FRINK Linked Data Fragments server

Matches in SemOpenAlex for { <https://semopenalex.org/work/W2069172368> ?p ?o ?g. }

• W2069172368 endingPage "814" @default.
• W2069172368 startingPage "811" @default.
• W2069172368 abstract "Arguing against the Proposition is Maria Kallergi, Ph.D. Dr. Kallergi received her Ph.D. in Physics and Electrical Engineering from the University of South Florida in 1990. Her background was in digital detectors and semiconductor devices but, in 1991, she started a career in medical imaging as a Postdoctoral Research Fellow at the H. Lee Moffitt Cancer Center & Research Institute and the Department of Radiology at the University of South Florida, where she is currently an Associate Professor of Radiology. Dr. Kallergi's research interests focus on two- and three-dimensional digital breast imaging technologies and computer methodologies for the automated detection and diagnosis of breast, lung, and pancreatic cancers. For CAD to be effective, two conditions must exist. Firstly, the CAD system must be capable of detecting cancers that are missed clinically. Secondly, the radiologist must be able to recognize when the computer has found a missed cancer. There is ample evidence in the literature to support the first condition. However, published sensitivities for the detection of clinically missed breast cancers range from 50%–80%.3–5 Hence, there is not conclusive evidence that the second condition is true. There have been four published clinical evaluations of the use of CAD for the detection of breast cancers.1,6–8 Two of these were inconclusive,6,7 one because it was only a pilot study,7 and two showed an apparent benefit for using CAD,1,8 there was an increase in the number of cancers detected. However, neither of these two studies was statistically significant, probably because breast cancer prevalence in a screening population is low. In one of the two CAD-beneficial studies, clustered microcalcifications were detected in seven of the eight missed cancers that were detected by the computer and only one was a mass.1 In the pilot study, the only CAD-detected cancer was a cluster of microcalcifications. The two other studies were retrospective temporal comparisons and it is not possible to determine what type of lesions the CAD system helped the radiologists find. CAD detection of clustered microcalcifications can be effective in helping radiologists find cancers. This is because the sensitivity of the scheme is high—up to 98%. Further, very little normal breast anatomy mimics calcifications in a mammogram. Therefore, radiologists can easily dismiss false detections. However, radiologists do not miss calcifications as often as they miss masses. In two large studies, 30% of missed cancers demonstrated calcifications, while 70% demonstrated masses.3,4 This implies that radiologists either do not look at the computer prompt for masses, or that they look at the prompted area but do not recognize that a cancer is present. As shown by Zhang et al., one reason why radiologists tend to ignore correct prompts is that the false detection rate of the CAD scheme is too high.9 The superposition of normal breast structure often mimics a mass. Not only does this increase the false-detection rate of the CAD scheme, but it also makes it difficult for radiologists to determine if the computer is prompting a true or false lesion. I believe that computer-aided diagnosis will one day be pervasive throughout radiology. However, it is important that we understand its limitations, both technical and clinical. In this way we can improve the technology and this will lead to faster clinical implementation and acceptance. In my opinion, further research is needed to improve CAD for mammography, particularly for mass detection, so that it can truly be an effective clinical tool for radiologists. CAD for screening mammography is not quite there yet. Success depends on a number of factors when assessing the effectiveness of any new medical technology, including computer-aided detection systems for screening mammography. Firstly, there is the question of whether CAD has fulfilled its initial goal. CAD was intended as an aid to the radiologist. Its primary goal was to help detect breast calcifications and spiculated masses at an earlier stage. There are now several testimonies that current CAD systems meet this challenge with a positive impact on cancer detection rates particularly for masses.8,10–12 In addition, callback rates, a major concern initially, have not been affected13,14 and the false positive signals, a major turn-off factor in the first generation of CAD systems, have been significantly reduced in subsequent upgrades (by almost a factor of 2) and can now be handled with greater comfort and dismissed easily by the trained reader.15 In terms of clinical acceptance, ten years ago, when these systems were still in the research laboratories, the medical community was divided, with unfriendly skeptics being an overwhelming majority. Today, radiologists are still divided but the friendly group has grown bigger and friendlier and many of the skeptics have become converts. This change is a consequence of the clinical implementation of CAD that has allowed the “learner” to practice and acquire the necessary skills for its use. It is also due to the positive results of recent studies that evaluated the systems clinically.10 Radiologists now find that CAD has increased their level of confidence and that its role is not only that of a “second reader” but also a “refocusing tool” in the often monotonous and repetitive environment of screening mammography, where external factors cause attention interruption and possible observational oversight. CAD has also had a positive impact beyond the radiologist: on the patient and the administrator. Patients appear more comfortable with the mammographic procedure, they demonstrate an increased level of security, and are more amenable to radiology decisions. Administrators are more likely to embrace CAD now because it can be seen as a marketing tool that could benefit the institution and, in some practices, it even makes good fiscal sense due to the higher reimbursement rate of the mammography package that includes CAD. Admittedly, the current commercial systems have problems and limitations. Both are well known and understood. It is also well known, but poorly understood by those outside the CAD community, that CAD requires continued commitment and investment to reach the desired levels of performance. Although far from perfect, today's CAD is steadily shaping its role as part of the standard of care.16 This is due to its demonstrated success in improving (i) the sensitivity of screening mammography and the confidence of the reader, (ii) the security and comfort of the patient, and (iii) the workflow and financial prospects of the administrator. My colleague, Maria Kallergi, raises a number of important points. The most important one on which we agree is that further research is needed to improve CAD, because existing systems have problems and limitations. I, however, disagree that these problems are all known and well understood. I think the biggest problem is that experience has shown that radiologists frequently ignore “missed” lesions detected by CAD that turn out to be true positives. As Birdwell et al. noted, the cancer detection rate increased by 7.4% in their clinical study, whereas, in their retrospective study, CAD was able to detect at least 27% of “actionable” cancers.4,11 This difference is one of the principle reasons why CAD is not effective in screening mammography. Further I do not agree that callback rates remain unaffected by CAD. Studies by Freer1 (18%), Helvie7 (9.7%), Birdwell11 (8.2%) and Cupples8 (10.9%) all found that the callback rate increased when CAD was used. Further research is needed to understand whether the increase in sensitivity justifies the increased callback rate, although I believe that it probably does. I do agree that there are intangible benefits of CAD to the radiologist. Higher confidence and less fatigue are just two. However, one would hope that this would result in higher performance. This has not yet been documented conclusively. Defining the benefits of CAD in screening mammography is difficult because of the low cancer prevalence in the screening population. Clinical evaluations are the only true measure of clinical effectiveness. Retrospective studies, two of which Dr. Kallergi quotes as proof of the benefits of CAD, are not definitive and are often overly optimistic. No clinical study has demonstrated a statistically significant benefit for using CAD. I believe that CAD systems will improve and become the standard of care in the future, but those CAD systems of the future will necessarily have higher performance than the systems studied to date. Newer versions with higher performance are now available and should be tested clinically. I do not disagree with my esteemed opponent on technical issues or future requirements of CAD systems for mammography. I disagree on the way success is judged. I argue that the usefulness of CAD technology today cannot be judged by diagnostic benefits alone. I further argue that benefits come in small “doses,” as is the case for the vast majority of medical technologies, and depend on a variety of factors including practice type and volume, readers’ experience, and method of interpretation. But even if we focus on the diagnostic benefits alone, recent studies provide convincing evidence that the two conditions set by Dr. Nishikawa as being necessary to demonstrate that CAD is effective, are met by current systems. Specifically, the work of Brem et al.10 supports the first condition related to the capabilities of CAD, and that of Birdwell et al.11 supports the second condition related to the readers’ ability to interpret the CAD results correctly. This was a large prospective study, which demonstrated the clinical benefits of CAD with the surprising outcome that CAD was shown to be helpful in detecting masses that had been missed by the radiologist, whereas previous studies had shown most benefit for the detection of calcifications. These findings, along with results from previous studies,8 strengthen the arguments against the Proposition and there is now substantial evidence that current CAD systems are effective clinically in more than one way. If one problem has to be identified that potentially impacts negatively on the current CAD systems, it is that the marketing part of the technology was pursued before the science. The result has been the creation of a negative bias on the side of users, mainly due to high false detection rates. This bias has been, and still is, difficult to overcome. Also, we have been clumsy and unprepared for the training process. This has led to a longer and more difficult learning curve that has generated more bias and skepticism. Taking these two elements into account, it is not difficult to explain the inconclusive and sometimes contradictory results of earlier studies. It cannot be denied that definitive diagnostic improvements will assuage all doubts regarding the usefulness of CAD in mammography. This, however, should not be our only criterion of success. If all factors are considered, we can conclude that current CAD systems, although not at optimum performance, are useful. This allows us to dream of what it will be like when optimum performance is reached." @default.
• W2069172368 created "2016-06-24" @default.
• W2069172368 creator A5012296515 @default.
• W2069172368 creator A5026442912 @default.
• W2069172368 creator A5079423135 @default.
• W2069172368 date "2006-03-10" @default.
• W2069172368 modified "2023-10-18" @default.
• W2069172368 title "Computer-aided detection, in its present form, is not an effective aid for screening mammography" @default.
• W2069172368 cites W1994054582 @default.
• W2069172368 cites W1994307679 @default.
• W2069172368 cites W1995225352 @default.
• W2069172368 cites W2016748517 @default.
• W2069172368 cites W2031808256 @default.
• W2069172368 cites W2047298105 @default.
• W2069172368 cites W2078710233 @default.
• W2069172368 cites W2086005837 @default.
• W2069172368 cites W2089887167 @default.
• W2069172368 cites W2100684166 @default.
• W2069172368 cites W2102018226 @default.
• W2069172368 cites W2113115183 @default.
• W2069172368 cites W2119077877 @default.
• W2069172368 cites W2122500316 @default.
• W2069172368 cites W2140259714 @default.
• W2069172368 cites W2168805745 @default.
• W2069172368 doi "https://doi.org/10.1118/1.2168063" @default.
• W2069172368 hasPubMedId "https://pubmed.ncbi.nlm.nih.gov/16696455" @default.
• W2069172368 hasPublicationYear "2006" @default.
• W2069172368 type Work @default.
• W2069172368 sameAs 2069172368 @default.
• W2069172368 citedByCount "40" @default.
• W2069172368 countsByYear W20691723682012 @default.
• W2069172368 countsByYear W20691723682013 @default.
• W2069172368 countsByYear W20691723682014 @default.
• W2069172368 countsByYear W20691723682020 @default.
• W2069172368 countsByYear W20691723682021 @default.
• W2069172368 countsByYear W20691723682022 @default.
• W2069172368 countsByYear W20691723682023 @default.
• W2069172368 crossrefType "journal-article" @default.
• W2069172368 hasAuthorship W2069172368A5012296515 @default.
• W2069172368 hasAuthorship W2069172368A5026442912 @default.
• W2069172368 hasAuthorship W2069172368A5079423135 @default.
• W2069172368 hasBestOaLocation W20691723681 @default.
• W2069172368 hasConcept C121608353 @default.
• W2069172368 hasConcept C126322002 @default.
• W2069172368 hasConcept C154945302 @default.
• W2069172368 hasConcept C19527891 @default.
• W2069172368 hasConcept C2780472235 @default.
• W2069172368 hasConcept C2781281974 @default.
• W2069172368 hasConcept C2910710802 @default.
• W2069172368 hasConcept C31601959 @default.
• W2069172368 hasConcept C41008148 @default.
• W2069172368 hasConcept C530470458 @default.
• W2069172368 hasConcept C71924100 @default.
• W2069172368 hasConceptScore W2069172368C121608353 @default.
• W2069172368 hasConceptScore W2069172368C126322002 @default.
• W2069172368 hasConceptScore W2069172368C154945302 @default.
• W2069172368 hasConceptScore W2069172368C19527891 @default.
• W2069172368 hasConceptScore W2069172368C2780472235 @default.
• W2069172368 hasConceptScore W2069172368C2781281974 @default.
• W2069172368 hasConceptScore W2069172368C2910710802 @default.
• W2069172368 hasConceptScore W2069172368C31601959 @default.
• W2069172368 hasConceptScore W2069172368C41008148 @default.
• W2069172368 hasConceptScore W2069172368C530470458 @default.
• W2069172368 hasConceptScore W2069172368C71924100 @default.
• W2069172368 hasIssue "4" @default.
• W2069172368 hasLocation W20691723681 @default.
• W2069172368 hasLocation W20691723682 @default.
• W2069172368 hasOpenAccess W2069172368 @default.
• W2069172368 hasPrimaryLocation W20691723681 @default.
• W2069172368 hasRelatedWork W1983939307 @default.
• W2069172368 hasRelatedWork W2012889557 @default.
• W2069172368 hasRelatedWork W2023780630 @default.
• W2069172368 hasRelatedWork W2029014456 @default.
• W2069172368 hasRelatedWork W2062909881 @default.
• W2069172368 hasRelatedWork W2144690133 @default.
• W2069172368 hasRelatedWork W2159722994 @default.
• W2069172368 hasRelatedWork W3096822600 @default.
• W2069172368 hasRelatedWork W1680833940 @default.
• W2069172368 hasRelatedWork W2116047071 @default.
• W2069172368 hasVolume "33" @default.
• W2069172368 isParatext "false" @default.
• W2069172368 isRetracted "false" @default.
• W2069172368 magId "2069172368" @default.
• W2069172368 workType "article" @default.