THE USE OF ONLINE IMAGE VERIFICATION TO ESTIMATE THE VARIATION IN RADIATION-THERAPY DOSE DELIVERY

In, J Rudrirtion On<olo~j, RIOI Phi,.! Vol 27. pp. 707-716 Pnnted I” the U.S.A. All nghts reserved.

0360.3016/93 $6.00 + .OO Copyright 0 I993 Pergamon Press Ltd.

0 Technical Innovations and Notes

THE USE OF ON-LINE IMAGE VERIFICATION TO ESTIMATE THE VARIATION IN RADIATION THERAPY DOSE DELIVERY

JEFF M. MICHALSKI, M.D., JOHN W. WONG, PH.D., R. L. GERBER, M.S., DI YAN, PH.D., ABEL CHENG, M.S., MARY V. GRAHAM, M.D., M. A. RENNA, B.S., R.T.T.,

P. J. SAWYER, R.T.T. AND CARLOS A. PEREZ, M.D.

Radiation Oncology Center, Mallinckrodt Institute of Radiology, Washington University School of Medicine, St. Louis, MO 63 I 10

Purpose: On-line radiotherapy imaging systems provide data that allow us to study the geometric nature of treatment sn. It is more clinically relevant to examine the resultant dosimetric variation. In this work, daily beam position as recorded by the on-line images is used to recalculate the treatment plan to show the effect geometric variation has on dose. Methods and Materials: Daily 6 MV or 18 MV x-ray portal images were acquired using a fiberoptic on-line imaging system for 12 patients with cancers in the head and neck, thoracic, and pelvic regions. Each daily on-line portal image was aligned with the prescription simulation image using a template of anatomical structures defined on the latter. The outline of the actual block position was then superimposed on the prescription image. Daily block positions were cumulated to give a summary image represented by the block overlap isofrequency distribution. The summary data were used to analyze the amount of geometric variation relative to the prescription boundary on a histogram distribution plot. Treatment plans were recalculated by considering each aligned portal image as an individual beam. Results: On-Line Image Verification (OLIV) data can differentiate between systematic and random errors in a course of daily radiation therapy. The data emphasize that the type and magnitude of patient set-up errors are unique for individual patients and different clinical situations. Head and neck sites had the least random variation (average O-100% block overlap isofrequency distribution width = 7 mm) compared to thoracic (average O-100% block overlap isofrequency distribution width = 12 mm) or pelvic sites (average O-100% block overlap isofrequency distribution width = 14 mm). When treatment delivery is analyzed case by case, systematic as well as random errors are represented. When the data are pooled by anatomical site, individuality of variations is lost and variation appears random. Recalculated plans demonstrated dosimetric deviations from the original plans. The differences between the two dosimetric distributions were emphasized using a technique of plan subtraction. This allowed quick identification of relative “hot and cold spots” in the recalculated plans. The magnitude and clinical significance of dosimetric variation was unique for each patient. Conclusions: OLIV data are useful to study geometric uncertainties because of the unique nature for individual patients. Dose recalculation is helpful to illustrate the dosimetric consequences of set-up errors.

On-line image verification, Treatment errors, Treatment plan recalculation.

INTRODUCTION

Accurate delivery of a specific radiation dose to a target volume is critical to the success of radiation therapy. Many clinical and radiobiologic data have demonstrated steep dose response curves for tumor control (13). This implies that an accurate estimation of the radiation dose delivered in everyday clinical practice is necessary. It is felt by some authors that the lack of a sigmoidal dose response in some clinical studies may be a reflection of inaccurate dosimetric information ( I, 6, 16).

Sources of dosimetric uncertainty in treatment planning have recently been reviewed (14). They may be related to physical processes such as inaccurate estimation of in- homogeneity corrections, CT to electron density conver- sions, dose computation algorithms, and treatment machine related uncertainties. They may also be patient specific, such as inaccurate definition of target volume, positional uncertainty, and field setup uncertainty. Of these variables, it is felt that beam misalignment may be the most critical to alter the likelihood of tumor control (I). Attempts have been made to introduce the impact of

Presented at the 33rd Meeting of ASTRO, Washington, D.C., Institute of Radiology, Washington University School of Med- 4 November I99 I. This work was supported in part by National icine, Radiation Oncology Center, Box 8224, 4939 Audubon Cancer Institute contract NOI-CM-97564. Ave., Ste. 5500, St. Louis, MO 63110.

Reprint requests to: Jeff M. Michalski, M.D., Mallinckrodt Accepted for publication 23 April 1993.

707

708 1. J. Radiation Oncology 0 Biology 0 Physics Volume 27, Number 3, 1993

dosimetric uncertainty in treatment planning (2, 8, 15). Dose uncertainty is estimated based on a theoretical average of patient positioning errors. The calculation provides valuable information to the radiation oncologist as he/she considers a treatment plan for a particular patient. However, these uncertainty models may not be applicable under all circumstances.

Several studies have demonstrated that frequent, even daily, port films can provide an accurate estimate of patient set-up variation and may eventually increase the accuracy of dose delivery (2, 9, 10). The recent advent of On-Line Image Verification (OLIV) has provided us with a wealth of information regarding patient treatment variability (4, 5). We propose that OLIV portal images obtained throughout a course of therapy can be used to give a more accurate estimation of dose delivered to the patient. This study describes our method used to recalculate and analyze radiation therapy treatment plans with the information obtained from the daily portal images.

METHODS AND MATERIALS

Patient population Twelve patients with various malignancies were treated

on a dual energy medical accelerator* equipped with an on-line imaging system. Treatment sites were evenly distributed to include malignancies of the head and neck region, thorax, and pelvis. All patients were treated with curative intent. Their ages range from 45 to 75. Patients treated to the head and neck region were immobilized with a customized thermoplastic head mask whereas patients treated to the thorax and abdominopelvic regions had no special immobilization devices. An opposed lateral treatment technique was employed for head and neck sites except in one patient where a pair of oblique wedge fields was used for a maxillary sinus tumor. All lung cancer patients were treated with opposed AP/PA treatments initially, followed by parallel opposed oblique boosts. Two patients with gynecologic cancers received AP/PA pelvic treatment and two other patients were treated with a four field box pelvic technique. All fields were imaged and an average of 59 portal images were obtained per patient, with a range of 38 to 79.

Simulation films were digitized using a CCD camera and a light box. Anatomical features and the prescription boundary on the films were outlined with a colored wax pencil to facilitate identification. During each treatment session, a single 4 set (approximately 25 cGy) exposure portal image of each portal was stored. No attempt was made to adjust the patient position or beam alignment based on the information available from the acquired images during this study. Adjustments were only made with information gathered from the routine weekly port film that continues in our clinic. Images were enhanced for display using simple gray scale windowing. We attempted to obtain a second large open field image for the small boost field treatments because of the lack of anatomical information in the small field. The open field image was obtained with a four monitor unit exposure (approximately 5 cGy). The treatment portal image and the open field image were superimposed to provide the silhouette of the actual beam aperture along with the additional anatomical structure information from the large open field. Since the open field technique required the administration of a small dose (less than 50 cGy total) to otherwise not irradiated areas, informed consent was obtained and this information was relayed to the prescribing physician who took this into account during the initial treatment planning.

After the digital prescription and portal images were optimally ‘windowed’ for display on the OLIV system, they were transferred to a high performance image work- station.* These images were analyzed using the Cumula- tive Verification Image Analysis (CVIA) method.

Cumulative verification image analwis The CVIA method has been previously described (5).

This method has been further developed as part of the NC1 Radiation Therapy Treatment Planning Tools contract (11).

The magnification difference between the portal image and the prescription image was determined using a fiducial grid. For each portal image the block outline was manually identified using a computer mouse. A block template defined from the first treatment was used to facilitate the process for subsequent daily images.

Image acquisition Daily portal images were obtained using a commercially

available fiberoptic imaging system.+ Operation of the system has been described in previous publications (5, 17) and a companion paper (18). In the present study the fiberoptic device was mounted on the beam stop of the accelerator. Digital on-line portal images were acquired and stored on optical disks in a personal computer.

Structural contours and intersections were identified and outlined on the prescription image (Fig. la). These contours were stored as a bit map template and their re- lationship to each other and the block edge was fixed. The structural template was overlaid on a portal image and adjusted by translational and rotational shifts using a computer mouse to align the same structures on the portal image (Fig. 1 b). Once the structures were adequately aligned, the position of the block outline was extracted and superimposed back on the prescription image (Fig.

* Varian Radiation Division, Palo Alto, CA 94303. f Fiber Imaging Inc., St. Louis, MO 63 103.

t Silicon Graphics, Inc. Mountain View, CA 94309.

On-line image verification 0 J. M. MICHALSKI rr al. 709

a

Fig. 1. (a) Anatomical template for portal image alignment defined on prescription image. (b) Anatomical template overlaid and aligned upon portal image. (c) Resultant block position relative to prescription image based upon anatomical structure alignment.

b

lc). Each daily custom block position was cumulated in a summary bit map and superimposed on the prescription image to give a cumulative verification summary. In this study all images were aligned by one operator. The reli- ability and reproducibility of the structural alignment technique will be the subject of a separate publication.

From the cumulative verification summary, the Block Overlap Isofrequency Distributions (BOID’s) for the entire course of treatment were determined (Fig. 2). Each curve in the figure represents the frequency that the area was covered by the actual treatment portals. The innermost curve (100% isofrequency) represents the area that was

covered throughout the course of treatment. The outer- most curve represents the extreme limits of variation (0% isofrequency). In other words, nothing outside the 0% isofrequency distribution was irradiated by the primary beam whereas the area inside the 100% isofrequency was always irradiated. The curves moving outward from the center represent SO%, 50%, and 20% of coverage. For the dosimetric studies, an axis of calculation was defined on the prescription image using a computer mouse. The translational difference between the prescription edge and each portal boundary was measured. The translational differences between the daily portal boundaries relative to the

I. J. Radiation Oncology 0 Biology 0 Physics Volume 27. Number 3. 1993

Fig. 2. Block overlap isofrequency distribution (BOID) on a prescription image. Colored contours represent the relative frequency of daily coverage. 100% isofrequency red, 80% isofrequency yellow, 50% isofrequency green 20% isofrequency blue. 0% isofrequency lavender.

prescription edge were expressed as a histogram distribution to determine the magnitude and direction of the variation.

Dose calculation Treatment plans were calculated using a commercially

available treatment planning system.$ Patient external contours, gross tumor volumes with a 1 cm margin, lymph node regions, and critical structures were delineated using cross sectional imaging data from computed tomography. Block edges and central axis location for the original treatment plan were taken from the prescription image. For plan recalculation, each daily block position was considered as a unique beam. To simplify this 2-dimensional (2-D) calculation process, the blocked fields were assumed to be rectangular fields with their central axis located symmetrically between the beam edges. On the recalculated plans, portal movement was handled by assuming the central axis shifted and the beam widths as defined by the primary collimators varied on a daily basis which preserved the dosimetric data near the block edge. This approach allowed us to overcome a deficiency in our 2- D treatment planning system which employs a fanline transmission with no scatter correction to calculate dose under blocks. Changes in beam divergence were insignificant over the range of field sizes measured. Modifying wedges, when used, were assumed to move with central

axis of the beam. Compensating filters are treated as bolus in our treatment planning system and were assumed to move with the patient.

When recalculating a treatment plan, patient position was assumed to be rigid. All positional variation was approximated by translational and rotational changes in block position. Portal misalignment may be caused by other factors such as changes in patient position relative to the treatment couch (i.e., pitch or roll) or changes in relative anatomical structures (i.e., jaw or clavicle movement). In this study, the effects of these changes on alignment were approximated in terms of block variation. Furthermore, compensating filter misregistration or the possibility of a small change in the source to skin distant was not considered.

On rare occasions, a portal image was not available for dose recalculation. In these instances, either an average position was used (as defined by the 50% block overlap isofrequency distribution) or the original prescription field size and position was used for dose calculation.

Han analysis The original treatment plan was compared to the plan

recalculated from information derived from the daily portal images. Side by side comparison was helpful when the differences in dose were strikingly different, however subtle changes were not easily detected.

To emphasize the differences for easy detection, the original plan dose distribution was subtracted from that of the recalculated plan. This method allows a quick sur- vey of the areas that received a dose different from that which was prescribed. “Hot and cold spots” could be identified with respect to target volume and critical structures. An overlay of the “hot and cold spots” over the original plan helped illustrate the magnitude of error relative to the dose delivered to these areas.

RESULTS

Block overlap isqjqmmcy distribution (BOID) The width of the BOID’s was measured along the axis

of calculation defined on the transverse plane of the prescription image. In some instances the magnitude of variation was more pronounced in other directions, but for purposes of this paper, the BOID widths were only measured on the transverse plane.

The mean width of the 0 to 100% BOID’s was site dependent. Table 1 summarizes the magnitude of variation relative to site. Because the width of the 0 to 100% BOID’s can be influenced by the effects of a large variation of a single treatment, the 20 to 80% BOID’s width are more indicative of typical treatment variation. In general, alignment was better in the head and neck region than either the thorax or pelvis.

g Computerized Medical Systems, St. Louis, MO 63 128.

On-line image verification 0 J. M. MICHALSKI et (11. 711

Table 1. Block overlap isofrequency distribution (BOID) width (mm)*

o- 100% Average (range)

20-80% Average (range)

Head and neck 7 (2-14) 3 (l-7) Thorax 12 (3-22) 6 (3-14) Pelvis 14 (7-25) 6 (3-8)

* Measurements taken along the axis of calculation.

Block-edge sh[jis Figures 3.4, 5, illustrate three examples ofthe histogram

distribution of block variation relative to the prescription edge. Figure 3 shows a patient’s treatment that was accurately reproduced with minimal variation relative to the prescription. Figure 4 shows an example of random variation. The actual portal boundaries are distributed almost symmetrically around the prescription edge. A systematic error was considered to predominate when the average position of the block edge was greater than one standard deviation from the prescription boundary. A block cutting error is one example of this type of variation and is illustrated by Figure 5.

The unique nature of the variation for the individual patient is lost when the data are pooled for any given site. Figures 3, 4, and 5 show the unique treatment variations of individual patients. Figure 6 shows the combined treatment variation of all patients at the different sites. The pooled data show a near Gaussian distribution about the prescription edge. This emphasizes the importance of using an individuals data to estimate dosimetric uncertainty on a case by case basis.

16 15

I- 14 f 13

8 12 11 5 10 i= 4 t

z 3

z :

i

:,

AP PELVIS 2

RIGHT PORT EDGE

25 PORTS MEAN = -0.92 mm

8.d. = 2.1 mm

111111111111111111 llllllllllllllllllr

-20 -15 -10 -5 0 5 10 15 20 SHIFT (mm)

-20 -15 -10 -5 0 5 10 15 20 SHIFT (mm)

Fig. 3. Histogram distribution of a single block edge position Fig. 5. Histogram distribution of a single block edge position for a patient irradiated to the pelvis. The shaded 5 HVL repre- for a patient irradiated for a brain tumor. Systematic error presents the intended position of the block. dominates.

AP LUNG 2

RIGHT PORT EDGE

24 PORTS MEAN = 0.8

ad. = 3.0 mm

lllll,,ll-

-20 15 -10 -5 0 5 10 15 20 SHIFT (mm)

Fig. 4. Histogram distribution of a single block edge position for a patient irradiated to the lung. Random error predominates.

Dosimetric variation Figure 7 illustrates an example of an overlay plan in a

patient who received pelvic irradiation for carcinoma of the cervix. A systematic shift resulted in relative under- dosing of the lymph node volume by approximately 200 cGy, whereas the opposite side received an unintended dose of 1000 cGy adjacent to the prescribed area. In Figure 8, an overlay plan for a patient treated for lung cancer, demonstrates small fluctuations in the dose to the adjacent lymph node volume, whereas an area of adjacent normal lung received a dose slightly higher than anticipated. This patient’s portal positions varied in a predominantly random fashion. In each of these cases the magnitude of do-

23 PORTS MEAN = 8.1 mm

s.d. = 1.7 mm

RIGHT LATERAL BRAIN ANTERIOR PORT EDGE

8

i -1

712 1. J. Radiation Oncology 0 Biology 0 Physics Volume 27, Number 3, 1993

CUMULATIVE LATERAL H&N CUMULATIVE AP PELVIS CUMULATIVE AP THORAX SINGLE PORT EDGE SINGLE PORT EDGE SINGLE PORT EDGE

25 25 25

520 5 20 20 FOUR PITIENTS FOUR PATIENTS FOUR PATIENTS

0 ? 15 YLIN z 3.. mm 15 MC&N = 2.0 mm MEAN=‘.Olllfll O.D. = ..o mm SD. = 4.3 mm

15 a*. E 1.2 mm

9 lo 10 10

2 5 5 5

0 0 0 -20 -15 -10 -5 0 5 10 15 20 -20 -15 -10 -5 0 5 10 15 20 -20 -15 -10 -5 0 5 10 15 20

SHIFT (mm) SHIFT (mm) SHIFT (mm) (4 (W (C)

Fig. 6. Histogram distributions of actual block edge positions relative to the prescription. Each graph represents data pooled from the four patients in each anatomic subsite.

simetric variation is small and the consequences are of minimal clinical significance. Similar minor dosimetric variations were seen in most of the patient studies.

Figure 9 demonstrates the dosimetric significance of a systematic error in a patient treated for a brain tumor. A customized block was fabricated with a magnification factor larger than that which was intended. This treatment port was unequally weighted to that side because of the eccentric position of the tumor. The overlay plan illustrates areas of overdose, relative to what was intended, at the anterior and posterior portions of the plan. A clinically significant dose was delivered to the patients’ orbits as a

DISCUSSION

On-line image verification (OLIV) is a new technical advance in modern radiotherapy that allows us to measure the extent of daily set-up variations. Several series have demonstrated that minor set-up variations are common in radiotherapy (2, 5, 7, 9, 10, 12). It has been suggested that frequent port film verification may improve the accuracy of portal placement (9. 10). The OLIV system maintains a daily record of portal alignment that allows in-depth study of treatment set-up variation for the individual patient. This information also can be used to more accurately estimate the radiation dose delivered to

result of the block cutting error. the patient.

Dose Difference lcGyl 0 Es3 1200 1 B 800 2 r 200 3 tzzz -200 4 - -800 \

Fig. 7. An overlay of the dosimetric difference from the recalculated plan on the original dosimetric distribution. The effects of a slight systematic lateral shift are evident in this patient treated for cancer of the cervix. The primary clinical target volume has received the intended dose. The nodal target volume is slightly underdosed.

On-line image verification 0 J. M. MICHALW ef al. 713

Fig. 8. An overlay of the dosimetric difference from the recalculated plan on the original dosimetric distribution. The effects of a small but predominantly random variation are evident in this patient treated for carcinoma of the lung, Small clinically insignificant dosimetric alterations are seen in the nodal target volume. The dose to the primary tumor is uneffected. A small region of adjacent normal lung has received a dose higher than initially intended.

(se Difference kGv1 0 Ezl 1500 1 EEEEi 800 2 m 200 3= -200 4 -500

Fig. 9. An overlay of the dosimetric difference from the recalculated plan on the original dosimetric distribution. The effects of a block cutting error and systematic shift are evident in this patient treated for a brain tumor. The primary target volume is uneffected dosimetrically. The adjacent orbital contents demonstrate clinically significant deviations from the intended treatment plan.

714 I. J. Radiation Oncology 0 Biology 0 Physics Volume 27, Number 3. 1993

Table 2. Magnitude of variation from prescription (single port edge)

Mean distance

from Standard Predominant prescription deviation systematic

Patient no. (mm) (mm) error

Head and neck (total) 3.9 4.1 No I 5.5 2.7 Yes 2 4.2 1.7 Yes 3 2.7 2.2 Yes 4 I.9 2.1 No

Lung (total) 1.0 4.2 No 1 3.5 2.8 Yes 2 3.1 3.0 Yes 3 2.9 4.0 No 4 2.9 3.7 No

Pelvis (total) 2.0 4.3 No 1 4.5 4.5 Yes 2 1.0 2.1 No 3 3.2 2.9 Yes 4 2.0 3.4 No

Treatment set-up uncertainty The amount of treatment variation was shown to be

site dependent by some studies (2, lo), but this was not demonstrated by others ( 12). The magnitude of variation may also vary significantly from the simulation to the initial treatment ( 12). This suggests a systematic error may be inherent in the transfer of a patient from the simulator to the treatment machine.

Our data demonstrate that the magnitude and type of treatment set-up variability depends on a number of factors unique to a specific clinical situation. Some degree of random variation was present in all cases. BOID widths provide a measure of the magnitude of random variation. These values reflect the range that a port has moved during the course of treatment without reference to the position of the prescribed boundary. Treatment of head and neck sites, on the average, had a smaller magnitude of random error than sites in the thorax or pelvis. This may be related to better immobilization techniques, field size, or the availability of more rigid anatomical landmarks for set- up in this site. The mean and one standard deviation of the translational shift of a single block edge for one port with respect to the prescription is shown in Table 2 for individual patients and for the group as a whole. In some circumstances the shift from the prescribed port edge is of equal or greater magnitude than the standard deviation about the actual positions. This suggests that a systematic error predominates in these set-ups. Our data also emphasizes that the use of pooled data to estimate variation negates the individual nature of set-up errors.

Systematic errors were present in several of the patients. These were represented by a unidirectional shift of the average port edge placement away from the prescribed

boundary. In one example, an error was made in cutting the custom block due to recording an erroneous source to tray distance. The cause of other systematic errors is more difficult to explain. They could represent difficulty in transferring a set-up designed on the simulator to the treatment machine as has been previously described ( 12). The presence of systematic errors is particularly problem- atic because our data as well as others demonstrate they impact most significantly on the dosimetric distribution (1). Systematic errors should be avoidable or corrected with improved treatment verification. However, in some circumstances they may not be apparent unless serial portal images are examined cumulatively. Setup corrections based on periodic port films inherently assume a purely systematic error and this concept may be flawed.

Dosimetric estimation Radiotherapy planning typically presents the physician

with a nominal distribution of dose around the target volume. These plans assume rigid patient anatomy and pre- cise beam alignment during a course of radiation therapy. Several authors have proposed methods that integrate an estimate of uncertainty in radiation therapy plans. Goitein has suggested that a physician be presented with three treatment plans (3). One would be a nominal plan calculated without variation and the others would represent estimates ofsystematic minimum or maximum coverage. Leong has devised a method for incorporating random set-up fluctuations into computerized dose calculations (8). lsodose curves would depict the “most probable” lo- cations ofthe isodoses with corresponding margins of uncertainty. van de Geijn likewise has described a radiation therapy planning method that accounts for random variation in patient position and set-up (15). Each of these methods generalizes the degree and type of variation. This information can be valuable to the radiation oncologist as he prepares a course of therapy for a patient. However, the actual distribution may vary from that which was pre- dicted. The magnitude and type of set-up errors do not become apparent until therapy is underway. Information from the OLIV system can be used to recalculate a patient’s treatment plan to verify that the target volume received the intended dose and dose to critical structures remains within the range of tolerance. The individual nature of the magnitude and type of variation, as demonstrated in our study, underscores the shortcomings of using a population based model to predict the uncertainty in radiation therapy planning.

The recalculated plans provide a unique measure of the accuracy of our intended treatment plan. Plan subtraction is a method that allows easy identification of dosimetric differences. Our data, based on a small number of patients, suggest that on simple treatment plans with a generous margin surrounding the target volume, the recalculated plans typically demonstrated adequate cov-

On-line image verification 0 J. M. MICHALSKI et al. 715

erage of the disease. On the contrary, initial plans that had narrow margin around the target volume or complex plans employing several beams with oblique orientation, had recalculated plans that more likely demonstrated a narrowing or loss of the dosimetric margin. We anticipate studying this with a larger population base to confirm these findings.

The technique in this study offers only a first order approximation of dose actually delivered to the patients’ target volumes. Ideally, this method should be expanded to take into account the treatment variation related to patient organ motion during each treatment session. Just as the daily setup error measured in this study demonstrates changes from the expected dose delivery, the very ability of target position at each treatment session may alter the expected cumulative target radiation dose. Ex- tending this dose recalculation method to this type of variation would require faster electronic portal imaging devices with better “real time” imaging capabilities. Ad- ditionally, the dosimetric impact of non-rigid body variation and out of plane rotations cannot be adequately estimated with the techniques described here.

Future clinical utility The current technique of dose recalculation is cum-

bersome and requires employing up to 70 individual beams for calculating the resultant plan. However, the BOID’s can be considered as an in air beam transmission distribution. The information can be used in a single con- volution calculation to provide the same result.

The use of OLIV provides information regarding daily variation that is specific to an institutional policy, a specific

clinical site, and individual patients. As more data are collected, this information will allow us to use radiation therapy planning systems with uncertainty algorithms that better predict the variation expected in any given clinical situation. Plan recalculation may act as interim step in the quality assurance procedure for a course of radiotherapy. The dosimetric information can be used to mod- ify the treatment to conform to the intended prescription.

As radiation therapy moves towards 3-D treatment planning and conformal treatment delivery, we may ex- pect more difficulties in estimating dose because of the possibility of increased set-up complexities and variations. An OLIV system will be critical to verify more complex radiation therapy techniques. Our method of dose recalculation will be particularly useful to show changes in dose volume histograms which are heavily relied upon for plan optimization.

CONCLUSION

On-line image verification provides us with information regarding daily set-up variation. The data can be used to provide a first order estimate of actual dose delivered to a target tumor volume and surrounding critical structures. The recalculation method proposed in this paper has an advantage over other planning systems that estimate dosimetric uncertainty because the data is derived directly from the patient under treatment. Eventually the information can be used to provide a more accurate estimate of dosimetric uncertainties in a variety of clinical situations.

REFERENCES

1.

2.

3.

4.

5.

6.

Brahme, A. Dosimetric precision requirements in radiation therapy. Acta Radiol. Oncol. 23:379-39 1; 1984. Byhardt, R. W.; Cox, J. D.: Hornburg, A.; Liermann, G. Weekly localization films and detection of field placement errors. Int. J. Radiat. Oncol. Biol. Phys. 4:881-887;1978. Goitein, M. Calculation of the uncertainty in the dose delivered during radiation therapy. Med. Phys. 12:608- 612;1985. Graham, M. L.; Cheng, A. Y.; Geer, L. Y.; Binns, W. R.; Vannier, M. W.; Wong, J. W. A method to analyze 2-dimensional daily radiotherapy portal images from an on-line fiber-optic imaging system. Int. J. Radiat. Oncol. Biol. Phys. 20:6 13-6 19; 199 1. Halverson, K. J.; Leung, T. C.; Pellet, J. B.; Gerber, R. L.; Weinhous, M. S.; Wong, J. W. Study of treatment variation in the radiotherapy of head and neck tumors using a fiberoptic on-line radiotherapy imaging system. Int. J. Radiat. Oncol. Biol. Phys. 2 1: 1327- 1336; 199 1. Herring, D. F. The consequences of dose response curves for tumor control and normal tissue injury on the precision necessary in patient management. Laryngoscope 85: 1112- 1118;1975.

7. Kinzie, J. J.; Hanks, G. E.; MacLean, C. J.; Kramer, S.

8.

9.

10.

Il.

12.

13.

Patterns of care study: Hodgkin’s disease relapse rates and adequacy of portals. Cancer 52:2223-2226; 1983. Leong, J. Implementation of random positioning error in computerised radiation treatment planning systems as a result of fractionation. Phys. Med. Biol. 32:327-334;1987. Marks, J. E.; Haus, A. G.; Sutton, H. G.; Griem, M. L. Localization error in the radiotherapy of Hodgkin’s disease and malignant lymphomas with extended mantle fields. Cancer 34:83-90; 1974. 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 3712755-276 1; 1976. National Cancer Institute contract NOl-CM-97564, Ra- diotherapy Treatment Planning Tools, March 1, 1989 thru February 28, 1994. Rabinowitz, I.; Broomberg, J.; Goitein, M.; McCarthy, K.; Leong, J. Accuracy of radiation field alignment in clinical practice. Int. J. Radiat. Oncol. Biol. Phys. 1 1:1857- 1867;1985. Thames, H. D.; Schultheiss, T. E.; Hendry, J. H.; Tucker, S. L.; Dubray, B. M.; Brock, W. A. Can modest excalations

716 I. J. Radiation Oncology 0 Biology 0 Physics Volume 27, Number 3. I993

14.

15.

16.

of dose be detected as increased tumor control. Int. J. Radiat. Oncol. Biol. Phys. 22:241-46; 199 1. Uric, M. M.; Goiten, M.; Koppke, K.; Kutcher, J. G.; LoSasso, T.; Mohan, R.; Munzenrider, J. E.; Sontag, M.; Wong, J. W. The role of uncertainty analysis in treatment planning. Int. J. Radiat. Oncol. Biol. Phys. 2 I:9 1- 107; 199 1.

van de Geijn, J. EXTDOS 7 1 Revised and expanded version of EXTDOS. A program for treatment planning in external beam therapy. Comp. Prog. Biomed. 2: 169-I 77;1972.

Williams, M. V.; Denekamp. J.: Fowler, J. F. Dose-response

17.

18.

relationships for human tumors: Implications for clinical trials of dose modifying agents. Int. J. Radiat. Oncol. Biol. Phys. 10:1703-1707;1984. Wong, J. W.; Binns, W. R.; Cheng, A.; Geer, L. Y.; Epstein, J. W.; Klarmann, J.; Purdy, J. A. On-line radiotherapy imaging with an array of fiberoptic image reducers. Int. J. Ra- diat. Oncol. Biol. Phys. 18: 1477-1484;1990. Wong, J. W.; Binns, W. R.; Cheng, A.; Epstein, J. W.; Klar- mann, J. A second generation high resolution fiberoptic radiotherapy imaging system (Abstr.). Int. J. Radiat. Oncol. Biol. Phys. 2 1 (S): 124; 199 1.

THE USE OF ONLINE IMAGE VERIFICATION TO ESTIMATE THE VARIATION IN RADIATION-THERAPY DOSE DELIVERY

Documents