Ini I Rodiarron Oncology EmI Phys.. Vol. 22, pp. 1139-l 146 Printed in the U.S.A. All rights reserved.
0360.3016/92 $5.00 + .OO Copyright 0 1992 Pergamon Press Ltd.
0 Technical Innovations and Notes ASSESSMENT OF GEOMETRIC TREATMENT ACCURACY USING TIME-LAPSE DISPLAY OF ELECTRONIC PORTAL IMAGES LAWRENCE E. REINSTEIN, PH.D.,
SUJATHA PAI, M.S. AND ALLEN G. MEEK, M.D.
Departmentof Radiation Oncology, University Hospital at Stony Brook, Stony Brook, NY 1 I794 During the past two years, several electronic portal imaging systems have been introduced to the market by therapy accelerator manufacturers and other vendors. While these systems differ substantially in their detection technology, they are all capable of displaying portal images on a video screen in near real-time, and of creating multiple static (or “movie”) images during each treatment. Major questions confront the users of such systems as to the best utilization of this wealth of information, and to its value in comparison to traditional weekly portal film methods. Using an “in-house” video based system, a new technique was established to aid in the assessment of on-lineiimages so that immediate “go/no-go” decisions can be made by the therapy technologist. A video “movie-loop” is displayed which consists of the static image of the initial (approved) set-up, and the current treatment image. Multiple~images of successive treatments can also be viewed in this “time-lapse” display mode to provide a quick visual means for review of an entire course of therapy. The on-line imaging system hardware is composed of a combination copperplate/fluorescent-screen detector, a front surface mirror angled at 45” to remove the camera from the direct radiation beam, and a high sensitivity SIT video camera. This assembly is attached to a rigid base and mounted directly to the isocentric gantry. The geometry is fixed to within +1 mm and assures the precise day-to-day reproducibility which is necessary for the success of the time-lapse display technique. Experience with this technique shows it to enhance the user’s ability to notice small changes in patient’s position with respect to the radiation field. Radiation treatment sites reviewed using this procedure were Hodgkin’s (mantle), Lung, Brain and extremities. Shifts in patient position on the order of several millimeters were readily detectable, as will be demonstrated in thi$ paper. Somewhat surprisingly, grosser movements (>I cm) were also noted despite overall technical excellence as assessed by weekly portal filming. The eye senses day-today movement with greater ease when the fields are seen in timelapse display than when compared as discrete portal images. Ultimately, persistent movement appreciated on the time-lapse display can suggest the need for a change in patient set-up or immobilization technique. Portal, Imaging, EPID, On-line imaging, Quality assurance.
INTRODUCTION
localization errors (on the order of 1 cm) occur relatively frequently ( lo%-36% depending upon treatment site). These studies imply a need for large treatment margins which increase the target volume, thus running counter to the aforementioned strategy. Somewhat encouraging, however, are the results of Marks et al. (5), which show significant reduction in the rate of localization errors as the frequency of portal/localization films is increased (a drop from 36% of the cases to 15% of the cases in errors greater than 1 cm by increasing the number of portal films during a course of treatment from 9 to 24). These results suggest the need for increasing the frequency of portal/ localization imaging to, perhaps, a daily basis. Although daily portal imaging may be desirable from the point of view of improved treatment accuracy, such a policy is by no means accepted practice. A study (8) of 158 institutions indicates that even weekly filming has not been universally accepted, with fewer than half of the institutions reporting that weekly films are taken on the
One of the goals of modem radiation therapy is to improve the patient’s local control probability by increasing the therapeutic ratio in order to deliver a higher tumor dose without producing unacceptable complications. New diagnostic modalities have made more precise tumor localization possible and, when combined with “state of the art” 3-D treatment planning, let the radiotherapist design a treatment volume that conforms closely to the actual tumor volume. By using custom shaped blocks, multileaf collimation, and dynamic therapy techniques, the volume of normal tissue being irradiated can be significantly reduced. Although tumor localization and dose calculation capabilities have advanced, the ability to deliver treatments with the same degree of precision on a daily basis over an entire course of therapy has not. Several historic studies have demonstrated (1, 2, 5-7) that clinically significant
Presentedat the 32nd Meeting of ASTRO, Miami, FL, 1519 October 1990.
Reprint requests to: L. E. Reinstein, Accepted 1139
for publication
6 September
Ph.D. 199 1.
I. J. Radiation Oncology 0 Biology 0 Physics
Volume 22, Number 5, 1992 up accuracy, giving the technologist the ability to make a go/no-go decision.
METHODS
Fig. 1. The prototype accelerator gantry.
EPID detector
assembly
mounted
to the
majority of their patients. Reluctance to take frequent portal films is based upon the inconvenient and timeconsuming nature of the process; they impede patient flow and, therefore, increase the cost of treatment. Thus, a major motivation for the development of realtime on-line imaging systems (also known as electronic portal imaging devices [EPID]) has been to make daily portal imaging both expedient and practical. Various commercial and prototype EPID’s are currently available which, ideally, would allow the accelerator technologist to take daily portal images with very little extra time or added inconvenience. But, given this new capability, it now becomes important to ask how to make best use of the suddenly overwhelming number of portal images without over-stressing the system? What is the most effective way to evaluate these daily images to improve the quality of treatment? We have introduced a new technique, analogous to “time-lapse” photography, which addresses these questions. By rapidly displaying a sequence of images taken on consecutive treatment days, a “movie-loop” is created which dramatically brings out any change in patient or block position. When reduced to a two-image sequence (the first being the “approved” and the second the current treatment image), the resulting “flip-book” like animation is a sensitive indication (approximately +2 mm) of set-
Table 1. Results of the test for accuracy
AND MATERIALS
Electronic portal imaging The on-line electronic portal imaging system is videobased, consisting of a 2 mm thick copper plate backed with a GdzOzS:Tb Phosphor (250 mg/cm’), a 45” front surfaced mirror used to remove the video camera from the direct radiation beam, and a silicon intensified target (SIT) video camera with an Fl.4 lens (9). (The field of view in the plane of the isocenter is approximately 2.5 cm X 30 cm.) The video signal is digitized using an image processing system* interfaced to a computer.+ This particular image processor allows software control of the gain and offset, important for optimization of the image acquisition. It also contains a frame buffer board which consists of two 5 12 X 5 12 X 8 bit image planes. The imaging software package (OPIUM)* includes a variety of routines for image acquisition, frame averaging, non-uniformity corrections, windowing, and contrast enhancement ( 10). SpeciJic requirements for time-lapse display For the “time lapse” display technique to work effectively, the following system requirements must be met: For the “flip-book” technique (alternation of the “approved” and current image) the frame buffer need only be capable of storing two images for immediate sequential display. In the system described, however, as many as sixteen 256 X 240 X 8 bit images can be displayed in rapid sequence in a “movie-loop” playback. (Although the resolution is slightly reduced its effect is virtually unnoticeable in the movie-loop mode.) This enables the user to review the day-to-day shifts for virtually an entire treatment course. &. Acquired images must be of similar radiographic quality. For this technique to be effective, successive images must have a similar spread of gray scale values. As a result any image-to-image changes are only caused by
of electronic
portal image (EPI) detector
positioning
Pixel position
Image no. 1
Image no. 2
Image no. 3
Image no. 4
Image no. 5
Origin (5 cm, 0) (-5 cm, 0) (0, 5 cm) (0, -5 cm)
143, 132 185, 129 102, 135 141,79 145, 186
143, 132 185, 129 101, 135 141,79 145, 185
143, 132 185, 129 101, 136 141,79 145, 186
143, 132 185, 129 101, 135 141,79 145, 185
143, 132 185, 129 101, 135 141,79 145, 185
* Imaging Technology Series 15 1. +IBM AT. *On-line
portal
imaging,
University
of Manitoba.
The de-
velopment of the portal imaging system was in collaboration with S. Shalev, Manitoba Cancer Treatment and Research Foundation.
Geometric treatment accuracy 0 L. E. REINSTEINet al.
Fig. 2. (a) Electronic portal image (EPI) of a RAND0 phantom with lung blocks. (b) Electronic portal image (EPI) of the same phantom after being translated in the longitudinal direction by 2 mm.
shifts in position (like in a cartoon animation), providing good detection sensitivity to block or patient movement. (Because of its very different gray scale dis-
I141
tribution, this technique will not work well if the reference image is a simulator film.) 3. Accurate repositioning of the EPID detector. The system described relies on precise repositioning of the video detection system rather than making use of software registration techniques (such as the identification of anatomic landmarks to generate coordinate transformations) which require too much time on the part of the user and create unacceptable delays in treatment. Instead, by mechanically repositiotnng the rigidly mounted detector (Fig. 1) the images are inherently registered, and displayed with respect to a single fixed beam coordinate system. Any apparent shift in block or patient position with respect to this system is, as a consequence, real. In order to test the detector positioning reproducibility the entire detector assembly was removed and then remounted on the gantry five times. A new EPI was taken after each move and the pixel coordinates of selected graticule pins recorded in Table 1. The results of this test reveal a repositioning precision of better than 1 mm.
Fig. 3. (a) The original reference image of the RAND0 phantom thorax with lung blocks. (b) Same reference image as in 3a but with key contours drawn in. (c) Key contours of reference image overlaid on electronic portal image (EPI) of phantom translated 2 mm longitudinally. Note that the 2 mm shift is easy to detect using this simulated “flip-book” technique compared to the conventional side-by-side review illustrated in Figure 2a. (d) Original reference contours overlaid on chest phantom electronic portal image (EPI) after 10 mm translation.
1142
I. J. Radiation Oncology 0 Biology0 Physics
Volume 22, Number 5, 1992
Fig. 4. (a) Electronic portal image (EPI) of RAND0 head phantom. (b) Key contours of RAND0 head drawn in for purposes of this paper only. (c) Contours of reference image overlaid on electronic portal image (EPI) of phantom after 2 mm translation in longitudinal direction. (d) Original reference contours overlaid on electronic portal image (EPI) of head phantom afier translation of 8 mm in longitudinal direction.
In actual practice, the initial physician-approved EPI is automatically loaded into the frame buffer during patient set-up. After a few seconds of treatment, the accelerator technologist views the current treatment EPI alternating with the approved image. Since the target-to-detector distance is fixed, it is simple to appreciate the magnitude of any shift, making it easy for the technologist to develop judgement as to whether the treatment should be allowed to continue.
RESULTS
With our current prototype system, reasonable quality images can be obtained in approximately 2 set at the rate of 30 frames/set. Although somewhat noisy (only 64 frames averaged), they are sufficiently useful to detect motion of relatively high contrast boundaries (e.g., bone, external surface, cerrobend blocks, etc.).
A major problem arises when trying to illustrate a dynamic video technique (such as the time-lapse display method) in a static illustration for journal article. There is no way to demonstrate the animation effect or the sensitivity of this technique to extremely small errors in field placement. The information can only be displayed as a “film-strip” sequence of images. (To improve reproduction quality for the sake of this article, the images shown were formed from an average of 5 12 frames. This acquisition time (17 set) is clearly impractical for real-time decisions.) To assist the reader’s understanding, key contours are drawn in on the original physician approved images. The contour lines are superimposed on a sampling of treatment images taken on subsequent days. These lines have been added only for the purpose of journal presentation and are not present (or needed) in the dynamic video display. The actual video display consists of the playback of images taken on consecutive treatment days as they appear
Geometric treatment accuracy 0 L. E. REINSTEINet al.
1143
Similar results were found when using the RAND0 head and shifting it in the longitudinal dimension. (See Fig. 4a, b, c, and d) Again, the small 2 mm shift is easily picked up.
Fig. 5. (a) Simulator image of patient being treated for carcinoma of the right lung. (b) Electronic portal image (EPI) of the approved treatment set-up. (c) Electronic portal image (EPI) of the same patient on a subsequent treatment day. (See Fig. 6f).
with respect to the fixed beam coordinate
system. It is hoped that the reader, in looking at the shifts between the drawn in reference contour and the current portal image in each figure, can image the “flip-book” effect described.
Phantom tests A RAND0 phantom was used to test whether small errors were detectable. An original reference image of the phantom thorax was made. It was then shifted in the longitudinal dimension first by 2 mm and then by 10 mm. Figures 2a, and b show the resulting EPID chest images. Note that if viewed in the conventional side by side manner these small changes are difficult to appreciate. The “flip-book” technique was applied by alternating the original with the 2 mm shift image and then the original with the 10 mm shift image, illustrated in Figures 3a, b, c, and d. Using this dynamic technique even the small 2 mm position change is easily detectable.
Patient images Because this particular EPID is a prototype under clinical evaluation, only a small number of our patients have undergone real-time imaging. Although not necessary for the dynamic “flip-book” technique, all images created for this paper were enhanced using the CLAHE technique (3, 4, 10). (With current hardware operation at 8 MHz, the CLAHE takes only about 30 sec. The speed can be improved severalfold using a faster 386 machine.) Figures 5a, b, and c show a simulator’ image side-byside with electronic portal images taken of a patient undergoing treatment for carcinoma of the right lung on the first and a subsequent day of treatment. Again, note that by viewing these images side-by-side in the conventional manner, it is difficult and time-consuming to evaluate field placement accuracy. The same patient images, when viewed using the “flip-book” technique,, are easier easy to assess. The apparent movement caused by the small shift in position is picked up immediately by the viewer. Figures 6 and 7 each show an original approved electronic portal image (EPI), the tracing of anatomic contours drawn in as a reference (for the purpose of this journal paper only), and the superposition of the reference contour with the image taken on each of four subsequent days. For the images seen in Figure 6b, the anatomical contours outlined (black and in white lines) are: the right clavicle, the tortuous trachea (deviating to the patient’s right), the apex of the right lung, the medal border of the left lung, the T%T9 interspace, and the bkfurcation of the trachea. The rather dramatic change in patient set-up position seen in Figure 6c was caught by tbe EPID on the next treatment day. (Ordinarily this unusual geometric error would not have been seen since portal films are taken on a weekly basis!) Note that the apex ofithe lung and the clavicle are completely out of the treatment field and that the carina has shifted approximately three vertebral bodies in the cephalad direction. (Overall shift is approximately 6 cm.) Observing the clavicle in image 6d indicates the possibility that the patient’s right arm is not in the original set-up position. The results show a small shift (1 cm) to the patient’s right (see the medial border of the left lung). Figure 6e reveals a shift in patient set-up about 1 cm in the caudad direction as seen by the shadow of the lung apex and approximately 4 cm to the right as revealed by the tracheal contour. In the final image of this series (Fig. 6f) the change in patient position is toward the left by 4 cm and the caudad direction by about 1 cm. The next example is of a Hodgkin’s disease patient. Only the central portion of the mantle field is seen in Figure 7 (the field of view on the prototype system is limited to approximately 25 cm X 30 cm). The outlines of
I. J. Radiation Oncology 0 Biology 0 Physics
Volume 22, Number 5, 1992
Fig. 6. (a) The electronic portal image (EPI) of a patient undergoing treatment for carcinoma of the right lung, after the set-up was approved. (b) Same electronic portal image (EPI) of approved set-up with contour outlines drawn in as a reference for purposes of this paper only. (c) Electronic portal image (EPI) of the second day of treatment with the reference lines superimposed. A large shift in patient position (approximately 6 cm) in the cephalad direction was picked up by this image. (d) Electronic portal image (EPI) of a subsequent treatment day showing a 1 cm shift in position to the patient’s right with respect to the reference image. (e) EPI image of 1 cm in the caudad direction with respect to reference image is seen. subsequent treatment. Shift of approximately (f) Electronic portal image (EPI) of patient on subsequent treatment day showing changes with respect to the reference image in both transverse and longitudinal dimensions from 4 cm to 1 cm. (Note that this is the same EPI as in Fig. 5c.)
Geometric treatment accuracy 0 L. E. REINSTEINet al.
Fig. 7. (a) Electronic portal image (EPI) of approved treatment set-up for Hodgkin’s disease. Only central portion of mantle field is seen in this image. (b) Same electronic portal image (EPI) with reference contours outlined for purposes of this paper only. (c) Superposition of reference lines with electronic portal image (EPI) on subsequent day showing patient position shift of 1 cm in cephalad direction with slight rotation to the left. (d) Electronic portal image (EPI) of subsequent treatment showing patient translation approximately 1 cm to the right. (e) Electronic portal image (EPI) of subsequent treatment day showing smaller patient translation (approximately .5 cm to the left). (f) Electronic portal image (EPI) of subsequent treatment similar to treatment shown in 7e.
1145
1146
I.
J. Radiation Oncology 0 Biology0 Physics
the clavicle, the apex of the lungs, the trachea, and the T6-T7 interspace are seen in Figure 7b. The image on the second day of treatment image (Fig. 7c) revealed a shift in patient position of approximately 1 cm in the cephalad direction (note the shifting of the lung apex and clavicles). There was also a movement to the patient’s left of approximately 0.75 cm. The next image (Fig. 7d) shows the patient shifting in the opposite (right) direction as seen by the outline of the trachea. Changes of similar magnitude are revealed in Figures 7e and 7f. Careful analysis of anatomic landmarks using a sequence of 10 electronic portal images result in a root mean square deviation in patient position of 8 mm and 4 mm in the longitudinal and transverse dimensions, respectively. DISCUSSION
AND CONCLUSION
Although observer studies must still be carried out to quantify the limits of its effectiveness, based upon the results of phantom imaging and clinical experience during
Volume 22, Number 5, 1992
the past year, the “time-lapse” or “flip-book” display technique appears to be is an effective means for detecting shifts in patient or block position as small as 2 mm. The technique is a simple extension of currently available EPI methods. While requiring almost no additional software, it does, however, necessitate detector repositioning capability that is fast and precise to within 1 mm. Sophisticated software-based image comparison techniques currently exist and may prove invaluable for off-line retrospective analyses of an entire treatment course. At the present time, however, they require excessive user interaction (identification and input of anatomic landmarks) and, therefore, impede treatment flow beyond what is acceptable in a busy clinic. Although the technique described in this paper does not make full use of computer capabilities (e.g., pattern recognition, spatial transformation, etc.), it does provide the accelerator technologist with a useful tool to make a quick determination of the treatment set-up accuracy, and thus to improve the overall quality of care.
REFERENCES I. Byhardt, R. W.; Cox, J. D.; Homberg,
2. 3. 4.
5.
6.
A.; Lierman, G. Weekly localization films and detection of field placement errors. Int. J. Radiat. Oncol. Biol. Phys. 4: 88 1; 1978. Lam, K. S.; Partowmah, M.; Loo, D. J.; Wharam, M. D. On line measurement of field placement errors in extended beam radiotherapy. Brit. J. Radiol. 60: 361-367; 1987. Leszczynski, K. W.; Shalev, S. S. Digital contract enhancement for on-line portal imaging. Med. Biol. Eng. Comput. 27: 507-512; 1989. Leszczynski, K. W.; Shalev, S. S. A robust algorithm for contrast enhancement by local histogram modification. Image Visual Comp. 7(3): 205-210; 1989. Marks, J. E.; Haus, A. G.; Sutton, H. G.; Griem, M. L. The value of frequent treatment verification films in reducing localization error in the irradiation of complex fields. Cancer 37: 2755; 1976. Marks, J. E.; Bedwinek, J. M.; Lee, F.; Purdy, J. A.; Perez,
C. A. Dose response analysis in nasopharyngeal carcinoma. Cancer 50: 1042-1050; 1982. Rabinowitz, R.; Bloomberg, J.; Goitein, M.; McCarthy, K.; Leong, J. Accuracy of radiation field alignment in clinical practice. Int. J. Radiat. Oncol. Biol. Phys. 11: 1857-1867; 1985. Reinstein, L. E.; Task Group 28. Radiotherapy imaging quality: report of AAPM Task Group No. 28. AAPM Report No. 24. AIP; 1987: 5. 9. Shalev, S.; Lee, T.; Leszczynski, K.; Cosby, S.; Chu, T.; Reinstein, L. E.; Meek, A. G. Video techniques for on-line portal imaging. Comput. Med. Imag. Graph. 13(3): 2 17226; 1989. 10. Sherouse, G. W.; Rosenman, J.; McMurry, H. L.; Pizer, S. M.; Chaney, E. Automatic digital contrast enhancement of radiotherapy films. Int. J. Radiat. Oncol. Biol. Phys. 13: 801; 1987.
0360.3016/92 $5.00 + .OO Copyright 0 1992 Pergamon Press Ltd.
0 Technical Innovations and Notes ASSESSMENT OF GEOMETRIC TREATMENT ACCURACY USING TIME-LAPSE DISPLAY OF ELECTRONIC PORTAL IMAGES LAWRENCE E. REINSTEIN, PH.D.,
SUJATHA PAI, M.S. AND ALLEN G. MEEK, M.D.
Departmentof Radiation Oncology, University Hospital at Stony Brook, Stony Brook, NY 1 I794 During the past two years, several electronic portal imaging systems have been introduced to the market by therapy accelerator manufacturers and other vendors. While these systems differ substantially in their detection technology, they are all capable of displaying portal images on a video screen in near real-time, and of creating multiple static (or “movie”) images during each treatment. Major questions confront the users of such systems as to the best utilization of this wealth of information, and to its value in comparison to traditional weekly portal film methods. Using an “in-house” video based system, a new technique was established to aid in the assessment of on-lineiimages so that immediate “go/no-go” decisions can be made by the therapy technologist. A video “movie-loop” is displayed which consists of the static image of the initial (approved) set-up, and the current treatment image. Multiple~images of successive treatments can also be viewed in this “time-lapse” display mode to provide a quick visual means for review of an entire course of therapy. The on-line imaging system hardware is composed of a combination copperplate/fluorescent-screen detector, a front surface mirror angled at 45” to remove the camera from the direct radiation beam, and a high sensitivity SIT video camera. This assembly is attached to a rigid base and mounted directly to the isocentric gantry. The geometry is fixed to within +1 mm and assures the precise day-to-day reproducibility which is necessary for the success of the time-lapse display technique. Experience with this technique shows it to enhance the user’s ability to notice small changes in patient’s position with respect to the radiation field. Radiation treatment sites reviewed using this procedure were Hodgkin’s (mantle), Lung, Brain and extremities. Shifts in patient position on the order of several millimeters were readily detectable, as will be demonstrated in thi$ paper. Somewhat surprisingly, grosser movements (>I cm) were also noted despite overall technical excellence as assessed by weekly portal filming. The eye senses day-today movement with greater ease when the fields are seen in timelapse display than when compared as discrete portal images. Ultimately, persistent movement appreciated on the time-lapse display can suggest the need for a change in patient set-up or immobilization technique. Portal, Imaging, EPID, On-line imaging, Quality assurance.
INTRODUCTION
localization errors (on the order of 1 cm) occur relatively frequently ( lo%-36% depending upon treatment site). These studies imply a need for large treatment margins which increase the target volume, thus running counter to the aforementioned strategy. Somewhat encouraging, however, are the results of Marks et al. (5), which show significant reduction in the rate of localization errors as the frequency of portal/localization films is increased (a drop from 36% of the cases to 15% of the cases in errors greater than 1 cm by increasing the number of portal films during a course of treatment from 9 to 24). These results suggest the need for increasing the frequency of portal/ localization imaging to, perhaps, a daily basis. Although daily portal imaging may be desirable from the point of view of improved treatment accuracy, such a policy is by no means accepted practice. A study (8) of 158 institutions indicates that even weekly filming has not been universally accepted, with fewer than half of the institutions reporting that weekly films are taken on the
One of the goals of modem radiation therapy is to improve the patient’s local control probability by increasing the therapeutic ratio in order to deliver a higher tumor dose without producing unacceptable complications. New diagnostic modalities have made more precise tumor localization possible and, when combined with “state of the art” 3-D treatment planning, let the radiotherapist design a treatment volume that conforms closely to the actual tumor volume. By using custom shaped blocks, multileaf collimation, and dynamic therapy techniques, the volume of normal tissue being irradiated can be significantly reduced. Although tumor localization and dose calculation capabilities have advanced, the ability to deliver treatments with the same degree of precision on a daily basis over an entire course of therapy has not. Several historic studies have demonstrated (1, 2, 5-7) that clinically significant
Presentedat the 32nd Meeting of ASTRO, Miami, FL, 1519 October 1990.
Reprint requests to: L. E. Reinstein, Accepted 1139
for publication
6 September
Ph.D. 199 1.
I. J. Radiation Oncology 0 Biology 0 Physics
Volume 22, Number 5, 1992 up accuracy, giving the technologist the ability to make a go/no-go decision.
METHODS
Fig. 1. The prototype accelerator gantry.
EPID detector
assembly
mounted
to the
majority of their patients. Reluctance to take frequent portal films is based upon the inconvenient and timeconsuming nature of the process; they impede patient flow and, therefore, increase the cost of treatment. Thus, a major motivation for the development of realtime on-line imaging systems (also known as electronic portal imaging devices [EPID]) has been to make daily portal imaging both expedient and practical. Various commercial and prototype EPID’s are currently available which, ideally, would allow the accelerator technologist to take daily portal images with very little extra time or added inconvenience. But, given this new capability, it now becomes important to ask how to make best use of the suddenly overwhelming number of portal images without over-stressing the system? What is the most effective way to evaluate these daily images to improve the quality of treatment? We have introduced a new technique, analogous to “time-lapse” photography, which addresses these questions. By rapidly displaying a sequence of images taken on consecutive treatment days, a “movie-loop” is created which dramatically brings out any change in patient or block position. When reduced to a two-image sequence (the first being the “approved” and the second the current treatment image), the resulting “flip-book” like animation is a sensitive indication (approximately +2 mm) of set-
Table 1. Results of the test for accuracy
AND MATERIALS
Electronic portal imaging The on-line electronic portal imaging system is videobased, consisting of a 2 mm thick copper plate backed with a GdzOzS:Tb Phosphor (250 mg/cm’), a 45” front surfaced mirror used to remove the video camera from the direct radiation beam, and a silicon intensified target (SIT) video camera with an Fl.4 lens (9). (The field of view in the plane of the isocenter is approximately 2.5 cm X 30 cm.) The video signal is digitized using an image processing system* interfaced to a computer.+ This particular image processor allows software control of the gain and offset, important for optimization of the image acquisition. It also contains a frame buffer board which consists of two 5 12 X 5 12 X 8 bit image planes. The imaging software package (OPIUM)* includes a variety of routines for image acquisition, frame averaging, non-uniformity corrections, windowing, and contrast enhancement ( 10). SpeciJic requirements for time-lapse display For the “time lapse” display technique to work effectively, the following system requirements must be met: For the “flip-book” technique (alternation of the “approved” and current image) the frame buffer need only be capable of storing two images for immediate sequential display. In the system described, however, as many as sixteen 256 X 240 X 8 bit images can be displayed in rapid sequence in a “movie-loop” playback. (Although the resolution is slightly reduced its effect is virtually unnoticeable in the movie-loop mode.) This enables the user to review the day-to-day shifts for virtually an entire treatment course. &. Acquired images must be of similar radiographic quality. For this technique to be effective, successive images must have a similar spread of gray scale values. As a result any image-to-image changes are only caused by
of electronic
portal image (EPI) detector
positioning
Pixel position
Image no. 1
Image no. 2
Image no. 3
Image no. 4
Image no. 5
Origin (5 cm, 0) (-5 cm, 0) (0, 5 cm) (0, -5 cm)
143, 132 185, 129 102, 135 141,79 145, 186
143, 132 185, 129 101, 135 141,79 145, 185
143, 132 185, 129 101, 136 141,79 145, 186
143, 132 185, 129 101, 135 141,79 145, 185
143, 132 185, 129 101, 135 141,79 145, 185
* Imaging Technology Series 15 1. +IBM AT. *On-line
portal
imaging,
University
of Manitoba.
The de-
velopment of the portal imaging system was in collaboration with S. Shalev, Manitoba Cancer Treatment and Research Foundation.
Geometric treatment accuracy 0 L. E. REINSTEINet al.
Fig. 2. (a) Electronic portal image (EPI) of a RAND0 phantom with lung blocks. (b) Electronic portal image (EPI) of the same phantom after being translated in the longitudinal direction by 2 mm.
shifts in position (like in a cartoon animation), providing good detection sensitivity to block or patient movement. (Because of its very different gray scale dis-
I141
tribution, this technique will not work well if the reference image is a simulator film.) 3. Accurate repositioning of the EPID detector. The system described relies on precise repositioning of the video detection system rather than making use of software registration techniques (such as the identification of anatomic landmarks to generate coordinate transformations) which require too much time on the part of the user and create unacceptable delays in treatment. Instead, by mechanically repositiotnng the rigidly mounted detector (Fig. 1) the images are inherently registered, and displayed with respect to a single fixed beam coordinate system. Any apparent shift in block or patient position with respect to this system is, as a consequence, real. In order to test the detector positioning reproducibility the entire detector assembly was removed and then remounted on the gantry five times. A new EPI was taken after each move and the pixel coordinates of selected graticule pins recorded in Table 1. The results of this test reveal a repositioning precision of better than 1 mm.
Fig. 3. (a) The original reference image of the RAND0 phantom thorax with lung blocks. (b) Same reference image as in 3a but with key contours drawn in. (c) Key contours of reference image overlaid on electronic portal image (EPI) of phantom translated 2 mm longitudinally. Note that the 2 mm shift is easy to detect using this simulated “flip-book” technique compared to the conventional side-by-side review illustrated in Figure 2a. (d) Original reference contours overlaid on chest phantom electronic portal image (EPI) after 10 mm translation.
1142
I. J. Radiation Oncology 0 Biology0 Physics
Volume 22, Number 5, 1992
Fig. 4. (a) Electronic portal image (EPI) of RAND0 head phantom. (b) Key contours of RAND0 head drawn in for purposes of this paper only. (c) Contours of reference image overlaid on electronic portal image (EPI) of phantom after 2 mm translation in longitudinal direction. (d) Original reference contours overlaid on electronic portal image (EPI) of head phantom afier translation of 8 mm in longitudinal direction.
In actual practice, the initial physician-approved EPI is automatically loaded into the frame buffer during patient set-up. After a few seconds of treatment, the accelerator technologist views the current treatment EPI alternating with the approved image. Since the target-to-detector distance is fixed, it is simple to appreciate the magnitude of any shift, making it easy for the technologist to develop judgement as to whether the treatment should be allowed to continue.
RESULTS
With our current prototype system, reasonable quality images can be obtained in approximately 2 set at the rate of 30 frames/set. Although somewhat noisy (only 64 frames averaged), they are sufficiently useful to detect motion of relatively high contrast boundaries (e.g., bone, external surface, cerrobend blocks, etc.).
A major problem arises when trying to illustrate a dynamic video technique (such as the time-lapse display method) in a static illustration for journal article. There is no way to demonstrate the animation effect or the sensitivity of this technique to extremely small errors in field placement. The information can only be displayed as a “film-strip” sequence of images. (To improve reproduction quality for the sake of this article, the images shown were formed from an average of 5 12 frames. This acquisition time (17 set) is clearly impractical for real-time decisions.) To assist the reader’s understanding, key contours are drawn in on the original physician approved images. The contour lines are superimposed on a sampling of treatment images taken on subsequent days. These lines have been added only for the purpose of journal presentation and are not present (or needed) in the dynamic video display. The actual video display consists of the playback of images taken on consecutive treatment days as they appear
Geometric treatment accuracy 0 L. E. REINSTEINet al.
1143
Similar results were found when using the RAND0 head and shifting it in the longitudinal dimension. (See Fig. 4a, b, c, and d) Again, the small 2 mm shift is easily picked up.
Fig. 5. (a) Simulator image of patient being treated for carcinoma of the right lung. (b) Electronic portal image (EPI) of the approved treatment set-up. (c) Electronic portal image (EPI) of the same patient on a subsequent treatment day. (See Fig. 6f).
with respect to the fixed beam coordinate
system. It is hoped that the reader, in looking at the shifts between the drawn in reference contour and the current portal image in each figure, can image the “flip-book” effect described.
Phantom tests A RAND0 phantom was used to test whether small errors were detectable. An original reference image of the phantom thorax was made. It was then shifted in the longitudinal dimension first by 2 mm and then by 10 mm. Figures 2a, and b show the resulting EPID chest images. Note that if viewed in the conventional side by side manner these small changes are difficult to appreciate. The “flip-book” technique was applied by alternating the original with the 2 mm shift image and then the original with the 10 mm shift image, illustrated in Figures 3a, b, c, and d. Using this dynamic technique even the small 2 mm position change is easily detectable.
Patient images Because this particular EPID is a prototype under clinical evaluation, only a small number of our patients have undergone real-time imaging. Although not necessary for the dynamic “flip-book” technique, all images created for this paper were enhanced using the CLAHE technique (3, 4, 10). (With current hardware operation at 8 MHz, the CLAHE takes only about 30 sec. The speed can be improved severalfold using a faster 386 machine.) Figures 5a, b, and c show a simulator’ image side-byside with electronic portal images taken of a patient undergoing treatment for carcinoma of the right lung on the first and a subsequent day of treatment. Again, note that by viewing these images side-by-side in the conventional manner, it is difficult and time-consuming to evaluate field placement accuracy. The same patient images, when viewed using the “flip-book” technique,, are easier easy to assess. The apparent movement caused by the small shift in position is picked up immediately by the viewer. Figures 6 and 7 each show an original approved electronic portal image (EPI), the tracing of anatomic contours drawn in as a reference (for the purpose of this journal paper only), and the superposition of the reference contour with the image taken on each of four subsequent days. For the images seen in Figure 6b, the anatomical contours outlined (black and in white lines) are: the right clavicle, the tortuous trachea (deviating to the patient’s right), the apex of the right lung, the medal border of the left lung, the T%T9 interspace, and the bkfurcation of the trachea. The rather dramatic change in patient set-up position seen in Figure 6c was caught by tbe EPID on the next treatment day. (Ordinarily this unusual geometric error would not have been seen since portal films are taken on a weekly basis!) Note that the apex ofithe lung and the clavicle are completely out of the treatment field and that the carina has shifted approximately three vertebral bodies in the cephalad direction. (Overall shift is approximately 6 cm.) Observing the clavicle in image 6d indicates the possibility that the patient’s right arm is not in the original set-up position. The results show a small shift (1 cm) to the patient’s right (see the medial border of the left lung). Figure 6e reveals a shift in patient set-up about 1 cm in the caudad direction as seen by the shadow of the lung apex and approximately 4 cm to the right as revealed by the tracheal contour. In the final image of this series (Fig. 6f) the change in patient position is toward the left by 4 cm and the caudad direction by about 1 cm. The next example is of a Hodgkin’s disease patient. Only the central portion of the mantle field is seen in Figure 7 (the field of view on the prototype system is limited to approximately 25 cm X 30 cm). The outlines of
I. J. Radiation Oncology 0 Biology 0 Physics
Volume 22, Number 5, 1992
Fig. 6. (a) The electronic portal image (EPI) of a patient undergoing treatment for carcinoma of the right lung, after the set-up was approved. (b) Same electronic portal image (EPI) of approved set-up with contour outlines drawn in as a reference for purposes of this paper only. (c) Electronic portal image (EPI) of the second day of treatment with the reference lines superimposed. A large shift in patient position (approximately 6 cm) in the cephalad direction was picked up by this image. (d) Electronic portal image (EPI) of a subsequent treatment day showing a 1 cm shift in position to the patient’s right with respect to the reference image. (e) EPI image of 1 cm in the caudad direction with respect to reference image is seen. subsequent treatment. Shift of approximately (f) Electronic portal image (EPI) of patient on subsequent treatment day showing changes with respect to the reference image in both transverse and longitudinal dimensions from 4 cm to 1 cm. (Note that this is the same EPI as in Fig. 5c.)
Geometric treatment accuracy 0 L. E. REINSTEINet al.
Fig. 7. (a) Electronic portal image (EPI) of approved treatment set-up for Hodgkin’s disease. Only central portion of mantle field is seen in this image. (b) Same electronic portal image (EPI) with reference contours outlined for purposes of this paper only. (c) Superposition of reference lines with electronic portal image (EPI) on subsequent day showing patient position shift of 1 cm in cephalad direction with slight rotation to the left. (d) Electronic portal image (EPI) of subsequent treatment showing patient translation approximately 1 cm to the right. (e) Electronic portal image (EPI) of subsequent treatment day showing smaller patient translation (approximately .5 cm to the left). (f) Electronic portal image (EPI) of subsequent treatment similar to treatment shown in 7e.
1145
1146
I.
J. Radiation Oncology 0 Biology0 Physics
the clavicle, the apex of the lungs, the trachea, and the T6-T7 interspace are seen in Figure 7b. The image on the second day of treatment image (Fig. 7c) revealed a shift in patient position of approximately 1 cm in the cephalad direction (note the shifting of the lung apex and clavicles). There was also a movement to the patient’s left of approximately 0.75 cm. The next image (Fig. 7d) shows the patient shifting in the opposite (right) direction as seen by the outline of the trachea. Changes of similar magnitude are revealed in Figures 7e and 7f. Careful analysis of anatomic landmarks using a sequence of 10 electronic portal images result in a root mean square deviation in patient position of 8 mm and 4 mm in the longitudinal and transverse dimensions, respectively. DISCUSSION
AND CONCLUSION
Although observer studies must still be carried out to quantify the limits of its effectiveness, based upon the results of phantom imaging and clinical experience during
Volume 22, Number 5, 1992
the past year, the “time-lapse” or “flip-book” display technique appears to be is an effective means for detecting shifts in patient or block position as small as 2 mm. The technique is a simple extension of currently available EPI methods. While requiring almost no additional software, it does, however, necessitate detector repositioning capability that is fast and precise to within 1 mm. Sophisticated software-based image comparison techniques currently exist and may prove invaluable for off-line retrospective analyses of an entire treatment course. At the present time, however, they require excessive user interaction (identification and input of anatomic landmarks) and, therefore, impede treatment flow beyond what is acceptable in a busy clinic. Although the technique described in this paper does not make full use of computer capabilities (e.g., pattern recognition, spatial transformation, etc.), it does provide the accelerator technologist with a useful tool to make a quick determination of the treatment set-up accuracy, and thus to improve the overall quality of care.
REFERENCES I. Byhardt, R. W.; Cox, J. D.; Homberg,
2. 3. 4.
5.
6.
A.; Lierman, G. Weekly localization films and detection of field placement errors. Int. J. Radiat. Oncol. Biol. Phys. 4: 88 1; 1978. Lam, K. S.; Partowmah, M.; Loo, D. J.; Wharam, M. D. On line measurement of field placement errors in extended beam radiotherapy. Brit. J. Radiol. 60: 361-367; 1987. Leszczynski, K. W.; Shalev, S. S. Digital contract enhancement for on-line portal imaging. Med. Biol. Eng. Comput. 27: 507-512; 1989. Leszczynski, K. W.; Shalev, S. S. A robust algorithm for contrast enhancement by local histogram modification. Image Visual Comp. 7(3): 205-210; 1989. Marks, J. E.; Haus, A. G.; Sutton, H. G.; Griem, M. L. The value of frequent treatment verification films in reducing localization error in the irradiation of complex fields. Cancer 37: 2755; 1976. Marks, J. E.; Bedwinek, J. M.; Lee, F.; Purdy, J. A.; Perez,
C. A. Dose response analysis in nasopharyngeal carcinoma. Cancer 50: 1042-1050; 1982. Rabinowitz, R.; Bloomberg, J.; Goitein, M.; McCarthy, K.; Leong, J. Accuracy of radiation field alignment in clinical practice. Int. J. Radiat. Oncol. Biol. Phys. 11: 1857-1867; 1985. Reinstein, L. E.; Task Group 28. Radiotherapy imaging quality: report of AAPM Task Group No. 28. AAPM Report No. 24. AIP; 1987: 5. 9. Shalev, S.; Lee, T.; Leszczynski, K.; Cosby, S.; Chu, T.; Reinstein, L. E.; Meek, A. G. Video techniques for on-line portal imaging. Comput. Med. Imag. Graph. 13(3): 2 17226; 1989. 10. Sherouse, G. W.; Rosenman, J.; McMurry, H. L.; Pizer, S. M.; Chaney, E. Automatic digital contrast enhancement of radiotherapy films. Int. J. Radiat. Oncol. Biol. Phys. 13: 801; 1987.
