0360-3016/90 $3.00 + .I0 copyright 0 1990 Pergamon Press plc
Inr. J, Radiarmn Oncology Eiol Phys.. Vol. 18. pp. 1411-1484 Printed in the U.S.A. All nghts reserved.
0 Technical Innovations and Notes ON-LINE RADIOTHERAPY IMAGING WITH AN ARRAY OF FIBER-OPTIC IMAGE REDUCERS JOHN W. WONG, PH.D.,’ W. ROBERT BINNS, PH.D.,~ ABEL Y. CHENG, M.SC.,~ LEWIS Y. GEER, B.A.,’ JOHN W. EPSTEIN,~ JOSEPH KLARMANN, PH.D.~ AND JAMES A. PURDY, PH.D.’ ‘Radiation Oncology Center, Mallinckrodt Institute of Radiology, ‘Department of Physics, Washington University, St. Louis, MO 63110; and ‘Fiber Imaging Inc., St. Louis, MO In the optical approach for on-line radiotherapy imaging, a large metal sheet-fluorescent screen combination is used to convert the radiation intensity distribution into a visible light image. Data are then captured via a mirror with a camera located out of the beam. Although usable portal images can be acquired, presence of the large mirror renders the system impractical in many treatment geometries. We have overcome this limitation by replacing the mirror with an array of 16 by 16 bundles of plastic fiber-optic image reducers. Each bundle, in turn, is made up of 16 by 16 individual optical fibers. The total of 256 by 256 fibers spans an input area of 40 cm by 40 cm with each individual fiber viewing an area of 1.6 mm by 1.6 mm. Within a height of 12 cm, each fiber is reduced to an area of 0.1 mm by 0.1 mm. The reduced portal image is then turned and “piped” to a final 3.0 cm by 3.0 cm output area. For data acquisition and digitization, the fiber output is directly coupled to the sensor of a TV camera interfaced to a small computer via a 512 by 512 frame grabber. In this initial evaluation, the imaging system has been characterized in terms of its line spread function, noise and resistance to radiation damage. Adequate phantom and patient images are presented. Radiotherapy treatment verification, On-line portal imaging.
patient repositioning is imperative for successful radiotherapy. Significant daily setup variations have been noted in the treatment of various disease sites, (2, 3, 7, 11) demonstrating the need for frequent treatment verification. As daily exposure of conventional port film is not practical, efforts have been made recently to develop on-line radiotherapy imaging devices (1, 4, 5, 9, 12, 13). These devices are likely to be available for clinical use in the near future. They will also play a critical role in the verification of new dynamic, or conformal, treatment approaches using therapy machines equipped with computer-controlled multileaf collimators. At present, scanning and optical systems are used for on-line radiotherapy imaging. In the scanning devices, ( 1, 4, 13) linear arrays of radiation detectors are used to measure, or scan, the portal radiation intensity sequentially at pre-determined spatial intervals. The data are then reconstructed to provide a digital portal image. In the optical
devices, ($9, 12) the entire radiation intensity is converted into an optical light image with a conventional x-ray fluorescent screen placed underneath the patient. Typically, a thin metal sheet is placed above the screen to act as a photon-to-electron converter to increase the light conversion efficiency of the screen. For data acquisition, the optical image is reflected out of the radiation field via a 45” mirror for capture with a TV camera interfaced to a small computer. The several optical devices differ mainly in the choice of TV camera tubes, such as the siliconintensified target vidicon or the plumbicon. Both the scanning and optical radiotherapy imaging devices have been used successfully in preliminary clinical applications. Adequate portal images can be acquired in less than 4 set of beam exposure. The optical system is relatively easy to assemble with all its parts commercially available. It contains no active mechanical scanning components. With the rapid l/30 set scanning sequence of the TV camera, fluctuation of radiation intensity during operation is less of a problem in comparison to other
Reprint requests to: John W. Wong, Ph.D., Mallinckrodt Institute of Radiology, Physics Section, 5 10 South Kingshighway Blvd., St. Louis, MO 63 110. Acknowledgements-The authors wish to thank Aaron Fenster and Peter Munro for their suggestions,Martin Israel for his initial
participation, and Brian Clevinger for his enthusiastic support and discussion throughout the course of the project. Supported in parts by the Alafi Washington Company, St. Louis. Accepted for publication 20 December 1989.
INTRODUCIION Accurate
1471
1478
I. J. Radiation Oncology 0 Biology 0 Physics
June 1990, Volume 18, Number 6
scanning devices. Munro d al (9) have demonstrated that their system is capable of resolving better than a one linepair per mm and detecting a 1.5 cm object at 0.8% contrast. Note that their system differs from the more conventional optical systems by integrating signal both on the camera sensor and in the computer buffer which would reduce readout noise. Unfortunately, despite the many desirable features of the optical systems, the presence of the bulky mirror assembly renders them cumbersome for general clinical use. This manuscript describes our effort to overcome the drawback of the mirror in the optical approach by replacing it with a 2-dimensional array of fiber-optic image reducers.
METHODS
AND
MATERIALS
A prototype system was constructed. Evaluations ol several imaging capabilities were made with this first model for the purpose of further refinement.
The fiber-optic imaging system Fiber-optic image reducers were made from plastics. Beginning with a “pre-formed boule”, consisting of a single clear polystyrene column encased within a thin acrylic cladding, single fiber units were produced by drawing the boule vertically through a ring-shaped oven. The crosssectional area of the single fiber was 1.6 mm by I .6 mm with a tolerance of about 0.05 mm. The difference in the refractive indices of the two plastics allowed for “lightpiping”, provided that the angle of the incident light satisfied the condition of internal reflection at the cladding. In the second stage, a reducer bundle was made by fusing 16 by 16 of the single fibers at elevated temperature into a “multi-boule”, and passing that through the fiber drawing device. The result was a “multi-fiber” consisting of 16 by 16 individual fibers of 0.1 mm by 0.1 mm crosssectional area. To form a compact image reducer that would pipe images out of the radiation field, the transition portion from the multi-boule into a multi-fiber was turned 90” within a height of 10.0 cm as shown in Figure 1. For viewing a large field, a 2-dimensional array of 16 by 16 bundles of the plastic fiber-optic image reducers was assembled within a light-tight housing as shown in Figure 2a. The exterior of each bundle was painted black to minimize inter-bundle crosstalk. The assembly, which will be referred to as the fiber-optic imager, is 12 cm tall which includes an excess of 2 cm thick of material. The extra height and length of the device was intended for convenient modifications during laboratory evaluations. In summary, the device consists of 256 by 256 fibers
* Hamamatsu Photonic Systems Corporation, Box 6910, Bridgewater, NJ 08807-09 10, USA. + International Business Machines Corporation, P.O. Box 1326-W, Boca Raton,
FL 33429-l 328, USA.
Fig. 1. A single fiber-optic reducer bundle with 16 by 16 fibers. each of I .6 mm by I .6 mm at the input and 0.1 mm by 0.1 mm at the output. The bundle had been turned 90’ within IO cm.
viewing an area of 40 cm by 40 cm. Each individual fiber has an input area of 1.6 mm by 1.6 mm. At the output. each fiber is reduced to an area of 0.1 mm by 0.1 mm. The output image area is 3 cm by 3 cm. For on-line radiotherapy imaging, a prototype optical system using the fiber-optic imager instead of a mirror was assembled, and is shown schematically in Figure 2b. A 3 mm copper metal sheet was used as the photon-toelectron converter. A fluorescent screen with 300 mg per cm’ of gadolinium oxysulphide was used as the optical light converter. For efficient light collection, the output end of the fiber-optic imager was coupled directly onto a TV camera* equipped with a glass optical fiber faceplate. Optical coupling grease was used. The camera was interfaced to a small personal computert via a frame grabber* for digitizing the output image into 5 12 by 5 12 pixels. The TV camera had a 2.5 cm diagonal sensor but only allowed viewing of a 20 cm by 16 cm input image in the present studies due to the manufacturer’s setting of the scanning limit of the TV tube.
Siudies of‘system characteristics Our present studies involved beams from a linear accelerator$. of the device were studied:
* Data Translation, 01752, USA. b Varian Radiation 94303, USA.
Inc.,
only the 4 MV x-ray Several characteristics
100 Locke Drive,
Marlboro,
MA
Division, 6 1I Hansen Way, Palo Alto, CA
1479
On-line radiotherapy 0 J. W. WONG et a/.
: : :
: : :
/Patient
escence Screen
Fig. 2. (a) The 40 cm by 40 cm fiber-optic imager. (b) Schematic of a prototype optical system using the fiber-
optic imager instead of a mirror.
Resistance to radiation damage was investigated by irradiating a single fiber bundle with 16 by 16 fibers to 1 million cGy with a high dose rate animal irradiator. Flood field measurements of light transmission from an irradiated fluorescent screen were made of the 40 cm
by 40 cm fiber-optic imager at different beam exposure times. The camera gain setting was representative of the one used in the phantom and patient studies. The data were then used to examine the noise characteristics of the system. However, in depth analysis of the noise contribution from the different components were not made in the present studies. The line spread function of the system was measured to determine its resolving power. The approach was similar to that used by Munro et al. (8) As the resolution was limited by the input fiber area, an 1 mm slit beam, collimated by two highly polished 12 cm tall copper blocks,
was used. The blocks were placed directly on the surface of the metal photon-to-electron converter. The slit beam was aligned parallel with one of the axis of the fiber-optic imager. Slit images were acquired as the slit beam was moved across individual fibers with the aid of a micrometer attachment. Translational line spread functions were determined. Diagonal line spread functions were not examined.
Phantoms and patients studies The fiber-optic imaging system was used to acquire 4 MV x-ray images of a therapy image quality assurance (Lutz) phantom, (6) and an anthropomorphic (Rando) head phantom**. Portal images of the lower neck field in the treatment of a patient with base of tongue cancer were also made. All phantoms and the patient irradiations were set at a source-to-surface (SSD) distance of 80 cm.
** Alderson Research Laboratories, Inc., 390 Ludlow St, P.O. Box 127 1, Stamford, CT 06904, USA.
The imaging system was set at a distance of 108 cm from the source. Images were acquired and digitized at the l/30 set video rate. The gain and the d.c. offset of the camera was adjusted to give a large count without saturating the g-bit resolution of the frame grabber. For improving data statistics, the 8-bit frames were averaged in 16-bit with the on-board utility function of the frame grabber to produce a final image. The frame averaging utility required an additional three frame-time for each averaging operation before the next frame could be acquired, thus limiting the full utilization of the beam time. Although the bundle to bundle packing at the input was within a tolerance of 0.05 mm, minor irregularities in the alignment of the fiber bundles at the output, of the order of 0.1 mm, would result in appreciable distortion of the image due to the small fiber output dimensions. The problem was further compounded by the minor irregularities of the bundle shapes. A software program was written to provide first order correction of the distortion based on the measured coordinates of the bundle corners at the input. The method corrected for the misalignment of the bundles but not that of the individual fibers. Residual image distortion at the fiber level remained. The imaging system was also inherently non-uniform due to variations in screen thickness, TV camera phosphor, light transmission in the bundles, etc. Therefore, in addition to the alignment correction, the phantom and patient images were corrected for the non-uniformity by first dividing the data with that of a flood field. Finally, to improve visualization of low contrast objects, the images were enhanced by either simple stretching of the display levels (i.e. windowing) or with the adaptive histogram equalization software imported from the University of North Carolina ( 10).
1480
1. J. Radiation Oncology 0 Biology 0 Physics
RESULTS
System characteristics There was less than 5% loss in the light transmission through a fiber bundle after it had been irradiated with 1 million cGy. Within a single fiber bundle, there was a variation of f 15% in light transmission with the larger deviations occurring near the periphery of the bundle. Figure 3a shows an in-air flood field image demonstrating the noticeable non-uniformity of the system. The mean light output of the bundle with the highest transmission was 1.6 that of the lowest. The image also showed the image distortion caused by misalignment of the output fibers. Figure 3b shows the same flood field after alignment correction. Since the dead spaces between bundles at the output did not contain information, they were also eliminated. The results showed that the operation of our first prototype was at a suboptimal level. The significant non-uniformity reduced the dynamic range of the present system from the ideal 256 levels to about 120 levels. The noise characteristics of the imaging system was examined with variance analysis of local areas on the flood field images. A baseline uniformity map, whose values were assumed to have zero uncertainty, was obtained by averaging 4,800 l/30 set flood field image frames. The measurements were equivalent to an exposure of 1,460 cGy to the imager, of which only 365 cGy was used to form the data. A separate set of flood fields acquired with 4 to 100 l/30 set frames, equivalent to 0.3 cGy to 7.5 cGy of useful data, were then normalized with the baseline uniformity map. Eight regions of interest on the normalized images, each approximately 2 cm by 2 cm of input area, were analyzed for the mean of the digitized signals and their standard deviations (s.d.). The ratio of 1 s.d. to the mean, that is the coefficient of variance, averaged over the 8 regions, is shown in Figure 4. The coefficient of
Fig. 3. (a) In-air flood field image before bundle alignment correction. The dead space between bundles was removed.
June 1990, Volume 18, Number 6
variance decreased from 1% to 0.5% for the dose range studied and did not behave as photon counting statistics. The gradient of decrease was particularly shallow after 1.5 cGy. Other factors, such as the structural noise due to the presence of the individual fiber cladding, contributed to the system noise characteristics. Figure 5 shows the best and the worst line spread functions of the 1 mm slit beam for the case when the beam was centered on one row of fibers, and when centered on the cladding between two rows. The central peak in the latter case is an inherent artifact of flood field normalization at the position of the cladding. For the former case, 10% of the light spilled over to the adjacent fiber. It is not clear in these initial studies whether the effects were due to fiber crosstalk or the spreading of the light from the fluorescent screen. Nevertheless, the results show that the present prototype is capable of resolving 0.3 line-pair per mm at the detector plane, and appears adequate for clinical applications where geometric magnifications usually exist. Phantom and patient studies Figures 6a, b, and c show the corrected images of the Lutz phantom filled with 20 cm depth of water. The Lutz phantom contained 13 1.6 cm diameter discs, each with a density 1.4 times that of water. The thickness of the discs ranged from 26 mm to 5 mm, representing an equivalent contrast of 10.4 mm to 2 mm thickness of water. Each disc also had a 3 mm by 3 mm notch cut out along its side. The maximum dose to the phantom was 33 cGy, representing 2 set of data acquisition out of 8 set of beam time. In (a), the image was enhanced with simple windowing; (b), the image of (a) was corrected for bundle misalignment; (c), the image of (a) was enhanced with adaptive histogram equalization followed by alignment correction. In the enhanced images, all the discs could be
corrections.
(b) The same flood field after alignment
1481
On-line radiotherapy 0 J. W. WONG et al.
Fig. 4. The best and worst line spread function of a 1 mm slit beam placed on top of the surface of the metal of the imaging system. The former was for the slit beam which traversed one row of fibers at the input while the
latter two rows.
seen, particularly in Figure 6c. The notches could be seen on the thicker discs. Figures 7a and b show alignment corrected images of the head region of the Rando phantom. The maximum dose to the phantom was 33 cGy. The image was enhanced with simple windowing in (a) and adaptive histogram equalization in (b). The frontal portion of the head phantom in the image was obscured due to saturation of the camera tube. For both Figures 6 and 7, the actual dose required to form the images was 8.5 cGy. Figures 8a and b show the alignment corrected portal images of the lower neck field in the patient treatment. In (a), windowing was used and (b) adaptive histogram equalization was used. The images were made by super-
ANALYSIS
OF
VARIANCES
positioning an open field image with one where the block was in place, thereby simulating the “double exposure” technique used in customary port film practice. In both cases, the open beam portion of the image was obtained with 6.7 cGy maximum dose to the patient. In the blocked portion, the maximum dose to the patient was 67 cGy. The actual doses required to form the open and blocked beam images were 1.7 cGy and 17 cGy respectively.
DISCUSSION As a first prototype, the fiber-optic imaging system was operating at a suboptimal level. Despite the limitations,
FOR
8
REGIONS
OF
INTEREST
1
0
d
Fig. 5. The noise characteristics
I
I
I
J
of the normalized flood fields expressed as the average coefficient of variance for 8 regions of interest as a function of actual dose required to form the flood field images. The dashed line shows the expected decrease in noise, with respect to the first data point, according to Poisson statistics.
1482
1. J. Radiation Oncology 0 Biology 0 Physics
June 1990, Volume 18. Number 6
Fig. 6. (a) Image of the Lutz phantom before alignment correction; (b) with windowing and bundle alignment correction; (c) with adaptive histogram equalization enhancement followed by alignment correction.
Fig. 7. (a) Image of the head region of a Rando phantom after windowing adaptive histogram equalization enhancement and alignment correction.
and alignment
correction;
(b) after
On-line radiotherapy 0 J. W.
WONG
et al.
1483
Fig. 8. (a) Superimposed image of the lower neck field to a patient after simple windowing enhancement and alignment correction. (b) The image in (a) after adaptive histogram equalization enhancement followed by alignment correction.
the feasibility studies were most encouraging. The Lutz phantom images showed that the system was capable of detecting about 1% contrast for objects of about 1.S cm diameter. The Rando head phantom and patient images also showed useful anatomical features. Note that with our present system, the exposure times were prolonged due to the 3 frames delay in the frame averaging utility of our particular frame grabber. The exposure, or dose to the patient and phantom would be reduced by a factor of 4 with a frame grabber capable of performing real-time frame averaging. Deficiencies of the present prototype are mainly due to the fact that every fiber-optic reducer that was drawn was used in its construction. Appreciable improvements can be readily, and are being, made. The non-uniformity in light transmission of the fiber reducer bundles which significantly limits the dynamic range of the system can be improved by preselecting the fiber reducers for uniform transmission with a standard light source. Image distortion due to the irregular output sizes of the fiber reducers can be reduced with the use of a laser micrometer to monitor the reducers for uniform dimensions during the drawing process. Simple quality control practices in the production of the fiber bundles will greatly enhance the quality of the final images. Finally, a rigorous correction of image distortion can be made by establishing a one-time geometric map relating the individual fiber position to a corresponding group of pixels in the frame grabber. The resolving capability of the present fiber-optic imager is inherently limited by the fiber dimension at the input and is inferior to that of a mirror system. Fibers of
++Shott Fiberoptics 01550, USA.
Inc., 122 Charlton
St, South Bridge, MA
1 mm by 1 mm input area, or smaller, can be readily produced to improve image resolution. On the other hand, for larger field of view, when the TV camera has to be moved further from the mirror, the characteristics of the two systems will be more comparable. Uniformity correction by division with a flood field image resulted in artifacts at the fiber cladding positions and contributed as structural noise in the image. These artifacts were more noticeable when the image pixel resolution was higher than that of the fiber, in our case four times, and can be alleviated by adjusting the resolution of the system to that dictated by the fiber dimensions. We have recently inserted a commercially available glass fiber reducer++ between the output of the plastic fibers and the faceplate of the camera so that the full 40 cm by 40 cm input area of the imager can be viewed. The corresponding reduction in pixel resolution of the frame grabber in effect acts as a low-pass filter and provides smoother images. Another obvious area of improvement would be the tuning of the electronics of the system for better signal-tonoise characteristics which were not made for the present studies. The optical on-line radiotherapy imaging system is relatively inexpensive and easy to assemble. It is appealing as a passive system. The use of fiber-optic reducers in the optical system enhances its potential as a practical imaging device for radiotherapy. Our feasibility studies were most encouraging, showing that the fiber-optic approach would be capable of adequate clinical imaging. Several areas of attainable improvement have been identified for successful clinical utilization with the next generation system.
1484
I. J. Radiation Oncology 0 Biology 0 Physics
June 1990, Volume 18, Number 6
REFERENCES 1. Bova, F. J.; Fitzgerald, L. T.; Mauderly, W. M.; Islam, M. J. Real time megavoltage imaging (Abstr.). Med. Phys.
14:707: 1987. 2. Huizenga, H.; Levendag, P. C.; De Porre, P. M. Z. R.; Visser. A. G. Accuracy in radiation field alignment in head and neck cancer: a prospective study. Radiother. Oncol. 11: 18 I187; 1988. 3. Kinzie, J. J.; Hanks, G. E.; Maclean, C. J.; Kramer, S. Patterns of care study: Hodgkin’s disease relapse rates and adequacy of portals. Cancer 52:2223-2226; 1983. 4. Lam, K. S.; Partowmah, M.; Lam, W. C. An on-line electronic portal imaging system for external beam radiotherapy. Br. J. Radiol. 59:1007-1013; 1986. 5. Leong, J. Use of digital fluoroscopy as an on-line verification device in radiation therapy. Phys. Med. Biol. 3 1:985-992: 1986. 6. Lutz, W. R.; Bjamgard, B. E. A test object for evaluation of portal films. Int. J. Radiat. Oncol. Biol. Phys. 11:63 l634; 1984. 7. Marks, J. E.; Haus, A. G.; Sutton, H. G.; Greim, M. L. The value of frequent verification films in reducing localization
8.
9.
10.
11.
12. 13.
error in the irradiation of complex fields. Cancer 37:27552761; 1976. Munro, P.; Rawlinson, J. A.; Fenster, A. Therapy imaging: a signal-to-noise analysis of metal plate/film detectors. Med. Phys. 14:975-984; 1987. Munro, P.; Rawlinson, J. A.; Fenster, A. A digital fluoroscopic imaging device for radiotherapy localization (Abstr.). Phys. Med. Biol. 33(Suppl. 1):45; 1988. Pizer, S. M.; Austin, J. D.; Perry, J. R.; Safrit, H. D.; Zimmerman, J. B. Adaptive histogram equalization for automatic contrast enhancement of medical images. Proc. SPIE 626:242-250; 1986. Rabinowitz, 1.; Broomberg, 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. Shalev, S.; Leszczynsky, K.; Lee, T. On-line portal verification (Abstr.). Phys. Med. Biol. 33(Suppl. 1):85; 1988. Van Herk, M.; Meertens, H. A matrix ionization chamber imaging device for on-line patient setup verification during radiotherapy. Radiother. Oncol. 1 1:369-378; 1988.
Inr. J, Radiarmn Oncology Eiol Phys.. Vol. 18. pp. 1411-1484 Printed in the U.S.A. All nghts reserved.
0 Technical Innovations and Notes ON-LINE RADIOTHERAPY IMAGING WITH AN ARRAY OF FIBER-OPTIC IMAGE REDUCERS JOHN W. WONG, PH.D.,’ W. ROBERT BINNS, PH.D.,~ ABEL Y. CHENG, M.SC.,~ LEWIS Y. GEER, B.A.,’ JOHN W. EPSTEIN,~ JOSEPH KLARMANN, PH.D.~ AND JAMES A. PURDY, PH.D.’ ‘Radiation Oncology Center, Mallinckrodt Institute of Radiology, ‘Department of Physics, Washington University, St. Louis, MO 63110; and ‘Fiber Imaging Inc., St. Louis, MO In the optical approach for on-line radiotherapy imaging, a large metal sheet-fluorescent screen combination is used to convert the radiation intensity distribution into a visible light image. Data are then captured via a mirror with a camera located out of the beam. Although usable portal images can be acquired, presence of the large mirror renders the system impractical in many treatment geometries. We have overcome this limitation by replacing the mirror with an array of 16 by 16 bundles of plastic fiber-optic image reducers. Each bundle, in turn, is made up of 16 by 16 individual optical fibers. The total of 256 by 256 fibers spans an input area of 40 cm by 40 cm with each individual fiber viewing an area of 1.6 mm by 1.6 mm. Within a height of 12 cm, each fiber is reduced to an area of 0.1 mm by 0.1 mm. The reduced portal image is then turned and “piped” to a final 3.0 cm by 3.0 cm output area. For data acquisition and digitization, the fiber output is directly coupled to the sensor of a TV camera interfaced to a small computer via a 512 by 512 frame grabber. In this initial evaluation, the imaging system has been characterized in terms of its line spread function, noise and resistance to radiation damage. Adequate phantom and patient images are presented. Radiotherapy treatment verification, On-line portal imaging.
patient repositioning is imperative for successful radiotherapy. Significant daily setup variations have been noted in the treatment of various disease sites, (2, 3, 7, 11) demonstrating the need for frequent treatment verification. As daily exposure of conventional port film is not practical, efforts have been made recently to develop on-line radiotherapy imaging devices (1, 4, 5, 9, 12, 13). These devices are likely to be available for clinical use in the near future. They will also play a critical role in the verification of new dynamic, or conformal, treatment approaches using therapy machines equipped with computer-controlled multileaf collimators. At present, scanning and optical systems are used for on-line radiotherapy imaging. In the scanning devices, ( 1, 4, 13) linear arrays of radiation detectors are used to measure, or scan, the portal radiation intensity sequentially at pre-determined spatial intervals. The data are then reconstructed to provide a digital portal image. In the optical
devices, ($9, 12) the entire radiation intensity is converted into an optical light image with a conventional x-ray fluorescent screen placed underneath the patient. Typically, a thin metal sheet is placed above the screen to act as a photon-to-electron converter to increase the light conversion efficiency of the screen. For data acquisition, the optical image is reflected out of the radiation field via a 45” mirror for capture with a TV camera interfaced to a small computer. The several optical devices differ mainly in the choice of TV camera tubes, such as the siliconintensified target vidicon or the plumbicon. Both the scanning and optical radiotherapy imaging devices have been used successfully in preliminary clinical applications. Adequate portal images can be acquired in less than 4 set of beam exposure. The optical system is relatively easy to assemble with all its parts commercially available. It contains no active mechanical scanning components. With the rapid l/30 set scanning sequence of the TV camera, fluctuation of radiation intensity during operation is less of a problem in comparison to other
Reprint requests to: John W. Wong, Ph.D., Mallinckrodt Institute of Radiology, Physics Section, 5 10 South Kingshighway Blvd., St. Louis, MO 63 110. Acknowledgements-The authors wish to thank Aaron Fenster and Peter Munro for their suggestions,Martin Israel for his initial
participation, and Brian Clevinger for his enthusiastic support and discussion throughout the course of the project. Supported in parts by the Alafi Washington Company, St. Louis. Accepted for publication 20 December 1989.
INTRODUCIION Accurate
1471
1478
I. J. Radiation Oncology 0 Biology 0 Physics
June 1990, Volume 18, Number 6
scanning devices. Munro d al (9) have demonstrated that their system is capable of resolving better than a one linepair per mm and detecting a 1.5 cm object at 0.8% contrast. Note that their system differs from the more conventional optical systems by integrating signal both on the camera sensor and in the computer buffer which would reduce readout noise. Unfortunately, despite the many desirable features of the optical systems, the presence of the bulky mirror assembly renders them cumbersome for general clinical use. This manuscript describes our effort to overcome the drawback of the mirror in the optical approach by replacing it with a 2-dimensional array of fiber-optic image reducers.
METHODS
AND
MATERIALS
A prototype system was constructed. Evaluations ol several imaging capabilities were made with this first model for the purpose of further refinement.
The fiber-optic imaging system Fiber-optic image reducers were made from plastics. Beginning with a “pre-formed boule”, consisting of a single clear polystyrene column encased within a thin acrylic cladding, single fiber units were produced by drawing the boule vertically through a ring-shaped oven. The crosssectional area of the single fiber was 1.6 mm by I .6 mm with a tolerance of about 0.05 mm. The difference in the refractive indices of the two plastics allowed for “lightpiping”, provided that the angle of the incident light satisfied the condition of internal reflection at the cladding. In the second stage, a reducer bundle was made by fusing 16 by 16 of the single fibers at elevated temperature into a “multi-boule”, and passing that through the fiber drawing device. The result was a “multi-fiber” consisting of 16 by 16 individual fibers of 0.1 mm by 0.1 mm crosssectional area. To form a compact image reducer that would pipe images out of the radiation field, the transition portion from the multi-boule into a multi-fiber was turned 90” within a height of 10.0 cm as shown in Figure 1. For viewing a large field, a 2-dimensional array of 16 by 16 bundles of the plastic fiber-optic image reducers was assembled within a light-tight housing as shown in Figure 2a. The exterior of each bundle was painted black to minimize inter-bundle crosstalk. The assembly, which will be referred to as the fiber-optic imager, is 12 cm tall which includes an excess of 2 cm thick of material. The extra height and length of the device was intended for convenient modifications during laboratory evaluations. In summary, the device consists of 256 by 256 fibers
* Hamamatsu Photonic Systems Corporation, Box 6910, Bridgewater, NJ 08807-09 10, USA. + International Business Machines Corporation, P.O. Box 1326-W, Boca Raton,
FL 33429-l 328, USA.
Fig. 1. A single fiber-optic reducer bundle with 16 by 16 fibers. each of I .6 mm by I .6 mm at the input and 0.1 mm by 0.1 mm at the output. The bundle had been turned 90’ within IO cm.
viewing an area of 40 cm by 40 cm. Each individual fiber has an input area of 1.6 mm by 1.6 mm. At the output. each fiber is reduced to an area of 0.1 mm by 0.1 mm. The output image area is 3 cm by 3 cm. For on-line radiotherapy imaging, a prototype optical system using the fiber-optic imager instead of a mirror was assembled, and is shown schematically in Figure 2b. A 3 mm copper metal sheet was used as the photon-toelectron converter. A fluorescent screen with 300 mg per cm’ of gadolinium oxysulphide was used as the optical light converter. For efficient light collection, the output end of the fiber-optic imager was coupled directly onto a TV camera* equipped with a glass optical fiber faceplate. Optical coupling grease was used. The camera was interfaced to a small personal computert via a frame grabber* for digitizing the output image into 5 12 by 5 12 pixels. The TV camera had a 2.5 cm diagonal sensor but only allowed viewing of a 20 cm by 16 cm input image in the present studies due to the manufacturer’s setting of the scanning limit of the TV tube.
Siudies of‘system characteristics Our present studies involved beams from a linear accelerator$. of the device were studied:
* Data Translation, 01752, USA. b Varian Radiation 94303, USA.
Inc.,
only the 4 MV x-ray Several characteristics
100 Locke Drive,
Marlboro,
MA
Division, 6 1I Hansen Way, Palo Alto, CA
1479
On-line radiotherapy 0 J. W. WONG et a/.
: : :
: : :
/Patient
escence Screen
Fig. 2. (a) The 40 cm by 40 cm fiber-optic imager. (b) Schematic of a prototype optical system using the fiber-
optic imager instead of a mirror.
Resistance to radiation damage was investigated by irradiating a single fiber bundle with 16 by 16 fibers to 1 million cGy with a high dose rate animal irradiator. Flood field measurements of light transmission from an irradiated fluorescent screen were made of the 40 cm
by 40 cm fiber-optic imager at different beam exposure times. The camera gain setting was representative of the one used in the phantom and patient studies. The data were then used to examine the noise characteristics of the system. However, in depth analysis of the noise contribution from the different components were not made in the present studies. The line spread function of the system was measured to determine its resolving power. The approach was similar to that used by Munro et al. (8) As the resolution was limited by the input fiber area, an 1 mm slit beam, collimated by two highly polished 12 cm tall copper blocks,
was used. The blocks were placed directly on the surface of the metal photon-to-electron converter. The slit beam was aligned parallel with one of the axis of the fiber-optic imager. Slit images were acquired as the slit beam was moved across individual fibers with the aid of a micrometer attachment. Translational line spread functions were determined. Diagonal line spread functions were not examined.
Phantoms and patients studies The fiber-optic imaging system was used to acquire 4 MV x-ray images of a therapy image quality assurance (Lutz) phantom, (6) and an anthropomorphic (Rando) head phantom**. Portal images of the lower neck field in the treatment of a patient with base of tongue cancer were also made. All phantoms and the patient irradiations were set at a source-to-surface (SSD) distance of 80 cm.
** Alderson Research Laboratories, Inc., 390 Ludlow St, P.O. Box 127 1, Stamford, CT 06904, USA.
The imaging system was set at a distance of 108 cm from the source. Images were acquired and digitized at the l/30 set video rate. The gain and the d.c. offset of the camera was adjusted to give a large count without saturating the g-bit resolution of the frame grabber. For improving data statistics, the 8-bit frames were averaged in 16-bit with the on-board utility function of the frame grabber to produce a final image. The frame averaging utility required an additional three frame-time for each averaging operation before the next frame could be acquired, thus limiting the full utilization of the beam time. Although the bundle to bundle packing at the input was within a tolerance of 0.05 mm, minor irregularities in the alignment of the fiber bundles at the output, of the order of 0.1 mm, would result in appreciable distortion of the image due to the small fiber output dimensions. The problem was further compounded by the minor irregularities of the bundle shapes. A software program was written to provide first order correction of the distortion based on the measured coordinates of the bundle corners at the input. The method corrected for the misalignment of the bundles but not that of the individual fibers. Residual image distortion at the fiber level remained. The imaging system was also inherently non-uniform due to variations in screen thickness, TV camera phosphor, light transmission in the bundles, etc. Therefore, in addition to the alignment correction, the phantom and patient images were corrected for the non-uniformity by first dividing the data with that of a flood field. Finally, to improve visualization of low contrast objects, the images were enhanced by either simple stretching of the display levels (i.e. windowing) or with the adaptive histogram equalization software imported from the University of North Carolina ( 10).
1480
1. J. Radiation Oncology 0 Biology 0 Physics
RESULTS
System characteristics There was less than 5% loss in the light transmission through a fiber bundle after it had been irradiated with 1 million cGy. Within a single fiber bundle, there was a variation of f 15% in light transmission with the larger deviations occurring near the periphery of the bundle. Figure 3a shows an in-air flood field image demonstrating the noticeable non-uniformity of the system. The mean light output of the bundle with the highest transmission was 1.6 that of the lowest. The image also showed the image distortion caused by misalignment of the output fibers. Figure 3b shows the same flood field after alignment correction. Since the dead spaces between bundles at the output did not contain information, they were also eliminated. The results showed that the operation of our first prototype was at a suboptimal level. The significant non-uniformity reduced the dynamic range of the present system from the ideal 256 levels to about 120 levels. The noise characteristics of the imaging system was examined with variance analysis of local areas on the flood field images. A baseline uniformity map, whose values were assumed to have zero uncertainty, was obtained by averaging 4,800 l/30 set flood field image frames. The measurements were equivalent to an exposure of 1,460 cGy to the imager, of which only 365 cGy was used to form the data. A separate set of flood fields acquired with 4 to 100 l/30 set frames, equivalent to 0.3 cGy to 7.5 cGy of useful data, were then normalized with the baseline uniformity map. Eight regions of interest on the normalized images, each approximately 2 cm by 2 cm of input area, were analyzed for the mean of the digitized signals and their standard deviations (s.d.). The ratio of 1 s.d. to the mean, that is the coefficient of variance, averaged over the 8 regions, is shown in Figure 4. The coefficient of
Fig. 3. (a) In-air flood field image before bundle alignment correction. The dead space between bundles was removed.
June 1990, Volume 18, Number 6
variance decreased from 1% to 0.5% for the dose range studied and did not behave as photon counting statistics. The gradient of decrease was particularly shallow after 1.5 cGy. Other factors, such as the structural noise due to the presence of the individual fiber cladding, contributed to the system noise characteristics. Figure 5 shows the best and the worst line spread functions of the 1 mm slit beam for the case when the beam was centered on one row of fibers, and when centered on the cladding between two rows. The central peak in the latter case is an inherent artifact of flood field normalization at the position of the cladding. For the former case, 10% of the light spilled over to the adjacent fiber. It is not clear in these initial studies whether the effects were due to fiber crosstalk or the spreading of the light from the fluorescent screen. Nevertheless, the results show that the present prototype is capable of resolving 0.3 line-pair per mm at the detector plane, and appears adequate for clinical applications where geometric magnifications usually exist. Phantom and patient studies Figures 6a, b, and c show the corrected images of the Lutz phantom filled with 20 cm depth of water. The Lutz phantom contained 13 1.6 cm diameter discs, each with a density 1.4 times that of water. The thickness of the discs ranged from 26 mm to 5 mm, representing an equivalent contrast of 10.4 mm to 2 mm thickness of water. Each disc also had a 3 mm by 3 mm notch cut out along its side. The maximum dose to the phantom was 33 cGy, representing 2 set of data acquisition out of 8 set of beam time. In (a), the image was enhanced with simple windowing; (b), the image of (a) was corrected for bundle misalignment; (c), the image of (a) was enhanced with adaptive histogram equalization followed by alignment correction. In the enhanced images, all the discs could be
corrections.
(b) The same flood field after alignment
1481
On-line radiotherapy 0 J. W. WONG et al.
Fig. 4. The best and worst line spread function of a 1 mm slit beam placed on top of the surface of the metal of the imaging system. The former was for the slit beam which traversed one row of fibers at the input while the
latter two rows.
seen, particularly in Figure 6c. The notches could be seen on the thicker discs. Figures 7a and b show alignment corrected images of the head region of the Rando phantom. The maximum dose to the phantom was 33 cGy. The image was enhanced with simple windowing in (a) and adaptive histogram equalization in (b). The frontal portion of the head phantom in the image was obscured due to saturation of the camera tube. For both Figures 6 and 7, the actual dose required to form the images was 8.5 cGy. Figures 8a and b show the alignment corrected portal images of the lower neck field in the patient treatment. In (a), windowing was used and (b) adaptive histogram equalization was used. The images were made by super-
ANALYSIS
OF
VARIANCES
positioning an open field image with one where the block was in place, thereby simulating the “double exposure” technique used in customary port film practice. In both cases, the open beam portion of the image was obtained with 6.7 cGy maximum dose to the patient. In the blocked portion, the maximum dose to the patient was 67 cGy. The actual doses required to form the open and blocked beam images were 1.7 cGy and 17 cGy respectively.
DISCUSSION As a first prototype, the fiber-optic imaging system was operating at a suboptimal level. Despite the limitations,
FOR
8
REGIONS
OF
INTEREST
1
0
d
Fig. 5. The noise characteristics
I
I
I
J
of the normalized flood fields expressed as the average coefficient of variance for 8 regions of interest as a function of actual dose required to form the flood field images. The dashed line shows the expected decrease in noise, with respect to the first data point, according to Poisson statistics.
1482
1. J. Radiation Oncology 0 Biology 0 Physics
June 1990, Volume 18. Number 6
Fig. 6. (a) Image of the Lutz phantom before alignment correction; (b) with windowing and bundle alignment correction; (c) with adaptive histogram equalization enhancement followed by alignment correction.
Fig. 7. (a) Image of the head region of a Rando phantom after windowing adaptive histogram equalization enhancement and alignment correction.
and alignment
correction;
(b) after
On-line radiotherapy 0 J. W.
WONG
et al.
1483
Fig. 8. (a) Superimposed image of the lower neck field to a patient after simple windowing enhancement and alignment correction. (b) The image in (a) after adaptive histogram equalization enhancement followed by alignment correction.
the feasibility studies were most encouraging. The Lutz phantom images showed that the system was capable of detecting about 1% contrast for objects of about 1.S cm diameter. The Rando head phantom and patient images also showed useful anatomical features. Note that with our present system, the exposure times were prolonged due to the 3 frames delay in the frame averaging utility of our particular frame grabber. The exposure, or dose to the patient and phantom would be reduced by a factor of 4 with a frame grabber capable of performing real-time frame averaging. Deficiencies of the present prototype are mainly due to the fact that every fiber-optic reducer that was drawn was used in its construction. Appreciable improvements can be readily, and are being, made. The non-uniformity in light transmission of the fiber reducer bundles which significantly limits the dynamic range of the system can be improved by preselecting the fiber reducers for uniform transmission with a standard light source. Image distortion due to the irregular output sizes of the fiber reducers can be reduced with the use of a laser micrometer to monitor the reducers for uniform dimensions during the drawing process. Simple quality control practices in the production of the fiber bundles will greatly enhance the quality of the final images. Finally, a rigorous correction of image distortion can be made by establishing a one-time geometric map relating the individual fiber position to a corresponding group of pixels in the frame grabber. The resolving capability of the present fiber-optic imager is inherently limited by the fiber dimension at the input and is inferior to that of a mirror system. Fibers of
++Shott Fiberoptics 01550, USA.
Inc., 122 Charlton
St, South Bridge, MA
1 mm by 1 mm input area, or smaller, can be readily produced to improve image resolution. On the other hand, for larger field of view, when the TV camera has to be moved further from the mirror, the characteristics of the two systems will be more comparable. Uniformity correction by division with a flood field image resulted in artifacts at the fiber cladding positions and contributed as structural noise in the image. These artifacts were more noticeable when the image pixel resolution was higher than that of the fiber, in our case four times, and can be alleviated by adjusting the resolution of the system to that dictated by the fiber dimensions. We have recently inserted a commercially available glass fiber reducer++ between the output of the plastic fibers and the faceplate of the camera so that the full 40 cm by 40 cm input area of the imager can be viewed. The corresponding reduction in pixel resolution of the frame grabber in effect acts as a low-pass filter and provides smoother images. Another obvious area of improvement would be the tuning of the electronics of the system for better signal-tonoise characteristics which were not made for the present studies. The optical on-line radiotherapy imaging system is relatively inexpensive and easy to assemble. It is appealing as a passive system. The use of fiber-optic reducers in the optical system enhances its potential as a practical imaging device for radiotherapy. Our feasibility studies were most encouraging, showing that the fiber-optic approach would be capable of adequate clinical imaging. Several areas of attainable improvement have been identified for successful clinical utilization with the next generation system.
1484
I. J. Radiation Oncology 0 Biology 0 Physics
June 1990, Volume 18, Number 6
REFERENCES 1. Bova, F. J.; Fitzgerald, L. T.; Mauderly, W. M.; Islam, M. J. Real time megavoltage imaging (Abstr.). Med. Phys.
14:707: 1987. 2. Huizenga, H.; Levendag, P. C.; De Porre, P. M. Z. R.; Visser. A. G. Accuracy in radiation field alignment in head and neck cancer: a prospective study. Radiother. Oncol. 11: 18 I187; 1988. 3. Kinzie, J. J.; Hanks, G. E.; Maclean, C. J.; Kramer, S. Patterns of care study: Hodgkin’s disease relapse rates and adequacy of portals. Cancer 52:2223-2226; 1983. 4. Lam, K. S.; Partowmah, M.; Lam, W. C. An on-line electronic portal imaging system for external beam radiotherapy. Br. J. Radiol. 59:1007-1013; 1986. 5. Leong, J. Use of digital fluoroscopy as an on-line verification device in radiation therapy. Phys. Med. Biol. 3 1:985-992: 1986. 6. Lutz, W. R.; Bjamgard, B. E. A test object for evaluation of portal films. Int. J. Radiat. Oncol. Biol. Phys. 11:63 l634; 1984. 7. Marks, J. E.; Haus, A. G.; Sutton, H. G.; Greim, M. L. The value of frequent verification films in reducing localization
8.
9.
10.
11.
12. 13.
error in the irradiation of complex fields. Cancer 37:27552761; 1976. Munro, P.; Rawlinson, J. A.; Fenster, A. Therapy imaging: a signal-to-noise analysis of metal plate/film detectors. Med. Phys. 14:975-984; 1987. Munro, P.; Rawlinson, J. A.; Fenster, A. A digital fluoroscopic imaging device for radiotherapy localization (Abstr.). Phys. Med. Biol. 33(Suppl. 1):45; 1988. Pizer, S. M.; Austin, J. D.; Perry, J. R.; Safrit, H. D.; Zimmerman, J. B. Adaptive histogram equalization for automatic contrast enhancement of medical images. Proc. SPIE 626:242-250; 1986. Rabinowitz, 1.; Broomberg, 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. Shalev, S.; Leszczynsky, K.; Lee, T. On-line portal verification (Abstr.). Phys. Med. Biol. 33(Suppl. 1):85; 1988. Van Herk, M.; Meertens, H. A matrix ionization chamber imaging device for on-line patient setup verification during radiotherapy. Radiother. Oncol. 1 1:369-378; 1988.
