| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Commentary |
1 Both authors: Department of Radiology, Imaging Research, Ste. 4200, University of Pittsburgh, Magee-Womens Hospital, 300 Halket St., Pittsburgh, PA 15213-3180.
Received October 18, 2006; accepted after revision October 20, 2006.
Address correspondence to D. Gur
(gurd{at}upmc.edu).
OBJECTIVE. The benefit and cost of computer-assisted detection (CAD) mammography screening remains a topic of great interest in breast imaging. Our purpose is to reflect on and interleave two articles in this issue of the AJR that highlight the difficulty in assessing the actual benefit of using CAD from either retrospective or prospective studies.
CONCLUSION. This commentary describes the possible benefit and some of the issues associated with the clinical use of current CAD technology while emphasizing the expectation of and need for future improvements in CAD performance.
Keywords: breast cancer • CAD • mammography • screening
When computerized chess games first appeared four decades ago, it was unimaginable to most that their performance would exceed that of some of the best players in the world. Similarly, computer-assisted detection (CAD) technology for the detection and classification of suspicious mammograms continues to evolve and improve. Computers are clearly better at tasks that require tedious, repetitious processes when the target of interest is well defined in a complex multidimensional feature domain. There are good reasons as to why the first U.S. Food and Drug Administration (FDA)-approved clinical use of CAD in diagnostic imaging was for aiding in the detection of suspicious regions on mammograms. The studies by Taplin et al. [1] and Ko et al. [2] present both prospective and retrospective assessments of CAD when used as a "second reader." Both studies add to the rapidly increasing body of literature suggesting that measuring actual improvement of cancer detection is not a simple task and the results of these studies are dependent on the performance of the CAD itself, the population of cases to which it is applied, and the reviewers who use it. Regardless, even at this early stage, we can state that similar to double readings, CAD can increase cancer detection rates during the first few years of its use, albeit in most instances it occurs at the cost of increasing recall rates as well. This will by definition increase the number of cancers that are detected earlier not only during the transition period (i.e., the first 2 or 3 years of CAD use) but also in a steady-state environment. However, while a small fraction of CAD-detected cancers are diagnosed at least 1 year earlier, at the current level of performance, on average, cancers will be detected earlier only by a small amount (e.g., zero to 2 months).
While the validity of the methodology used in many studies can be criticized, that the studies converge in their conclusion toward some positive findings cannot be ignored. Most studies published to date suggest that there is a clear benefit to the use of CAD in terms of increased cancer detection rates, in particular with less experienced and/or low-volume reviewers. The actual demonstrated benefit increases from none to approximately 20% with most studies suggesting an increase between 2% and 10%. The two articles in this issue [1, 2] are actually on the lower side of this range but they add to our growing confidence that, if nothing else, CAD is beneficial.
CAD in mammography clearly increases the efficiency and confidence level of radiologists while searching for subtle microcalcification clusters and, when asked, most users will subjectively comment that they are less fatigued at the end of a CAD-supported reading session. This by itself constitutes an important benefit in an environment that requires great alertness and attentiveness to details and suffers from significant shortages in qualified readers.
Clearly, improvement of CAD performance in detecting masses is highly desirable and it is possible that multiimage-based analyses will increase reviewers' confidence in the CAD results, as well. In this regard, we are at the beginning of the road rather than at the end and significant research support is needed despite the fact that there are already several commercial CAD systems on the market. The two studies in this issue [1, 2] underscore the need for additional research to improve our understanding of optimal clinical use of current CAD systems and ultimately to focus on increasing early detection of those cancers whose outcome is more likely to be improved by earlier detection.
Our experience has been that if you tell a radiologist that CAD could be eliminated as a diagnostic option, most would adamantly object. However, it is not clear if they would take the same position it were to become nonreimbursable or if they had to pay for it themselves. The societal question that should be considered is, since CAD is primarily used as a "second reader," should CAD be judged in terms of reimbursement similar to double reading? In other words, should CAD be reimbursed just as double reading is (and vice versa)? The FDA approved the use of CAD based on limited data and the early decision by payers to reimburse its use generated a large commercial market. Reimbursement for CAD use (technical and professional combined) in the U.S. alone could exceed $800 million per year. The question of whether the use of CAD or other aiding practices or tools such as double readings, magnifying glasses, or image processing routines like edge enhancement that make detection easier and possibly more accurate, should be reimbursed is beyond the scope of this commentary, but it will eventually have to be addressed in a larger context.
Despite the issues raised to date, CAD is here to stay. Computerized decision support and eventually computerized decisions will inevitably be pervasive in many areas of medicine in general and diagnostic imaging in particular. Ultimately, CAD will be used in ways we now consider unacceptable and/or inappropriate. It is just a matter of time.
References
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
