Medical Dosimetry, Vol. 19, No.4, pp. 223-226, 1994 Copyright © 1994 American Association of Medical Dosimetrists Printed in the USA. All rights reserved 0958-3947/94 $6.00 + .00
EVALUATION OF TABLETOP MATERIALS FOR AUTOCONTOURING IN CT TREATMENT PLANNING INDRA J. DAs, PH.D., and KIARAN MCGEE Department of Radiation Oncology, Fox Chase Cancer Center, 7701 Burholme Avenue, Philadelphia, P A 19111
BETH SWANN-D'EMILIA, C.M.D.,
Abstract-Computerized tomography (CT) treatment planning has been proven to be essential in precision radiotherapy. Typically CT planning requires a large number of CT scans (between 3 and 50 slices) to be entered into the treatment planning system. A significant amount of time in the planning process is devoted to outlining internal structures, organs, and the external patient contour in each CT slice. In principle, exterual contours could be generated easily using autocontouring routines; however, in reality it does not always provide a satisfactory contour due to the limitations of the CT tabletop having nearly the same CT number as that of the body structure. A solution to this problem is to create a large CT number gradient interface between the patient and CT tabletop by inserting a thin sheet of low CT number material. The optimum material for a tabletop was investigated from a range of low-density and low atomic number media. Various materials were studied by placing them underneath an unsliced humanoid phantom. A portion of the phantom abdomen was imaged and analyzed on a Picker Premier IQ CT scanner. Results indicate that the tabletop should be made of the material that has a CT number at least 10 times lower than the tissue in contact with the table. A simple and cost-etTective method of avoiding failures in autocontouring is to place a thin sheet of low-density material such as cardboard or foam board on the tabletop. Such an insert creates a large CT gradient resulting in a significant improvement in the accuracy of the edge detection algorithm used for autocontouring. Detailed analysis is presented. Key Words: CT, Autocontour, Tabletop, Treatment planning.
can take, from clinical experience, over 1 hour. Although internal structures usually require manual contouring, the external patient contour can be generated with the help of edge detection techniques. 5 - 9 These techniques require a significant CT number gradient between the structure being contoured and its surrounding in order to delineate it. When the CT numbers of the tabletop and the tissue in contact with the tabletop are similar, edge detection methods fail. The failure in autocontouring is more pronounced in body sections with external air cavities as well as those with large diameters. In these instances contours are either modified, edited, or reentered manually. The manual contouring process could take several minutes for each slice, adding a significant burden and time commitment. This additional time spent outlining by the dosimetrist is not practical, because one must also outline as many as eight internal structures on each slice. The time spent outlining is the single most burdensome factor in 3-D treatment planning. In a busy department, such a practice could not be tolerated. It seems logical that autocontouring is an essential tool for CT treatment planning. The success of an autocontouring routine depends on the site and the quality of the tabletop as well as the CT number gradient at the patient/table interface. Most manufacturers use a tissue equivalent material for the tabletop to eliminate image distortion, which
Previous studies have shown that CT-based treatment planning can produce a better radiation treatment in terms of target localization and normal tissue sparing l ,2 as well as an improvement in clinical outcome. 3 As a result, the use of CT images for treatment planning has increased significantly for curative treatments. These images are also the backbone of any three dimensional (3-D) radiotherapy treatment planning system, since they provide these systems with the 3-D patient data set that allows a range of functions to be performed. These include dose calculations, dose volume histogram analysis, and surface and volume rendering of the target and the critical normal tissues. In the past CT data has been obtained from traditional CT scanners, but recently dedicated scanners and computer workstation systems known as CT simulators4 have been developed. CT scans are contoured either at the time of simulation using a CT simulator or later, after being transferred to the treatment planning system. The process of contouring should take several minutes but for curative cases, with up to 50 CT slices and multiple organ and structure contours, the process
Reprint requests to: Beth Swann-D'Emilia, C.M.D., Department of Radiation Oncology, Fox Chase Cancer Center, 7701 Burholme Avenue, Philadelphia, P A 19111.
Volume 19, Number 4, 1994
Table 1. Characteristics of the materials used for the insert. Materials
Density (kglm 3 )
Cost ($/m 2 )
Tabletop Lucile Plywood Styrofoam Foamboard
0.78 0.63 2.5 0.32
1180 750 12.5 135
660 10 20 5
is very critical for diagnostic radiology. This is in contrast to the requirement of having a large CT number gradient between a structure and its surroundings (such as a patient external contour and the CT tabletop) in order for autocontouring to be successful. Although it is impractical to replace the scanner table material, the CT gradient at the patient/table interface can be maximized by inserting a sheet of material between the patient and tabletop prior to scanning. The insert material should have the following characteristics: I. 2. 3. 4. 5.
Readily available Inexpensive Nontoxic to skin Low Z (Z < tissue) Lowest backscatter factor
6. Durable/rugged for frequent usage 7. Easy to machine or cut to fit the CT table The efficacy of various materials for use in external autocontouring for treatment planning is investigated in this paper.
MATERIALS AND METHODS A Picker IQ Premier™ CT scanner (Picker International, St. Davids, PA 19087) was used for scanning the thoracic region of a humanoid unsliced phantom. To maximize the CT gradient (air gap) between the scanner tabletop and the phantom's surface, an Alpha Cradle (Alpha Cradle, Smithers Medical Products Inc. Tallmadge, OH) immobilization cast was created. The
Fig. l. A CT slice of humanoid phantom placed on the CT tabletop along with the outcome of the autocontouring using an improper mask. It shows the distorted contour covering the entire table and alpha-cradle immobilizer. The small squares located along the contour are the vertices of the polygon that define the contour.
Autocontouring in CT planning. B.
casted phantom was placed on the CT tabletop, and nine slices were obtained. The scanning process was then repeated exactly with the casted phantom placed on the various insert materials. Based on the characteristics defined in the introduction section, Lucite, plywood, Foamboard (Foamboard, Beinfang, Hunt, Statesville, NC 28677), and Styrofoam (Styrofoam, Dow Chemical, Midland MI 48674) were used in this study. Some of the properties of these materials are listed in Table 1. Each study was analyzed using the diagnostic functions incorporated into the CT scanner. A circular region of interest (ROI) covering 26 pixels was selected for averaging the CT numbers of the material and the phantom at the interface between the two. The slices from each scan study were then transferred to an image file server via an Ethernet network interface, which were in tum read into a ROCS (ROCS, Radiation Oncology Computer Systems, Carlsbad, CA 92(09) treatment planning system. The autocontouring process was investigated for each slice for a fixed mask suitable for clinical use. The mask is a CT number threshold where values below a certain CT number are typically displayed as a color (green, blue, red, etc.) for quick visualization by the operator. Autocontouring then involves tracing the external contour by tracking the interface between CT numbers above and below
the mask until the routine reaches the point at which it started. To evaluate the effect of the various insert materials on autocontouring, a qualitative score of pass or fail was assigned to each study. A pass score means the ROCS autocontouring routine successfully contoured the external patient contour, while a fail means the process was unsuccessful and the routine contoured not only the patient external contour but the CT tabletop also. A typical failed autocontour is shown in Fig. 1 while the pass autocontour is shown in Fig. 2.
RESULTS For autocontouring the external contour, a fixed mask corresponding to the density of 0.29 g/cm3 and a CT number of 50 was used on the ROCS treatment planning system. The CT slices were read into the treatment planning system blindly to eliminate any personal bias. In each slice a score of pass, fail, or semifail was used. The scores were then averaged for nine slices of each study. Table 2 shows the compiled results of the CT numbers from the scanner and the blind score from the treatment planning system. It was observed that autocontouring failed when the CT tabletop only (no insert) was used. Also, some of the inserts
-- - - '"-
Fig. 2. The outcome of autocontouring using an appropriate mask. The humanoid phantom is placed on top of the Foamboard. It shows an accurate contour of the phantom slice. The small squares on the contour are for the visual purpose of showing the vertices of the polygon that define the contour.
Volume 19, Number 4, 1994
Table 2. Analysis of the CT numbers and the outcome. Material Tabletop Lucite Plywood Styrofoam Foamboard
CT number of tissue in contact 56::':: 56::':: 44 ::':: 44::':: 48::'::
13 13 19 12 12
did not help in contouring, whereas others did. The striking observation was that if the ratio of CT numbers of the phantom and the insert medium was of the order of 10, autocontouring was successful. These materials were Foamboard, plywood, and Styrofoam. This suggests that any material with air or CT numbers similar to that of air could be used for the insert. The thickness of the material is not critical as long as it can create a good interface with the weight of the patient. CONCLUSIONS
Based on the criteria set in the introduction section, foamboard was found to be the most cost-effective insert medium for autocontouring. This material is simply a layer of foam sandwiched between two layers of cardboard. It provides a cushion and comfort to the patient. If this material is not available, any rigid foamy medium such as Styrofoam, which is commonly used for molding cerrobend electron cut-outs, could be used. If any other medium is chosen as an insert, the CT number of the material as a simple rule of thumb should be at least 10 times lower than tissue.
CT number of table top
-227::':: 12 -275::':: 51 -484::':: 38 -939::':: 27 -767 ::':: 7
-4.1 -4.9 -11.0 -21.0 -16.0
Fail Fail Pass Pass Pass
REFERENCES 1. Jani, S.K. CT Simulation for radiotherapy. Madison, WI: Medical Physics Publishing; 1993. 2. Goitein, M. The utility of computed tomography in radiation therapy: an estimate of outcome. Int. 1. Radiat. Oncol. BioI. Phys. 5:1799-1807; 1979. 3. Hobday, P.; Hodson, N.J.; Husband, J.; Parker, R.P.; Macdonald, J.S. Computed tomography applied to radiotherapy treatment planning: techniques and results. Radiology 133:4777-482; 1979. 4. Sherouse, G.W.; Bourland, D.; Reynolds, K.; McMurray, H.L.; Mitchell, T.P.; Chaney, E.L. Virtual simulation in clinical setting: some practical considerations. Int. 1. Radiat. Oncol. BioI. Phys. 19:1059-1065; 1990. 5. Roy, J.N.; Steiger, P.; Ling C.c. Overlap detection and contourtracking algorithms for critical organs-application to kidney. Compo Med. Imag. Graphics 14:153-161; 1990. 6. Udupa, J.K. Interactive segmentation and boundary surface formation for 3D digital images. Compo Graph. Image Process. 18:213-235; 1982. 7. Mills, P.H.; Fuchs, H.; Pizer, S.M.; et al. IMEX: A tool for image display and contour management in a windowing environment. In: Schneider, R.H.; Dwyer, S.J.; Jost, G.R., editors. Proceedings of Medical Imaging III: image capture and display. Newport Beach, CA: SPIE Vol 1091; 1989:132-142. 8. Canny, J. A computational approach to edge detection. IEEE Trans. Pat. Anal. Mach. Intell. 8:679-698; 1986. 9. Kass, M.; Witkin, A.; Terzopoulos, D. Snakes; active contour models. Int. 1. Compo Vision. 1:321-331; 1987.