| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Editorial |
1 From the Diagnostic Radiology Department, Warren Grant Magnuson Clinical Center, National Institutes of Health, Bldg 10, Rm 1C660, 10 Center Dr, MSC 1182, Bethesda, MD 20892-1182. Received January 3, 2003; accepted January 16. Supported by the intramural research programs of the Diagnostic Radiology Department, Warren Grant Magnuson Clinical Center. Address correspondence to the author (e-mail: rms@nih.gov).
Index terms: Computers, diagnostic aid • Diagnostic radiology • Editorials • Picture archiving and communication system (PACS) • Quality assurance
Radiologic diagnostic decision making is difficult. The breadth of possible abnormalities that are identifiable in a radiologic image is large. Hence, the length of radiology residencies is longer and the trend toward subspecialization is increasing. Radiologic images are becoming more complex, both in amount of information potentially gleaned and in data provided. Use of radiologic imaging is accelerating without a concomitant increase in the number of trained radiologists.
Computer-aided detection (CAD) has been proposed as a solution to interpretation of the ever-expanding amount of radiologic information. CAD is best for two types of tasks: tedious tasks, such as looking for a "needle in a haystack" (eg, a very small lung nodule at chest computed tomography [CT]), and tasks that involve a complex combination of multiple image features (eg, breast mass detection at mammography). CAD is also helpful if there is high interobserver variability or a lack of trained observers.
In this editorial, I postulate that CAD that is more comprehensive and versatile (ie, that includes a much broader range of detection tasks than do current CAD applications) would be more desirable to the radiology community, to patients, and to referring physicians. Rather than focusing CAD research on a narrow task, such as lung nodule detection, I propose a more global approach to CAD development that encompasses the totality of the radiologic image and the diagnostic tasks of the radiologist. Just as a radiologist looks for more than lung nodules at chest CT so also should CAD.
To make this vision a reality, I introduce the concept of the CAD road map, a strategic plan. The road map addresses not only the technical approach to solving a specific imaging problem but also the way to clinically validate the technique and make it available, to ensure quality, and to integrate it with the picture archiving and communication system (PACS) to make it more efficient. Following a brief description of the current status of CAD, I will give a sample road map for total CAD with a chest CT examination.
Current Status of CAD
CAD has a record of investigation dating back to the 1960s when articles on computer analysis of radiographic images appeared (1). As of December 2002, there were more than 170 literature references to CAD in radiology on PubMed. Major areas of research include detection of lung nodules at chest radiography and CT, of breast masses and microcalcifications at mammography, of breast masses at ultrasonography, and of quantitation of interstitial lung disease at chest CT (2–6). New CAD applications are emerging in colonic polyp detection at CT colonography and in pulmonary embolus detection at pulmonary CT angiography (7–9). CAD schemes that assess serial change and use multimodality registration are likely to be important and are also being developed (10,11). Technology for CAD in regard to breast lesion and lung nodule detection that has been approved by the Food and Drug Administration currently is commercially available.
The current approach to CAD research and development is somewhat fragmented. Researchers identify a clinically useful problem (eg, lung nodule detection), and then a large fraction of the research community devotes their efforts to solve it. This approach has two major drawbacks. First, it promotes neglect of a wide range of clinically important problems amenable to CAD. Second, it promotes development of CAD products with a narrow scope for single-task applications. The remedy is a broader view of the radiologists’ tasks and development of a list of CAD-amenable problems (of which there are many) and useful combinations. An integrated approach such as this would provide a more balanced and productive CAD research effort and would help less clinically oriented researchers in the pattern-recognition community to structure their efforts. This is the concept behind the road map, a path to more versatile radiologic CAD.
Sample Road Map: CAD in Interpretation of Chest CT Findings
Of CT examinations, the chest CT examination is typically most rapidly interpreted by the radiologist. Nevertheless, the chest CT examination serves as a useful paradigm for understanding the current status and possible future roles of CAD.
MacMahon et al (12) enumerated the range of pathologic findings of importance to CAD with chest radiography. An enumeration for the chest CT examination is likely to be quite similar to that for chest radiography and would be a useful guide for researchers, but this enumeration currently does not exist.
In the absence of such an enumeration, we can begin with an anatomic classification of areas where findings would have clinical importance (Fig 1). Attention must be paid to the lungs, the airways, the great vessels, the mediastinum and hila, the lymph vessels, the spine, the ribs, the sternum, the scapulae, the shoulders, the muscles, the subcutaneous tissues, the axillae, and the skin. In regard to these anatomic sites, CAD has been applied to chest CT examinations of the lungs (nodules [10,13,14] and interstitial disease [4]), the airways (15), and the vasculature (pulmonary embolus detection [8,9]). Quantitation of atherosclerosis (coronary calcium scoring) is performed routinely, but the plaques are generally located manually; this task too could be automated. Although CAD has been used successfully to identify pneumothorax at chest radiography (16), it has not yet been routinely applied to the chest CT examination. There is little work known concerning CAD of pneumonia (1).
|
A chest CT CAD system could also be used to locate sclerotic and lytic bone metastases. Such CAD would be beneficial for cancer patients and might help to reduce or eliminate the need to review bone window CT images, thereby increasing efficiency of interpretation. CAD for bone abnormalities would need to distinguish metastatic lesions from benign entities such as Schmorl nodes, bone islands, osteophytes, and hemangiomas. Automated vertebral body and rib numbering, vertebral body height assessment, bone densitometry for osteoporosis, and change analysis would also be of value with CAD for bone abnormalities.
The example tasks previously described do not apply only to the chest CT examination. Similar lists of diagnostic tasks can be developed for examination of the head, the neck, the abdomen, the pelvis, the bones, and the soft tissues.
Other Desirable Features of More Versatile CAD
Additional aspects of more versatile CAD are described in this section. Included are quality assurance, integration with PACS, validation, commercialization, and reimbursement. Some of these aspects require a coordination of resources, as shown in Figure 2.
|
Integration of CAD with PACS is also important. Radiologists do not want to learn to use a different physical workstation for each interpretation task; they would rather press a button on their existing PACS that would call up different types of CAD. By packaging the CAD with the PACS, clinical efficiency is improved. Current PACS typically have only rudimentary measurement tools and interpretation aids. The CAD road map should include development of an open interface so that CAD "plug-ins" (software modules) can be added to PACS that are from different vendors.
A critical step in a CAD road map is making CAD available. Unless it is available through commercialization or other means, CAD cannot help the patient. Both to clinically validate CAD and to meet regulatory requirements, the CAD road map must describe how to accumulate a clinical database to validate CAD. Such databases are costly to create. Publicly accessible databases of case material would assist CAD researchers to solve these problems. Example databases that are either funded or planned include the Lung Image Database Consortium for lung nodule detection, which is available at www3.cancer.gov/bip, and the ACRIN CT Colonography Database for Colonic Polyp Detection, which is Protocol A6664 and is available at www.acrin.org. These databases are supported by the National Cancer Institute. Many issues regarding such databases are well covered in articles by van Ginneken et al (1) and Sutton (17). The database should contain sufficient material from multiple institutions. Clinical assessment of observer performance should involve multiple observers with different skill levels.
New diagnostic tools such as CAD often face barriers to acceptance unless their superiority is clear. If reimbursement for CAD is not available or if CAD is expensive to acquire or to use, if it produces limited benefit compared with standard clinical interpretation, or if it impedes efficiency, then radiologists will be reluctant to use it. However, device manufacturers and patient advocacy groups often can influence the medical and regulatory community if they perceive the benefits to be great. Residency training programs and radiology societies also are likely to play a role in familiarizing radiologists with CAD and helping them integrate CAD into routine clinical practice. Manufacturers can play a leading role by making CAD easy to use. Reimbursement for mammographic CAD has led to an acceleration in the number of commercial products available. Strategies for reimbursement and clinical acceptance should be part of a CAD road map.
In summary, the more tasks that can be performed with CAD, the more useful it will be. Future CAD will likely include PACS integration and quality assurance, as well as time series change analysis, registration, and functional and molecular imaging. New problems and challenges will appear as technology develops. These will place new demands on both radiologists and CAD and will add to the urgency to automate old tasks. Comprehensive road maps that lay out the extent of and potentially fruitful combinations of CAD tasks will help accelerate the pace of research and identify the steps and support needed for development, validation, and dissemination.
ACKNOWLEDGMENTS
I thank Andrew Dwyer, MD, for helpful discussions and editing and review of the manuscript.
FOOTNOTES
The author has pending and/or been awarded patents for the subject matter described in this editorial.
REFERENCES
This article has been cited by other articles:
|
|
 |
|
  S. D. O'Connor, J. Yao, and R. M. Summers Lytic Metastases in Thoracolumbar Spine: Computer-aided Detection at CT--Preliminary Study Radiology, March 1, 2007; 242(3): 811 - 816. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. P. McAdams, E. Samei, J. Dobbins III, G. D. Tourassi, and C. E. Ravin Recent Advances in Chest Radiography Radiology, December 1, 2006; 241(3): 663 - 683. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K. L. Weiss, J. M. Storrs, and R. B. Banto Automated Spine Survey Iterative Scan Technique Radiology, April 1, 2006; 239(1): 255 - 262. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. L. Partain, H.-P. Chan, J. G. Gelovani, M. L. Giger, J. A. Izatt, F. A. Jolesz, K. Kandarpa, K. C. P. Li, M. McNitt-Gray, S. Napel, et al. Biomedical Imaging Research Opportunities Workshop II: Report and Recommendations Radiology, August 1, 2005; 236(2): 389 - 403. [Full Text] [PDF]
|
|
|
|
 |
|
 K Doi Current status and future potential of computer-aided diagnosis in medical imaging Br. J. Radiol., January 1, 2005; 78(suppl_1): S3 - s19. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. A. Krupinski Computer-aided Detection in Clinical Environment: Benefits and Challenges for Radiologists Radiology, April 1, 2004; 231(1): 7 - 9. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
