Abdom Imaging 27:275–283 (2002) DOI: 10.1007/s00261-001-0169-6
Abdominal Imaging © Springer-Verlag New York Inc. 2002
Reader strategies for CT colonography E. G. McFarland Mallinckrodt Institute of Radiology, Washington University, 510 South Kingshighway Boulevard, St. Louis, MO 63110, USA
Computed tomographic (CT) colonography currently faces the critical challenge of a rapidly evolving technique being validated in generalizable environments. These environments include the evaluation of the data by less expert readers who have undergone training, with standardized image display and reader protocols. This article reviews the current image display techniques most widely used, the reader protocol for detection and characterization of colorectal lesions, and standardization issues for reporting findings. Important issues of test performance expectations are also reviewed, which greatly affect the development and optimization of the reading paradigm.
Image display techniques In an era of continual advancements of volumetric image acquisition capabilities, image display techniques are another area of rapid flux. Multiple image display techniques are currently available to evaluate CT colonography image data [1– 4]. One of the most widely used image display techniques in CT colonography is two-dimensional multiplanar reformation (2D MPR) [5–12]. Concordant with the input of volumetric image data, this display extends beyond standard axial slice-based imagery and provides a time-efficient, seamless interactivity of axial, coronal, and sagittal views in a soft copy viewing environment (Fig. 1). Most investigators of CT colonography would state that this image display technique is the primary method used, given its multiple advantages. These advantages include ease of use, wide availability, and high sensitivity to detect polyps with adequate CT data (e.g., CT data not hindered by colonic collapse, excessive fluid, or stool retention; Fig. 2). Improved characterization is provided by the ability to simultaneously cross-register across the three image planes for each focal finding. If a finding is visualized and appears to be present in at least two of the three image planes, one’s confidence that this represents a focal finding is increased. Improved orientation of the
location of focal findings is provided by the extralumenal orientation of 2D MPR, which can be a strong advantage in areas of collapse or tortuosity. Important information of the inherent density of focal findings is provided by use of defined display window width and level settings. A standardized high-contrast display window width and level (at Washington University, we use a width of 1500 and a level of ⫺200) is best to impart optimal contrast for polyps within the air–soft tissue interface. This is also the best display to detect focal pockets of air within lesions, which is strongly characteristic of stool. Use of a standard soft tissue display window width (400) and level (10) is useful to better characterize mural involvement of more advanced lesions and fat content of lipomas. This is also the display window setting used to evaluate extracolonic findings. In addition to single-slice 2D MPR, three-dimensional (3D) MPR is a further application that provides a thickslab image display technique with a capability of simultaneous viewing in the axial, sagittal, and coronal planes (Fig. 1D). This can be displayed with a monochromatic or color opacity assignment laid into the thick slabs. This has the advantage of the extralumenal orientation of 2D MPR, with a finite volumetric sliding stack of the data providing imagery beyond a single slice. Prior investigations have shown good reader results with this technique [8], and further investigation is warranted. The evolution of 3D endoscopic rendering techniques has evolved to volume rendered techniques in CT colonography [1, 13]. Beyond the threshold dependency of shaded surface rendered techniques, volume rendered techniques have the inherent advantage of displaying a more inclusive range of the voxels within a data volume. The additional advancements of endoscopic perspective volume rendered imagery permit differentiation of nearfield from far-field objects, which can be useful in endoscopic applications. Although the initial technologic advancements of 3D endoscopic CT imagery attracted wide attention, it has also created some of the largest perceived gaps in capabilities between experienced and inexperienced radiologists.
276
Fig. 1. Illustration of an 18-mm sigmoid polyp (arrows) at CT colonography, using interactive 2D and 3D image display techniques. A Axial 2D multiplanar reformation (MPR). B Coronal 2D MPR. C Sagittal 2D MPR. D Axial thick-slab 3D MPR. E 3D endoscopic perspective volume rendered (PVR).
E. G. McFarland: Reader strategies for CT colonography
E. G. McFarland: Reader strategies for CT colonography
Fig. 2. A small 4-mm polyp (arrowhead) in the transverse colon is clearly depicted with the axial 2D MPR image (single detector CT, 5 min slice), in a region with excellent insufflation and minimal fluid and stool retention.
Currently the value of the 3D endoscopic view is the ability to further characterize focal findings detected from the 2D MPR evaluation [6, 8]. One of the most common dilemmas in CT colonography is the differentiation of a nodular fold from a focal polyp on a fold (Fig. 3). Although 2D MPR can be useful to cine through a region and differentiate between these two entities, the 3D endoscopic view adds value by improving characterization. The 3D endoscopic views alone, however, are currently hindered by more time-consuming and tedious navigation through the length of the colon. As advances in navigational tools [14, 15] and extended field of view displays [16] continue, the diagnostic performance of the 3D endoscopic image display for improved characterization and detection will merit further systematic investigation.
Reader protocol Given the current image display techniques, the most commonly used reader protocol is primary use of 2D MPR to detect and localize focal lesions, with selective use of 3D endoscopic displays to characterize those entities further. This process involves the key components of systematic navigation through the entire colon, followed by detection and characterization of each focal finding. Systematic and continuous navigation through the colon is the primary and most challenging skill to attain. The colon is a discipline—a long obedience in the same direction. The direction starts from the rectum, with con-
277
tinuous retrograde evaluation of the colon to the cecum. This style provides a logical organization to cover the entire surface area. It also provides an early indication of the degree of stool retention in the rectosigmoid colon, which may influence the evaluation of more proximal focal findings. The ability to navigate and efficiently visualize the colonic surface area is greatly aided by the facile use of 2D MPR. Initially, a 2- to 3-min scroll through the volumetric data of the axial, coronal, and sagittal planes can give a quick overview of the colonic anatomy. Then starting from the rectum, typically in the axial plane, contiguous navigation through each section of the colon is performed. As each section of the colon is evaluated, complete cine motion through the tubular contour of each wall is done to visualize the entire surface area. This is analagous to real-time ultrasound evaluation, where a complete sweep through an organ is performed to completely view the entire volume of interest. If there is disorientation of what the path of the next contiguous colonic segment is, cross-registration in the other two MPR planes can greatly facilitate orientation. Certainly future image processing advancements will aid the navigation process, with tools to better guide the path and illustrate the completeness of the evaluation of the visualized surface. The ability to detect and characterize lesions along the path of the colon during navigation is the next task. Different styles can be used to achieve this, which may depend on differences in image quality and number of detected focal findings. One strategy is to stay in the initial MPR plane (typically axial), from start to finish of the entire colon, for the first viewing of the colon. With this method, detected lesions would be initially marked with electronic arrows imbedded into the data. After completing the entire colonic survey in this first MPR plane, each marked lesion detected would be systematically evaluated with use of the other 2D MPR planes and the 3D endoscopic views in the supine and prone datasets. Another reading strategy is to start in continuity from the rectum to the cecum in the initial MPR plane of choice, and each lesion detected along the path is marked and evaluated using the other MPR planes and 3D views. After completion of the initial evaluation with the axial MPR, it may be helpful to then briefly sweep through the coronal and sagittal planes (with embedded arrows of the detected lesions still in place) to evaluate whether any more lesions can be detected. Typically, the high-contrast display window settings are used to detect the polyps. However, the soft tissue window display settings can be helpful to characterize specific colonic findings, such as higher-grade lesions with mural involvement. The logical strategy of how to completely survey the colon for the prone and supine datasets needs to be strategized and optimized within the specific workstation environment being used. The workstation environments
278
E. G. McFarland: Reader strategies for CT colonography
Fig. 3. Detection and improved characterization between a 12-mm sigmoid polyp and a thick haustral fold. A On the axial 2D MPR image, the polyp (arrow) appears as a sessile lesion along the wall of the colon. B On the sagittal 2D MPR image, the thick fold (arrowhead) and polyp (arrow) are similar in appearance. C On another sagittal 2D MPR image, the haustral fold (arrowheads) is seen stretching across the lumen. D A 3D endoscopic view clearly depicts the focal morphology of the polyp (arrow). E The 3D endoscopic view clearly characterizes the fold (arrows) (used with permission from McFarland et al. [8]).
E. G. McFarland: Reader strategies for CT colonography
vary from single to dual monitors, with different abilities to simultaneous view or register between supine and prone datasets [17]. Although novel advancements in the workstation environments to facilitate these tasks are being developed, the essential components of the organization and discipline inherent in the reader deserve emphasis. Reader training protocols to efficiently train and calibrate the performance of less-experienced readers is a current area of importance. Critical to dissemination is the ability for less-experienced readers in community environments to become adept with a new technique. Development of a training set of cases with different sizes and morphologies of lesions will be important. After introduction of the computer software skills for 2D MPR and 3D visualization techniques, the navigation skills and time-efficient strategies for lesion detection and characterization are learned. The number and type of cases needed to train a reader are being investigated. Computeraided diagnosis has shown promising early results in aiding reader evaluations in CT colonography applications [18]. Further comprehensive evaluation of computer-aided diagnosis among readers of different levels of experience will be important to determine whether increased sensitivity is achieved without a decrease in specificity.
Reporting of findings Each significant lesion detected must be properly characterized, localized, and measured. The reporting of findings is an essential final endpoint of the CT colonographic evaluation. Although still evolving, the issues involved in the reporting of findings are critical and will enable radiologists to continue to play a primary role in CT colonography. One of the most important aspects of the reporting of findings, generalizable to other 3D imaging techniques, is that the radiologic report provides a service to the subspecialists and has components that add value beyond existing capabilities. After standardization, the CT colonographic report can consistently and logically provide a widely accessible and comprehensive road map of the intralumenal findings, augmented with images providing an extralumenal registration. These advantages extend beyond the selective record of individual endoscopic images provided by an endoscopist at the time of a colonoscopy, where orientation of focal findings within tortuous colonic anatomy can be challenging. These CT radiologic reports might provide a longitudinal archive of imagery, which may be especially helpful in specific cohorts of patients who undergo frequent surveillance. The first component of the CT colonography report is identification of each focal intralumenal finding. The intralumenal findings should consistently contain specific
279
information about the location, morphology, density, and size of each lesion of interest. A sample of a potential CT colonographic report is shown in Figure 4, which is time and space efficient and comprehensively locates and characterizes each lesion. After the interactive evaluation of the volumetric data containing hundreds of source images, selected hard copy or electronic images are provided with specific descriptive annotations. Selective 2D MPR views are given for each finding, which may consist of an axial view and a longitudinal view, such as a coronal or sagittal 2D MPR. This helps to confirm the focality of the lesion in two views and provides additional localizing information. The measurement of each lesion has a profound effect on clinical management and requires a well-defined protocol. It also adds another component in which CT colonography can provide more accurate and reliable information than existing modalities. Prior studies have documented differences in reliability of different endoscopic methods of measurement. In a study of 100 polyps comparing different endoscopic methods of measurement to the reference method of an in vitro ruler, mean differences in measurement ranged from 3.4% with a linear probe to 12.3% with the forcep method [19]. We previously documented very good intra- and interobserver agreements of the measurement of CT colonography by using a defined measurement protocol with high-contrast display window width (1500) and level (⫺200); measurement of the long axis along the base of the head of the polyp with exclusion of the stalk, was performed (McDermott RA, McFarland EG, Brink JA, et al. Evaluation of polyp size measurement in multi-observer prospective study of CT Colonography. Society of Computed Body Tomography, 2001). The flexibility to cross-register within the center of the lesion simultaneously in the three MPR planes and then choose the optimal plane for measurement provides a facile and reliable manual method of measurement. Further automation of the CT measurement with image processing advancements may extend the reliability and accuracy of the measurement. The second major component of the radiologic report is a set of overview images of the entire colon, with arrows demarcating where the focal findings are. Similar to the advantages provided by the barium enema, CT colonography can provide a road-map of the entire colon using an edge-enhanced algorithm (Fig. 4G–H). With the same volumetric data acquisition, a different opacity assignment can be seamlessly applied to selectively assign opacity to the boundary voxels of the edge interface of the colonic mucosa, resulting in CT imagery that simulates a barium enema [20]. These summary images can be extremely helpful to give global orientation of the focal findings relative to the entire colon. In addition, these images can provide useful anatomic information of areas with marked tortuoisity, which can aid navigation for the endoscopist during colonoscopy (Fig. 4G–I).
280
E. G. McFarland: Reader strategies for CT colonography
E. G. McFarland: Reader strategies for CT colonography
281
Fig. 4. Potential CT colonography report page of colonic findings, with images of individual focal findings, followed by overview images of each polyp’s location and colonic anatomy. A Lesion 1 is a 20-mm proximal transverse colon polyp (axial 2D MPR; arrow). B Lesion 1 (arrow) in a coronal 2D MPR view. C Lesion 1 (arrow) in 3D endoscopic view. D Lesion 2 (arrows) is a circumferential, advanced, wall cancer, measuring 31 mm in width (coronal 2D MPR view). E Lesion 2 (arrowheads) from an axial 2D MPR view, with soft tissue window level, demonstrates mural involvement of cancer. F Lesion 2 (arrows) from a 3D endoscopic view demonstrates lumenal compromise. G Edgeenhanced 3D overview image (simulating barium enema) of entire colon demarcates locations of lesions 1 and 2 (arrows). H Edge-enhanced 3D view in right posterior oblique (RPO) projection. I Edge-enhanced 3D view in LPO projection. H, I These views provide overall orientation of the colonic anatomy in areas of tortuosity.
The third component of the CT colonography report that needs to be established is the appropriate and consistent reporting of extralumenal findings. Hara et al. showed the incidence and follow-up of extracolonic findings on CT colonography by using a low-dose technique (70 mA) [21]. In that study of 264 patients, 30 (11%) very important findings, 49 patients (17%) had moderately important findings, and 55 patients (21%) had minimally important findings. Most of the extracolonic findings in that study were of low importance, and overall the find-
ings did not prompt extra, unnecessary, expensive examinations. Although the benefit of early detection of significant disease is an advantage, the potential costs of additional, unnecessary examinations from unimportant findings could greatly offset cost effectiveness on a broad scale. Using CT acquisition protocols with effective 20 –50 mA, excellent image quality of the intralumenal, high-contrast depiction of polyps has been achieved [22, 23]. However, the increased noise imparted in the low-contrast interfaces might cause difficulties when
282
E. G. McFarland: Reader strategies for CT colonography
Fig. 5. Extracolonic findings of a renal mass (arrows) on CT colonography (100 mA, effective). A Axial 2D MPR image shows upper pole renal mass on the left. B Follow-up ultrasound clearly delineates an anechoic mass with through transmission, consistent with simple renal cyst.
evaluating the extralumenal findings. One problem could be the evaluation of the very common renal cyst, which can appear as a solid mass with low-dose techniques, and require ultrasound for further evaluation (Fig. 5). Of note, in the study by Hara et al, only two of 79 patients required ultrasound to confirm the cystic nature of renal lesions [21]. During the evaluation of whole-body screening applications, there are growing concerns that “more could be less” if well-defined protocols are not established to safeguard against the excessive costs of unnecessary additional examinations. Further investigation with CT colonography in larger cohorts will be needed to estimate the incidence of extracolonic findings and costs for additional examinations.
priori risk, age, and comorbidity of the patient. Hence, we have the dilemma of determining an optimized reader protocol that is strongly influenced by test expectations that are not clearly standardized. To standardize performance evaluation and properly guide the dissemination of CT colonography, multidisciplinary collaboration and evidence-based study designs will be essential [26 –27]. With guidelines established from multidisciplinary collaborations, expectations of reader performance can be better defined. Evidence-based medicine provides a framework for implementing efficient study designs to help estimate diagnostic test performance at a societal level. Following further validation, CT colonography could become an effective image-based modality to help decrease colorectal cancer mortality.
Future issues of test performance expectations The reading paradigm of any evolving imaging technique is shaped by expectations about test performance. The evolution of CT colonography is influenced largely by the expectation of what sizes and morphologies of polyps are important to detect and report. The standardization of these expectations are greatly challenged by the continual change of acquisition and image display capabilities. The reading paradigm to a large part depends on whether the expectation is to find every polyp or only “clinically significant” lesions. If the task is to find even dimunitive polyps (ⱕ5 mm in diameter), this can promote greater reader fatigue, decreased time efficiency, and more false positives. The difficulty is that the lower limit of what constitutes a clinically significant polyp size is controversial [24 –28] and certainly would depend on a
Acknowledgments. I thank Tom K. Pilgram, Ph.D., and James A. Brink, M.D., for their editorial assistance of this manuscript and their continued support throughout the years of our efforts at Washington University.
References 1. Rubin GD, Beaulieu CF, Argiro V, et al. Perspective volume rendering of CT and MR images: applications for endoscopic imaging. Radiology 1996;199:321–330 2. Reed JE, Johnson CD. Virtual pathology: a new paradigm for interpretation of computed tomographic colography. In: Hanson KM, ed. Medical imaging 1998: image processing. Washington, DC: SPIE, 1998:439 – 449 3. Paik DS, Beaulieu CF, Jeffrey RB Jr, et al. Vizualization modes for CT colonography using cylindrical and planar map projections. J Comput Assist Tomogr 2000;24:179 –188
E. G. McFarland: Reader strategies for CT colonography 4. Wang G, McFarland EG, Brown BP, et al. GI tract unraveling with curved cross-sections. IEEE Trans Biomed Eng 1998;17:318 –322 5. Hara AK, Johnson CD, Reed JE, et al. Detection of colorectal polys with CT colography: initial assessment of sensitivity and specificity. Radiology 1997;205:59 – 65 6. Dachman AH, Kuniyoshi JK, Boyle CM, et al. CT colonography with three-dimensional problem solving for detection of colon polyps. AJR 1998;171:989 –995 7. Royster AP, Fenlon HM, Clarke PD, et al. CT Colonoscopy of colorectal neoplasms: two-dimensional and three-dimensional virtual-reality techniques with colonoscopic correlation. AJR 1997; 169:1237–12426 8. McFarland EG, Brink JA, Heiken JP, et al. Spiral CT colonography: reader reliability and diagnostic performance with 2D and 3D image displays. Radiology 2000;218:375–383 9. Macari M, Milano A, Lavelle M, et al. Comparison of time-efficient CT colonography with two- and three-dimensional colonic evaluation for detecting colorectal polyps. AJR 2000;174:1543–1549 10. Rex DK, Vining D, Kopecky KK. An initial experience with screening for colon polyps using spiral CT with and without CT colography (virtual colonoscopy). Gastrointest Endosc 1999; 50:309 – 313 11. Fletcher JG, Johnson CD, Welch TJ, et al. Optimization of CT colonography technique: prospective trial in 180 patients. Radiology 2000;216:704 –711 12. Yee J, Akerkar GA, Hung RK, et al. Colorectal neoplasia: performance characteristics of CT colonography for detection in 300 patients. Radiology 2001;219:685– 692 13. McFarland EG, Brink JA, Loh J, et al. Visualization of colorectal polyps with spiral CT colography: evaluation of processing parameters with perspective volume rendering. Radiology 1997;205:701– 707 14. Paik DS. Automated flight path planning for virtual endoscopy. Med Physics 1998;25:629 – 637 15. Summers RM. Navigational aids for real-time virtual bronchoscopy. AJR 1997;168:1165–1170 16. Beaulieu CF, Jeffrey RB Jr, Karadi C, et al. Visualization modes for CT colonography, part II: blinded comparison of axial CT, virtual endoscopy, and panoramic-view volume rendering. Radiology 1999;21:203–212
283 17. Reed JE, Johnson CD. Image display system for synchronous interpretation of supine and prone computed tomographic colography. In: Hanson KM, ed. Medical imaging 1998: image processing. Washington, DC: SPIE, 1998:306 –315 18. Summers RM, Beaulieu CF, Pusanik LM, et al. Automated polyp detector for CT colonography: feasibility study. Radiology 2000; 216:284 –290 19. Gopalswamy N, Shenoy VN, Choudhry U, et al. Is in vivo measurement of size of polyps during colonoscopy accurate? Gastrointest Endosc 1997;46:497–50219 20. McFarland EG, Brink JA. Spiral CT colonography (virtual colonoscopy): challenge between advancing technology and generalizability. AJR 1999;173:549 –559 21. Hara AK, Johnson CD, MacCarty RL, et al. Incidental extracolonic findings at CT colonography. Radiology 2000;215:353–357 22. Hara AK, Johnson CD, MacCarty RL, et al. CT colonography: single-versus multi-detector row imaging. Radiology 2001;219: 461– 465 23. Macari M, Bini EJ, Xue X, Milano A, Katz S, Resnick D, Chandarana H, Klingenbeck K, Krinsky G, Marshall CH, Megibow AJ. Prospective comparison of thin-section low-dose multislice CT colonography to conventional colonoscopy in detecting colorectal polyps and cancers. Radiology 2002 (in press) 24. Read TE, Read JD, Butterly LF. Importance of adenomas 5 mm or less in diameter that are detected by signoidoscopy. N Engl J Med 1997; 336:8 –12 25. Rex DK, Cummings OW. The controversy regarding distal hyperplastic polyps. Gastrointestinal Endosc Clin North Am 1993;3:639 – 649 26. Rex DK, Smith JJ, Ulbright TM, et al. Distal colonic hyperplastic polyps do not predict proximal adenomas in asymptomatic averagerisk subjects. Gastroenterology 1992;102:317–319 27. Glick S. Double-contrast barium enema for colorectal cancer screening: a review of the issues and a comparison with other screening alternatives. Am J Gastroenterol 2000;174:1529 –1537 28. McMahon PM, Bosch JL, Gleason S, et al. Cost-effectiveness of colorectal cancer screening. Radiology 2001;219:44 –50 29. Sonnenberg A, Delco F, Bauerfeind P. Is virtual colonoscopy a cost-effective option to screen for colorectal cancer? Am J Gastroenterol 1999;94:2268 –2274