Huge sets of slices will transform interpretations

As we prepare to celebrate our 10th year anniversary with PACS and filmless operation at the Baltimore VA Medical Center, we look back on our trip from film-based to filmless imaging and try to predict where we and the medical imaging community will be heading in the next few years. The acronym TRIP, for transforming the radiology interpretation process, came out of a recent strategic planning meeting of the Society for Computer Applications in Radiology (SCAR).

Filmless radiology has allowed us to reinvent an inefficient workflow process, particularly the radiology interpretation aspect. During the past 10 years, we and other early adopters observed and documented the transformation of the radiologist's interpretation process.[1,2] Changes occurred in progressive phases:

1. Film-based interpretation uses alternators and multiple viewboxes.

2. Static soft-copy interpretation uses workstations rather than film. In this phase, cross-sectional images are still presented in static tile mode similar to film. Radiologists typically request four or more monitors to simulate film alternators and viewboxes as closely as possible. They judge the image quality of conventional radiographs and cross-sectional images based on how closely they could simulate the "look" of film.

3. Dynamic interpretation of images characterizes the next phase. Radiologists learn to rapidly, even unconsciously window and level and to use other workstation tools. Pomerantz et al at the University of Maryland demonstrated that the application of additional window/level settings resulted in improved conspicuity and characterization in 67% of CT examinations that were abnormal, and that their use substantially affected the final diagnosis in 18% of these cases without a significant increase in interpretation time.[3]

4. Stack mode characterizes the next phase in image interpretation for CT and other cross-sectional studies taking advantage of the visual system's enhanced ability to detect motion or other changes in a visual field. Sequential images within a study or sequence are "stacked" and reviewed as a "movie," permitting review of multiple sequences or comparison of current and prior studies using only one or two monitors.

The use of stack mode has been shown to increase interpretation efficiency as well as accuracy. Beard et al from the University of North Carolina School of Medicine and Mathie and Strickland at the Hammersmith Hospital in London documented improved speed and performance in stack mode in comparison with tile mode.[4,5] Linked stack mode is a further enhancement that synchronizes multiple stacked images (e.g., MRI sequences) within a single examination or across a current and one or more prior examinations.

Today's radiology residents and their counterparts in medical and surgical subspecialties learn to interpret radiographs using stack mode without having to unlearn the film-based paradigm. This gives them a nice advantage over those of us who have had to unlearn the interpretation process using film, employing such antiquated skills as the use of a hot light, positioning of film at an oblique angle to see life support lines, and the use of a minifying lens to better detect lung nodules.

5. Volumetric navigation characterizes the fifth phase in the TRIP, in which the process of image review is separated from the manner in which the images were acquired and reconstructed. This phase has been accelerated by the extraordinarily rapid transition to the use of multidetector CT scanners. Radiologists across the country are finding the first four phases of interpretation inadequate for the large numbers of images generated from these systems. A routine CT of the thorax using a multidetector system can generate 30 sheets of film each for lung, mediastinum, and liver settings. Our experience has found even stack mode insufficient for review of the 300 to 500 images acquired for a routine CT of the chest or the abdomen and pelvis and even less sufficient for the 1500 to 2000 images acquired for a CT angiography runoff study.

COPING WITH IMAGE OVERLOAD

Radiologists use several strategies to cope with this image overload. The most common is to acquire images using a multi-detector scanner with thin collimation and then reconstruct the images sent to PACS using much thicker (e.g., 5 mm or 8 mm) sections, resulting in a three- to 10-fold reduction in the number of images sent to PACS. Technologists can then use a dedicated CT workstation to perform additional reconstructions or rendering. These can include sagittal and coronal reconstructions of the spine, CTA of the vasculature, coronal images of the bowel for CT colonography, fly-through images of the colon for colonoscopy, and 3D rendering.

This approach is unsatisfactory for several reasons: It requires a large amount of additional technologist time, especially for angiographic rendering, analogous to the extra time required for the technologists to produce films in multiple window/level settings. Due to the complexity and time required, technologists only perform this rendering in a small percentage of cases. The reconstructed images unnecessarily take up a good deal of archival, network, and workstation memory space. Radiologists should have flexibility from case to case to determine whether the images should be reviewed in the sagittal or coronal or oblique planes or using a 3D perspective, but they are constrained to review images in the axial plane at a predetermined, usually relatively thick, plane of section, which negates the added value of the new generation of multi-detector scanners.

Volumetric navigation using an advanced workstation frees the radiologist from the shackles of conventional CT, which has traditionally been limited to sending axial images at a fixed slice thickness to a film printer or a PACS. An image of the spine, for example, can be rapidly and interactively rendered and reviewed as a sagittal or coronal data set at any slice thickness. This can be augmented with 3D and maximum intensity projection (MIP) rendering as well. The area being examined and the clinical history can determine viewing perspective. The pulmonary arteries, for example, can be reviewed using relatively thick slice coronal or oblique perspectives, with or without the use of MIP rendering. The colon, in our experience, is best depicted in the coronal plane, while the liver and spleen may be best reviewed in the axial plane and improved with the use of MIP. The vasculature of the thorax, abdomen, pelvis, and other areas may be optimally displayed according to their orientation within the body and are also probably best rendered as a MIP image.

The paucity of data in the medical imaging literature on the clinical value of advanced workstation tools is surprising, with the exception of MIP processing for lung nodule detection. Gruden et al found that the use of a MIP algorithm resulted in increased ability to detect central pulmonary nodules for both "senior" and "junior" readers as well as an improved ability to detect peripheral nodules for their junior readers.[6] We believe that similar advantages will be found for other advanced workstation algorithms such as ray sum and minimum intensity projection, as well as routine application of multi-planar imaging, and we plan to further investigate these. Once considered by radiologists to be eye candy for referring clinicians and patients, 3D imaging can be very useful in providing a general survey of an area and for portraying anatomic structures such as ribs that course in an oblique plane.

Although volumetric navigation, the fifth phase of our TRIP, has tremendous potential, it poses some unique and daunting challenges as well, especially the concern that we might be trading image volume overload for clinical image information overload. We can use these workstations to render images of the spine and other bones, pulmonary and abdominal and pelvis vasculature, and other structures that are comparable to dedicated studies of the spine and CT angiograms. Our abdominal and thoracic subspecialists have asked whether they are 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 they specifically and routinely comment, for example, on the renal arteries, aortic and iliac arteries, superior and inferior mesenteric arteries? What are the implications on the time required to dictate a study? How should these cases be billed, when 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?

Another major challenge is access to these workstations and their integration into the workflow of a PACS. In most departments, high-end workstations are much more expensive than typical PACS workstations. These workstations are usually not networked to each other, and comparison studies are rarely available due to the limited archival space. Nonradiology healthcare providers do not have access to the workstations and can see only rendered images that are pushed from the workstation to the PACS. On the whole, the typical setup is inefficient and expensive and provides limited accessibility to images.

At the Baltimore VA, we use an enterprise-wide 3D/multiplanar solution in which a separate, 3D/multiplanar "PACS" provides the horsepower for advanced processing using a client/server model. Approximately a dozen PC clients can share the server at a time, allowing each generic "vanilla" PC connected on the network to function as though it had the power and speed of a high-end 3D/multiplanar advanced workstation. All CT studies are pushed to the central server and thus made available to PCs located throughout the department and hospital. This system has been received enthusiastically by our medical and surgical colleagues, who have learned to use the workstation effectively in their own subspecialty areas for clinical and educational purposes. Although we have not performed a formal study, utilization of CT seems to have increased slightly due to the perceived advantages of 3D and multiplanar review by the vascular surgeons, orthopedic surgeons, podiatrists, and others. The nonradiologists, in particular, seem to appreciate and readily adapt to the more intuitive perspective provided by 3D, MIP, and multiplanar rendered images.

Some of our radiology colleagues have expressed discomfort with the practice of giving nonradiologists access to these volumetric navigation capabilities, suggesting that the perceived added value and control of the radiologist will be diminished. Part of their apprehension is based on the belief that the cross-sectional images are relatively obscure and thus need to be interpreted by a radiologist, unlike 3D and multiplanar images. Clinicians may be less wedded than radiologists to axial sections and need not unlearn a dependence on axial anatomy.

A similar concern was also expressed in the radiology community 10 years ago when PACS was introduced. A few early PACS adopters in academic radiology departments decided that images should not be made available outside the radiology department to clinicians at all, or that they should be made available only after a report was generated and signed. Just as most departments have decided to grant nonradiology healthcare providers access to images prior to a report being generated, it seems inevitable that nonradiologists will eventually be given access to the full volumetric data set.

Perhaps the biggest barrier to the transition to volumetric navigation has been the lack of integration of this capability in the current generation of PACS workstations. It is not practical for a radiologist interpreting a study using a PACS workstation to walk (or even slide) over to a dedicated 3D/multiplanar workstation for each case. Image navigation is typically not the linear, sequential process of review of a set of axial images?it may be performed in a haphazard fashion, with a radiologist reviewing a portion of a data set in one plane and other portions using other views. A fascinating exhibit at the RSNA meeting's InfoRad 2002 addressed this problem by using a cube that was progressively filled in to keep track of which anatomic areas had been reviewed during the course of an interpretation session.

A number of PACS vendors, including ours at the Baltimore VAMC, are working on an integrated solution in which volumetric navigation would become a routine part of the radiologist display or hanging protocols. The extraordinary challenge to vendors and the radiology community will be to make these powerful tools available in the routine reading process but not add significantly to the time required to interpret the study.

This phase of the radiology interpretation process, when integrated into the routine "hanging protocol" for a CT study of the chest, for example, could follow the sample user-selected display protocol:
A slowly rotating 3D rendered color image of the chest and upper abdomen appears automatically on the right upper third of the large LCD monitor. For the first few seconds, it appears as a surface reconstruction, and then it changes to a 3D rendering that depicts the bones, heart, and other vasculature of the chest and upper abdomen.

Additional views representing the coronal, sagittal, and axial planes appear in three other windows. These begin to slowly display sequential images in a cine fashion until the radiologist moves a cursor over one of these windows (e.g., the coronal view). This window then replaces the larger 3D rotating image, as the radiologist quickly moves through the stack of coronal images, automatically updating the other windows.

The radiologist then reviews the axial images using a thick-slab MIP algorithm in order to discern small lung nodules. Images from a previous CT of the thorax are displayed in a linked fashion on a second monitor.

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 CT/PET, the next major phases will focus on decision support tools such as computer-aided detection (CAD) and cuing and intelligent applications of informatics. Computer-aided cuing may take many forms, including an overlay in which microcalcifications are circled on a mammogram and lung nodules are colored in a shade of red that depends on their probability of malignancy.

CAD programs will come into routine use in the next few years, especially in the detection of lung nodules and breast cancers. Clinical information from the electronic medical record, results of previous examinations, and clinical and imaging expert systems and the indication for the current study will optimize image navigation and computer cuing and CAD programs and suggest diagnostic possibilities. Comparison with large computerized reference image data sets may also be used routinely by radiologists to facilitate more rapid, accurate diagnosis. These future additions to armamentarium of the radiologist will also create additional challenges and concerns.

TRIP will be a major theme at this year's SCAR meeting in Boston, and the challenge of the information explosion in radiology will be the focus of what promises to be a superb closing session. Although our continuing TRIP will take us into new territory, we believe that the radiology community will be able to stay in the driver's seat as long as we keep our eyes and our minds open to change and continue to allocate time and energy and resources to investigate the roads that lie just around the bend.

References
1. Siegel E, Reiner B. Work flow redesign: the key to success when using PACS. AJR 2002;178(3):563-566.
2. Bennett WF, Vaswani KK, Mendiola JA, Spigos DG. PACS monitors: an evolution of radiologist's viewing techniques. J Digit Imaging 2002;15[Suppl]1:171-174.
3. Pomerantz SM, White CS, Krebs TL, et al. Liver and bone window settings for soft-copy interpretation of chest and abdominal CT. AJR 2000;174(2):311-314.
4. Beard DV, Molina PL, Muller KE, et al. Interpretation time of serial chest CT examinations with stacked-metaphor workstation versus film alternator. Radiology 1995;197(3):753-758.
5. Mathie AG, Strickland NH. Interpretation of CT scans with PACS image display in stack mode. Radiology 1997;203(1):207-209.
6. Gruden JF, Ouanounou S, Tigges S, et al. Incremental benefit of maximum-intensity-projection images on observer detection of small pulmonary nodules revealed by multidetector CT. AJR 2002;179(1):149-157.

Dr. Siegel is vice chair of information Systems University of Maryland Department of Radiology and Chief of Radiology for VA Maryland Healthcare System, and Dr. Reiner is Director of Research at the VA Medical Center both at the Baltimore VA Medical Center.