design and data analysis are undoubtedly the causes for the differences in these studies' findings and conclusions.
In summary, the study by Taplin et al. [1] contributes new and important information to research in CAD. The study by Ko et al. [2] further supports the conclusion that CAD results in the improved detection of earlier and smaller breast cancers and should be implemented in clinical practice.
References
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
