|Khan M. Siddiqui, MD|
Radiology is changing rapidly, especially with recent developments in advanced image processing and image visualization techniques. These advancements can be both a blessing and a curse. They help us improve patient care by enhancing diagnostic accuracy but, on the other hand, decrease productivity and efficiency because of the sheer amount of information that needs to be reviewed and analyzed. This problem has grown even larger with the advent of multislice CT, which allows us to image at submillimeter increments and generate more than 1,000 images per acquisition.
|Eliot L. Siegel, MD|
At the same time that the numbers of images per study are growing exponentially, the shortage of radiologists in the United States continues to worsen. Statistics suggest that radiologists nationwide are reading more studies than ever before. The challenge is to keep up with the volume without sacrificing the quality of their interpretation. Relative value work units associated with medical imaging are rising by 4% to 6% annually with a doubling time of 14 years.1 Diagnostic imaging workload is expected to increase by more than 50% between now and 2010.1 During the same time period, the pool of radiologists will increase by only 20%. 1
|Bruce I. Reiner, MD|
Researchers and vendors have tried to address this problem by developing advanced visualization software that offers the ability to quickly identify, quantify, and report abnormalities. But very little work has addressed the use of these applications in day-to-day radiology practice or investigated the workflow implications of such innovations. When radiologists switch from film-based to soft-copy reading, most try to mimic the familiar film-based paradigm and fail to take full advantage of the capabilities of soft-copy interpretation to improve productivity. Filmless radiology offers the possibility of entirely reinventing a previously inefficient workflow process, particularly the work involved in radiology interpretation. In many instances, however, the transition to filmless radiology has been made by simply plugging new technologies into old mind-sets. Those who have made such “partial” transitions are among those hardest hit by the advent of increasingly large image datasets.
To better understand how we can develop strategies for interpreting large or complex datasets, it is important that we review the transition from film to filmless imaging. The initial phase was based on film and used alternators and multiple view boxes that had been streamlined (but essentially unchanged in terms of the interpretation process) over the course of 100 years. The second phase used computer workstations to achieve “soft-copy” interpretation. This consisted of static interpretation that mimicked film-based reading using a tile- or frame-mode display of images on a monitor. The monitor displayed images in much the same way that cross-sectional studies were printed on film.
In phase three, interpretation was more dynamic, and radiologists learned to use window-level manipulation and other workstation tools with facility. The use of dynamic interpretation has been reported to have a significant effect on accuracy and productivity. A study performed at the University of Maryland demonstrated that the application of additional window/level manipulation resulted in improved conspicuity and characterization in 67% of abnormal CT examinations. 2 The authors also showed that rapid and dynamic manipulation of window/level settings substantially affected the final diagnoses in 18% of these cases without significantly affecting overall interpretation time.
Phase 4 is stack mode interpretation, the mode most frequently used by radiologists for soft-copy interpretation. With stack mode interpretation, images are stacked on top of each other and are revealed (in a manner similar to frames in a movie) as the radiologist scrolls through the dataset. Stack mode interpretation takes advantage of the visual system’s enhanced ability to detect motion or other changes in such a visual field compared with viewing stationary objects. Stack mode interpretation increases efficiency as well as accuracy when compared with tile mode interpretation. Further enhancements to this mode of interpretation have also resulted in improved workflow, such as the ability to link multiple stacks of images (current and prior studies or multiple series in a study). The use of predefined, modality- and organ system-specific display protocols has also impacted workflow. For example, a study performed at the Baltimore Veterans Affairs Medical Center (BVAMC) showed a greater than 10% improvement in reading time using simple default display protocols for intensive care unit portable chest radiographs (unpublished data).
|One of the benefits of volume rendering for musculoskeletal CT images is that it decreases the visibility of streak artifacts caused by metallic hardware.|
Radiologists across the country are finding that the first four phases of interpretation are inadequate for handling the large numbers of images generated from multislice CT systems. A routine CT of the thorax using an 8- or 16-detector scanner can generate 30 sheets of film each for lung, mediastinum, and bone window/level settings. Stack mode interpretation is not adequate for review of the 300 to 500 images acquired for a routine CT of the chest, abdomen, and pelvis and is even less so for the 1,500 to 2,000 images acquired for a CT angiogram “runoff” study. This is also true for other modalities. Current MRI studies with multiple dynamic imaging sequences, such as fMRI or cardiac MRI, cannot be adequately interpreted using stack mode interpretation. Even general radiography is becoming increasingly complex, especially with the introduction and gradual adoption of applications such as dual-energy subtraction and computer-aided detection (CAD).
SHIFTING INTO FIFTH
To cope with these increasingly complex images, we must shift to phase five of the interpretation process: volumetric navigation. The real power of a current generation digital modality such as multidetector CT can be realized only by abandoning the slice-based paradigm entirely and taking the plunge into a volumetric world, where the scan is a volume of information that covers the anatomy in question, not just a set of arbitrary sections through it. The volume of data acquired at the modality should be a navigation opportunity that allows the diagnosing radiologist or surgeon to choose the view he or she wants in order to select the correct two-dimensional (2D) cut or curved plane that most efficiently and accurately represents the morphology or finding in question. Once used in this way, all the cross-sectional knowledge base of radiology can be brought to bear on the phenomenally detailed volumetric information collected by modern scanners.
Unfortunately, most picture archiving and communications systems (PACS) vendors have not made the transition to volumetric interpretation. In fact, most of the development in volumetric navigation applications has been accomplished by stand-alone three-dimensional (3D) vendors. The result is a discordance between the product capabilities of PACS and 3D vendors. Most of the current 3D vendors target highly specialized practices used by dedicated personnel to perform image processing. Such advanced processing is most often performed by radiology technologists, who take images acquired at submillimeter collimation on a multislice CT, reconstruct them at thicker slices, and also create multiplanar and 3D images. This method works well for some specialty studies but is unsatisfactory for routine imaging. First, it requires a large amount of additional technologist time, which, given the current shortage in technologists, can be ill afforded by most institutions and practices. Second, technologists cannot predict how an individual radiologist wants to see specific sets of multiplanar images. Radiologists, therefore, may be constrained to review routine images in the axial plane at a predetermined and usually relatively thick plane of section, which negates the added value of the new generation of multidetector scanners. Finally, technologists must factor in anatomic variations during reconstruction to provide an acceptable plane of view, which in turn increases processing time because they now have to partially interpret the study. This method of preprocessing images by the technologists is not ideal for volumetric interpretation.
Moreover, these reconstructed images take up a good deal of archival, network, and workstation memory space, which, in turn, leads to poor network performance and increased cost of storage. Radiologists should have the flexibility of determining on a case-by-case basis whether images should be reviewed in the sagittal, coronal, or oblique planes at any arbitrary slice thickness or to display a study with a 3D perspective. In most radiology departments, these 3D workstations are not networked to each other, and comparison studies are rarely available because of limited archival space. Integration of these advanced workstation tools (which were typically built as features for dedicated modality workstations for technologists) into the workflow of a PACS can be very difficult. Nonradiology health care providers do not have access to the workstations and can see only rendered images that are pushed from the workstation to the PACS. As more clinicians and surgeons demand access to advanced image-processing applications, such a configuration becomes increasingly inefficient and expensive at the same time that it limits the accessibility of images.
Multiple studies and reports have indicated advantages associated with the use of advanced image processing techniques during routine review. A study performed at the BVAMC showed that, during routine sagittal review of body CT examinations, 19% of patients had what were deemed to be significant findings in the spine that had not been commented upon initially using only axial images.3 This confirms the importance of reviewing the spine in the sagittal plane to identify compression deformities and other significant pathology. Review of the spine in the sagittal and coronal planes provides a better assessment of spine alignment and dislocation and can be done quickly, because the number of images is relatively small compared with images reconstructed in the axial plane.
Studies have also shown the increased value of maximum intensity projection (MIP) images in the detection of pulmonary nodules when compared with routine axial images. 4 MIP processing reduces the number of overlooked small nodules, particularly in the central lung. Similarly, MIP and minimum intensity projection images of CT datasets may help in the diagnosis of a wide spectrum of diffuse lung diseases. 5
Three-dimensional volume rendered processing of vascular structures is also thought to be a helpful adjunct to interpretation of 2D images. Three-dimensional volume rendered images can provide an immediate assessment of vascular pathology so that more time can be spent quantifying a stenotic or aneurismal lesion. CT angiography with volume rendering enables accurate evaluation of vascular disease, even in the presence of dense calcifications. 6 Volume rendering also benefits musculoskeletal CT imaging, because it eliminates most streak artifacts caused by metallic hardware or large body habitus and produces high-quality images on which the relationships among hardware, bones, and bone fragments are well demonstrated. 7,8 Sagittal and coronal plane interpretation may also reduce interpretation time and improve diagnostic accuracy for evaluation of musculoskeletal and abdominal trauma. 9-11
The advantages accrued from volumetric interpretation do not come without a number of accompanying controversies and concerns. Are we trading image volume overload for clinical image information overload? These workstations readily provide images of the spine and other bones; pulmonary, abdominal, and pelvic vasculature; and other structures that are comparable with dedicated studies of the spine and CT angiograms. Are radiologists now “responsible” for detailed reports of the musculoskeletal system and spine and of the individual vessels now visualized on a routine “body” CT study? Should we specifically and routinely comment, for example, on the renal arteries, aorta and iliac arteries, superior and inferior mesenteric arteries? What are the implications of such capabilities on the time required to dictate a study? How should these cases be billed, given the fact that a single acquisition can generate many types of studies? Should subspecialists, such as angiographers or neuroradiologists, overread or review each routine CT of the thorax or abdomen? Only time will resolve these questions, through a process that is likely to be difficult and accompanied by discussions about the fundamental nature of the radiologist’s task in interpreting images.
|Studies have shown the increased value of MIP images in the detection of pulmonary nodules when compared with routine axial images. MIP processing reduces the number of overlooked small nodules, particularly in the central lung.|
PACS vendors who have not made the paradigm shift to phase 5 will find themselves unable to keep pace with the demands made by multidetector CT and other imaging modalities. If volumetric review is to truly take hold, it must be delivered directly into the diagnostic reading workflow, so that radiologists do not have to move from one location to another to access this new approach to managing huge datasets. Volumetric review may one day replace conventional diagnostic reading, but we must first live through an era in which traditional means are still available for reference, and, more important, all the other tasks a radiologist needs to do are properly addressed. The key is not only to incorporate these advanced image processing techniques in the PACS workstation but to address and streamline interpretation workflow.
The transformation of the radiology interpretation process will continue to evolve at a rapid pace. Although image navigation and enhancement will continue to improve (including better support for multimodality fusion, such as that with CT/PET), the next major phases will focus on decision-support tools, such as CAD, cueing, and more sophisticated integration with the electronic medical record. CAD programs will come into routine use in the next few years, especially in the detection of lung nodules and breast cancer. Image display systems in phase 6 of soft-copy interpretation will observe and “learn” the radiologist’s interpretation process and automate image data navigation and analysis. This next generation of intelligent workstations will be able to respond to a radiologist’s personal interpretation style and personality and change with the radiologist’s mood and emotional state. These developments will undoubtedly create additional challenges and concerns that will require the creativity and expertise of the entire medical imaging community.
Khan M. Siddiqui, MD, is chief, imaging informatics and body MR imaging, VA Maryland Health Care System, Baltimore.
Eliot L. Siegel, MD, is chief of imaging, VA Maryland Health Care System, Baltimore.
Bruce I. Reiner, MD, is director of research, VA Maryland Health Care System, Baltimore.
- Maynard CD. Radiology: future challenges. Radiology. 2001;219:309-12.
- Pomerantz SM, White CS, Krebs TL, et al. Liver and bone window settings for soft-copy interpretation of chest and abdominal CT. Am J Roentgenol. 2000;174:311-4.
- Gerber JD, Stallmeyer M, Siegel E, Obuchowski A. Spinal pathology assessed from routine multidetector body CT volumetric data via sagittal reconstructions. Paper presented at: Annual meeting of the American Roentgen Ray Society; Miami; May 5, 2004.
- Gruden JF, Ouanounou S, Tigges S, Norris SD, Klausner TS. Incremental benefit of maximum-intensity-projection images on observer detection of small pulmonary nodules revealed by multidetector CT. Am J Roentgenol. 2002;179:149-57.
- Bhalla M, Naidich DP, McGuinness G, Gruden JF, Leitman BS, McCauley DI. Diffuse lung disease: assessment with helical CT–preliminary observations of the role of maximum and minimum intensity projection images. Radiology. 1996;200:341-7.
- Leclerc X, Godefroy O, Lucas C, et al. Internal carotid arterial stenosis: CT angiography with volume rendering. Radiology. 1999;210:673-82.
- Pretorius ES, Fishman EK. Volume-rendered three-dimensional spiral CT: musculoskeletal applications. RadioGraphics. 1999;19:1143-60.
- Falchi M, Rollandi GA. CT of pelvic fractures. Eur J Radiol. 2004;50:96-105.
- Larici AR, Gotway MB, Litt HI, et al. Helical CT with sagittal and coronal reconstructions: accuracy for detection of diaphragmatic injury. Am J Roentgenol. 2002;179:451-7.
- Rozylo-Kalinowska I. The influence of application of 3D CT reconstructions on classification of maxillofacial fractures. Ann Univ Mariae Curie Sklodowska [Med]. 2002;57:549-55.
- Romano L, Giovine S, Guidi G, Tortora G, Cinque T, Romano S. Hepatic trauma: CT findings and considerations based on our experience in emergency diagnostic imaging. Eur J Radiol. 2004;50:59-66.