Clinical Importance on CT

Transcription

1 183 Truncated-View Clinical Importance on CT..,..,.... Artifacts:.... James L. Lehr1 A truncated-view artifact in CT is produced whenever any part of the patient or imaged object is present in some but not all of the views obtained for a slice. The potential to create images with this artifact exists for any CT scanner in which the fan beam (or its equivalent) does not cover the entire gantry aperture. This includes most CT systems currently on the market. Although the artifact may not create a severe visual disturbance in the image, It can alter the CT numbers in a manner that will compromise the accuracy of quantitative analyses. This report describes the nature of the truncated-view artifact and presents simulated examples for both mathematical phantoms and clinical scans. The artifact can be eliminated by assuring that the entire patient and all foreign objects are included in the field of view, or it can be minimized by placing objects that cannot be entirely within the field of view as close to the edge of the gantry aperture as possible. CT systems reconstruct images from many x-ray projections. Intuitively one might expect that these projections should be obtained from all possible directions through all parts of the object to be imaged, and that if some part of the object were not viewed from all directions, errors would result. This is in fact the case, and the reconstruction methods used to produce CT images assume that the object lies entirely within a region that is viewed from all directions [1 ]. In real CT systems, this assumption may not be correct, and errors or artifacts in the image then will be produced. Figure 1 represents a typical third-generation or fan-beam CT system in which the x-ray tube rotates around a circle of radius D and is coupled to a curved detector array which intersects an angle a of the fan beam. As the x-ray tube rotates, the fan beams will intersect to form a circle (of radius r in fig. 1 ) centered at the center of rotation. An object inside the circle will lie within the fan beam for all positions of the tube, and its projection will be obtained from all directions. Therefore, an object lying entirely within this circular field of view can be reconstructed accurately, if nonlinear effects such as beam hardening [2] and scatter [3] are neglected. On the other hand, if the gantry aperture is larger than the field of view, it is possible for an object (such as that labeled A in fig. 1 ) to be in the gantry but outside the field of view. The x-ray projection of such an object will be detected for some but not all positions of the tube-detector apparatus. For example, the object A in figure 1 will lie in the beam when the tube is in position 1, but it will lie out of the beam when the tube has rotated to position 2. A reconstruction Received October ; accepted after re- made from projection data collected under these circumstances will produce an vision February 4, image that contains artifacts [4-7]. One approach to this situation 15 to regard Chi- the object as being too large for the field of view, and this perspective leads one to speak of the fat-man artifacts which are produced by the oversized object. AJR 141: , July X/83/ Another approach 5 to note that If the detector array were longer, so that the American Roentgen Ray Society field of view would be larger, the artifacts would not occur. This leads one to

2 184 LEHR AJR:141, July 1983 Fig. 1 -Geometry of third-generation CT system. X-ray tube moving along outer solid circle of radius D produces detected fan beam of angle a which defines field of view indicated by dashed circle of radius r. If gantry aperture (inner solid circle of radius G) is larger than field of view, object inside gantry such as that labeled A will be detected for some positions of tube such as that labeled 1, but it will be undetected for other tube positions, such as that labeled 2. This produces truncated-view artifact. speak of incomplete data, missing projections, limited field of view, or simply truncated-view artifacts. The artifact also can be generated by parallel-beam and by fourth-generation inverted-fan-beam scanners under similar circumstances. Various methods to correct at least partially for truncatedview artifacts have been described in the physics literature [4-7], but no thorough assessment of the potential effect of such artifacts in the interpretation and analysis of CT studies has appeared in the clinical literature. This report focuses on these aspects by presenting several simulated examples of truncated view artifacts for both simple mathematical objects and clinical scans. In addition, practical methods to detect, minimize, or avoid the effects of such artifacts in clinical work are suggested. Materials and Methods The approach taken to study truncated-view artifacts for this report begins by collecting a complete set of projection values, that is, a set in which the view from each tube position includes the entire object. A simple computer program then simulates a smaller field of view by setting the projection values associated with a certain number of the outer detectors to zero. By zeroing a larger number of projection values associated with a larger number of detectors at the ends of the array, the program simulates narrower fan beams and smaller fields of view. These truncated views are then used to create a reconstructed image that produces the artifacts that would be obtained if the outer detectors were not present. Fig. 2.-Clinical scans reconstructed from complete projections covering shoulders in pediatric patient displayed with wide window (window level 0, 53-cm field of view indicated by superimposed circle. A, Scan of abdomen in window width 2000). obese patient (window level 0, window width 51 2). B, Chest scan at level of

3 AJR:141, July 1983 Cl TRUNCATED-VIEW ARTIFACTS 185 C Fig. 3.-Images of mathematical 2-cm disk with CT number 91 2 H located cm from center of rotation in background of air. A, Reconstruction for 53-cm-diam field of view which corresponds to gantry aperture indicated by superimposed circle. Graph of CT numbers along dashed line passing through disk and center of rotation shows no truncated-view artifact (window level 0, window width 1 024). B, Reconstruction made from truncated views corresponding to 24-cm-diam field of view indicated by inner circle. (Outer circle again indicates gantry aperture.) Graph of CT numbers along dashed line shows that CT numbers in area of disk are too small (about -400 H) and that The objects for which truncated-view images have been produced include both mathematical phantoms and patient scans. Patient scans that were used included an abdominal scan in an obese patient(fig. 2A) and a chest scan atthe level ofthe shoulders in a small pediatric patient (fig. 2B). The mathematical phantoms comprised small and large circular disks, for which complete projection values were computed using appropriate formulae. The advantages of using mathematical phantoms are that the size, D CT numbers in air near disk are too high because of truncated-view artifact (window level , window width 32). C, Artifact image formed by subtracting A from B. Graph of CT numbers shows negative artifact in location of disk and positive artifact (about +600 H) near disk, but falls off nearly to zero at center of rotation (window level 0, window width 32). D, Same artifact image as C with graph of CT numbers along dashed line shown extending diagonally from center of rotation to just beyond field of view. Largest artifact of about 70 H occurs at periphery of field of view (window level 0. window width 32). location, and CT value of the object can be controlled relatively precisely, and that the simulated projection values do not include nonlinear effects such as beam hardening [2] or scatter [3]. The small disks had a diameter of 2 cm and a CT number of 912 Hounsfield units (H). They were placed in a background of air at various distances from the center of notation (fig. 3A). A small disk should approximate the effects of a foreign object such as a syringe containing dilute contrast material or an extremity in the gantry

4 186 LEHR AJR:141, July 1983 aperture. The large disk had a CT value of 0 H and a diameter of 52 cm. It was centered at the system s center of rotation, again in a background of air. This disk should approximate the effects of scanning a very large patient. The scans on patients were performed on a Siemens Somatom II system in which the field of view corresponds to the gantry aperture. The abdominal scan used a slice thickness of 8 mm, a peak voltage of 1 25 kev, and 360 views of 51 2 projections, each collected in 5 sec. The chest scan used a thickness of 4 mm, a peak voltage of 1 25 kev, and 720 views of 51 2 projections, collected in 1 0 sec. The mathematical scans were computed for 720 views. For all objects, reconstructions have been made at 256 x 256 pixels for the entire field of view, a circle with a diameter of 53 cm. Truncated projections corresponding to smaller fields of view then were used to reconstruct images with the truncated-view artifacts. Subtracting the image produced with complete views from the image made with truncated views produces an image of the artifact. Images were photographed at selected settings, often with superimposed graphs of pixel values produced by analysis programs supplied with the CT system in order to demonstrate the magnitude of the artifact. All neconstructions were made using the filtered backpnojection software supplied with the Somatom II system and using the standard body-mode kennel. For some images comesponding to smaller fields of view, an extended scale option, which doubles the upper CT number range so that it reaches 2048 H, was used to avoid or minimize overshoot artifacts which occur when the computed CT number is larger than H. In addition, the average value of five pixels at the center of the artifact image was computed. This is referred to as the central error. Results The image reconstructed from complete views of a 2-cmdiam mathematical disk with its center located cm from the center of rotation is shown in figure 3A. The large circle superimposed on this figure shows the location of the gantry aperture, which is the same as the field of view. The superimposed graph shows the CT numbers along the dashed line passing through the center of the disk and the center of rotation. In this and similar graphs CT values that exceed the maximum or minimum values on the scale are shown as the maximum or minimum values, respectively. The image of this disk reconstructed from views that have been truncated to produce a limited field of view 24.0 cm in diameter is shown in figure 3B, along with a graph of CT values and larger and smaller circles representing the gantry aperture and field of view, respectively. Note that the window width has been narrowed and that the scale of the graph is different from the one in figure 3A. Because the projection of the disk is detected in some but not all of the truncated views, its density is not reconstructed accurately but is smeared out over a larger area. The artifact image formed by subtracting figure 3A from 3B is shown in figure 3C, again with superimposed circles and a graph of Cl values along the horizontal diameter of the image. It is apparent that most of the dense part of the artifact occurs in streaks extending from the disk tangent to the field of view. FIgure 3D shows the same artifact with a superimposed graph of the CT values along a line extending radially from the center of the field of view through the 0 Li 0 C a, r6cm 0 r:9cm 0 r:i2cm & A r:l8cm r:24cm. I I I I I Disk Distance, d (cm.) Fig. 4.-Central error generated from mathematical 2-cm disk with CT value of 91 2 H lying outside various-sized fields of view as function of distance between center of disk and center of rotation. For fixed field of view, central error decreases (or remains constant) as disk is moved further from center of rotation. For fixed distance between disk and center of rotation, central error increases as field of view becomes smaller. upper streak. The magnitude of the artifact along this line falls off rapidly from its maximum value of about 60 H near the edge of the field of view to a value of about 2 H near the center. Although the artifact outside the field of view is sometimes negative, the artifact within the field of view is always positive or zero. Because CT reconstruction is a linear operation, the magnitude of the truncated-view artifact should be a linear function of the CT number of the disk. Varying the CT value of a disk confirms that the magnitude of the artifact at any point is, as expected, directly proportional to the disk s CT number plus The additive term of is needed because in the Hounsfield scale, air, which has a linear attenuation of zero, is assigned the value of units. A CT value in attenuation units (AU) may be defined as the CT number in Hounsfield units plus In this scale air has a value of 0 AU, water has a value of AU, and the disk would have a value of AU. A disk with half the linear attenuation of that shown in figure 3A would then have a CT value of /2 or 856 AU, corresponding to a CT number of -44 H. If such a disk were scanned under the conditions for figure 3B, the magnitude of the artifact would be half as large, but the spatial distribution would be the same as that shown in figure 3C. In a series of experiments, the distance of the 2-cm disk from the center of rotation and the size of the limited field of view were varied (fig. 4). For a fixed size of the field of view, the total central error decreases (or remains constant) as the disk is moved away from the center of rotation. For a fixed disk location, the central error increases as the diameter of the field of view decreases. The large disk and clinical scans were reconstructed for limited fields of view of various sizes (figs. 5-7). Figure 5A I A

5 AJR:141, July 1983 Cl TRUNCATED-VIEW ARTIFACTS 187 shows the image of the 52-cm, water-density (0 H) disk reconstructed from truncated views corresponding to a limited field of view 48 cm in diameter. The only artifact that can be seen is an increase in the CT numbers near the edge of the field of view. The graph of CT numbers across the horizontal diameter of the image shows that the magnitude of the artifact near the edge of the limited field of view is about H, but that the central error is nearly 0 H (actually 1 2 H). Because the large disk has a CT number of 0 H, the limited-field-of-view image and the artifact image are identical inside the limited field. The image of the large disk for a limited field of view 24.0 cm in diameter (fig. SB) shows a more pronounced artifact. This image was made using the extended scale option. Even then, the artifact near the edge shows an overshoot effect, and one can state only that the edge artifact is greater than 2048 H. The central artifact is H. Limited-view images and superimposed graphs of the corresponding truncated-view artifacts for the abdominal scan in an obese patient (fig. 2A) are shown in figure 6. The image for a 48-cm-diam field of view (fig. 6A) shows a small edge artifact with a maximum value of about H and a central error of only 2.6 H. Indeed, if the patient had been more carefully centered in the aperture, the truncated-view artifact could have been eliminated. For the 24-cm diam field of view (fig. 6B) the artifact is much more pronounced, with an overshoot artifact indicating a maximum value of more than H near the edge and 243 H near the center. The 1 8-cm-diam field of view (fig. 6C) has a severe cupping artifact that again exceeds H near the edge of the field and is 509 H at the center. In all truncated-view images Fig. 5.-Limited-field-of-view reconstructions of mathematical 52-cmdiam disk with density corresponding to 0 H. A, Reconstruction made from truncated views corresponding to 48-cm-diam field of view indicated by outer margin of disk s image. Graph of CT values along horizontal diameter mdicated by dashed line shows that artifact is primarily evidenced by cupping near edge of field of view (window level 0, window width 51 2). B, Reconstruction made with extended CT scale option from truncated views corresponding to 24-cm-diam field of view indicated by outer margin of disk s image. Graph shows much more pronounced cupping artifact due to smaller field of view and to larger amount of disk lying outside field of view (window level 1400, window width 1024). produced from this scan, the artifact is slightly asymmetrical, being greater in the upper left part of the field. This is because the patient is not centered, so that slightly more of the out-of-field density lies beyond the upper left border of the field of view. Limited-view reconstructions of the pediatric chest scan (fig. 2B) are shown in figure 7. A 36-cm-diam field of view was used for figure 7A. The artifact shows high-density streaks near the edge of the field and resembles the sum of two small disk artifacts with the disk placed in the location of the patient s shoulders. The absence of out-of-field matenial above or below the patient leads to virtually no artifact along a vertical line through the center of rotation. The 18- cm-diam field of view (fig. 7B) is just large enough to include cross sections of both lungs, but as indicated by the graph of the artifact, if the limited-field scan were used to estimate the CT number of, say, a pulmonary nodule, the results would be incorrectly high by about 200 H for a nodule near the midline and by about H for a peripheral nodule. The 1 2-cm-diam field of view (fig. 7C) is associated with a severe artifact. Figure 8 shows x5 reconstructive magnifications of the central part of the abdominal scan (fig. 2A). It is important to distinguish between a limited field of view and a restricted area of reconstruction [8] which may be used to produce magnification images. No truncated-view artifact is produced as long as the magnified reconstruction is made from projections comprising complete views. Figure 8A shows a x 5 reconstructive magnification of the central part of the abdominal scan (fig. 2A) made from complete views and using the extended scale option. No artifact is present.

6 188 LEHR AJR:141, July 1983 Fig. 6.-Limited-field-of-view reconstructions for abdominal scan shown in fig. 2A. A. Reconstruction for 48-cm-diam field of view indicated by inner circle. Graph shows CT values of artifact image (not shown, but formed by subtracting fig. 2A from fig. 5A) along dashed line extending from upper left to lower right part of gantry aperture (outer circle). Artifact appears as increased density near upper left edge of field of view where patient contour extends beyond field (window level 0, window width 51 2). B, Reconstruction for 24-cm-diam field of view. Graph again shows magnitude of artifact along line extending diagonally across field. Artifact is more severe with central error of 243 H and value exceeding H at periphery. Window center has been raised to compensate for artifact (window level 300, window width 512). c, Reconstruction for 1 8-cm-diam field of view. Graph shows very severe artifact of about 509 H at center which again exceeds H near edge of field (window level 700, window width 51 2). Figure 8B shows a x 5 reconstructive magnification made from truncated views corresponding to a limited field of view 1 2 cm in diameter. The graph of artifact values shows a central error of about H and a relatively wide peripheral rim in which the artifact exceeds 2048 H. The image has been displayed at the same window width as figures 8A and 2A, in an attempt to maintain contrast, but the window level has been raised to H to compensate for the central error. However, the cupping artifact is so severe that only structures in the central part of the image are visible. Figure 8C shows the same image as figure 8B displayed at a wider window of 2000 H. Although more of the structures are visible, overall image contrast is greatly reduced. Discussion The largest error due to truncated-view artifact will occur near the edge of the field of view. Indeed, this may be the only part of the artifact seen under normal viewing conditions. Although it is possible to conceal this edge effect by computing reconstructed values over a region smaller than the field of view, the artifact will still be present in the central part of the image as a relatively constant value added to the CT numbers. One result of this additive constant will be to decrease image contrast. Although in theory this can be compensated for to a large extent by raising the window level and narrowing the window width when the image is

7 AJR:141, July 1983 CT TRUNCATED-VIEW ARTIFACTS 189 C displayed, this maneuver will also increase the visual impact of any cupping effect. Another result of the truncated-view artifact will be to frustrate attempts to perform quantitative analyses, because the CT number will depend not only on the absorption properties at the point represented, but also on the density, composition, and location of those parts of the patient or of other objects that lie outside the field of view. There are many benefits that can be achieved by using a limited field of view [9]. Certainly in a very obese patient, a Fig. 7.-Limited-field-of-view reconstructions for chest scan in fig. 2B. A, Reconstruction for 36-cm-diam field of view. Graph shows CT values along horizontal diameter indicated by dashed line in artifact image (not shown) formed by subtracting fig. 2B from fig. 7A. Artifact is primarily peripheral and in region of shoulders, which lie partially outside field of view (window level -300, window width 2000). B, Reconstruction for 18-cm-diam field of view. Graph shows more pronounced artifact with central error of about 200 H and value exceeding H at edge of field (window level , window width 2000). C, Reconstruction for 1 2-cm-diam field of view. Graph of artifact values shows central error greater than 500 H and 1 -cm-thick rim at periphery of field of view in which artifact exceeds H (window level 200, window width 2000). CT scan containing a truncated-view artifact would be more useful than no scan at all, and this may justify the use of a large gantry aperture. Furthermore, those parts of the patient that lie outside the x-ray beam in some of the views obviously will receive a lower dose. Finally, if the field of view is reduced by moving the x-ray tube closer to the patient while maintaining the same detected fan-beam angle (e.g., by moving the detector array an equal distance away from the center of rotation), the x-ray tube output can be used more efficiently, the sampling geometry will improve, and the ratio of scattered-to-primary photons will decrease [3]. Achieving these benefits without incurring artifacts is the goal of the correction techniques that have been proposed [4-7]. However, such techniques either are inexact or create additional technical problems in the construction of CT systems, and to my knowledge, no CT system that

8 190 LEHR AJR:141, July 1983 f &d*....:1 incorporates an accurate truncated-view correction procedure is available commercially. Because most CT systems currently on the market have a gantry aperture that is larger than the field of view, they are subject to truncated-view artifacts. Although this may be obvious from reviewing the system s technical specifications, it may be simpler to perform scans of reference objects (e.g., air or a water phantom) both with and without a syringe or a small bottle filled with diluted contrast material located at various distances from the center of the aperture. By subtracting the images formed without the object in the aperture from those with the object present, artifact images can be obtained and compared to those presented above (fig. 3). If a scanner is subject to truncated-view artifacts, the Fig. 8.-Reconstructive magnifications X 5 of central part of abdominal scan in fig. 2A. All reconstructions have been done using extended scale option which permits maximum CT value of A. Reconstruction from complete views. No artifact is present (window level 0, window width 51 2). B, x5 Reconstruction made from truncated views corresponding to 1 2-cm-diam field of view. Graph shows values of artifact along horizontal diameter indicated by dashed line (window level 1 200, window width 51 2). C, Same reconstruction and graph as B displayed at wider window setting to make more of structures visible. Note loss of image contrast (window level 1 600, window width 2048). artifacts can be avoided by carefully centering the patient in a large enough field of view to encompass both the patient and any other needed objects such as pads, blankets, restraining devices, drains, and intravenous tubing. If it is necessary to have some part of the patient, such as an extremity, or some foreign object inside the gantry aperture but outside the field of view, the magnitude of the artifact can be reduced by placing those objects as far from the center of rotation as possible (i.e., at the edge of the gantry aperture). When interpreting CT scans, the radiologist should watch carefully for increased Cl numbers near the outer edge of the reconstructed image, indicating the possibility that a truncated-view artifact is present. However, if the entire field of view is not reconstructed, the only indication of the

9 AJR:141, July 1983 Cl TRUNCATED-VIEW ARTIFACTS 191 artifact s presence may be a relatively uniform increase in the CT numbers which is difficult to detect visually. If a large enough area can be reconstructed to demonstrate that the entire patient is in the field of view, and if it is known that no other objects were in the gantry aperture during the scan, then the radiologist can be certain that no truncated-view artifact is present. Unless this is the case, radiologists should be quite cautious about basing any clinical opinion on quantitative analysis of CT numbers in the image. ACKNOWLEDGMENTS I thank Paul Davis who wrote the truncating program, and Jeffrey Crass, John Fennessy, Paul Frank, and Jonathan Rubin for review of the manuscript. REFERENCES 1. Herman GT. Image reconstructions from projections. New York: Academic, 1980:68 2. Stonestrom JP, Alvarez RE, Macovski A. A framework for spectral artifact connections in x-ray CT. IEEE Trans Biomed Eng 1981 ;28: Johns PC, Yaffe J. Scattered radiation in fan beam imaging systems. Med Phys I 982;9 : Herman GT, Lewitt AM. Evaluation of a preprocessing algonithm for truncated CT projections. J Comput Assist Tomogr 1981;5: 1 27-i Nalcioglu 0, Cho ZH, Lou RY. Limited file of view neconstnuction in computer tomography. IEEE Trans NucI Sci 1979;26 : Wagner W. Reconstructions from restricted region scan datanew means to reduce the patient dose. IEEE Trans NucI Sci 1 979;26 : Lewitt AM. Processing of incomplete measurement data in computed tomography. Med Phys 1979;6:4i Lehn JL, Davis P. CT magnification techniques. In: Proceedings of the seventh conference on computers in radiology. Chicago: American College of Radiology, 1982: , Zonneveld FW Jr. Computed tomography. Eindhoven, Netherlands: Philips Medical Systems, 1980:26

10 This article has been cited by: 1. J. Liebregts, T. Xi, R. Schreurs, B. van Loon, S. Bergé, T. Maal Three-dimensional virtual simulation of alar width changes following bimaxillary osteotomies. International Journal of Oral and Maxillofacial Surgery 45:10, [CrossRef] 2. Roya Zandparsa Latest Biomaterials and Technology in Dentistry. Dental Clinics of North America 58:1, [CrossRef] 3. Prashant P. Jaju, Mayuri Jain, Ajita Singh, Akanksha Gupta Artefacts in cone beam CT. Open Journal of Stomatology 03:05, [CrossRef] 4. AK Hunter, WD McDavid Characterization and correction of cupping effect artefacts in cone beam CT. Dentomaxillofacial Radiology 41:3, [CrossRef] 5. DAVID D. BARBEE, JENNIFER R. ALLEN, PATRICK R. GAVIN COMPUTED TOMOGRAPHY IN HORSES. Veterinary Radiology 28:5, [CrossRef]