Int J Radiation Oncology Biol Phys , Vol 45, No 3, pp 773–789, 1999
`Copyright © 1999 Elsevier Science Inc
`Printed in the USA All rights reserved
`0360-3016/99/$–see front matter
`
`PII S0360-3016(99)00118-2
`
`PHYSICS CONTRIBUTION
`
`A RADIOGRAPHIC AND TOMOGRAPHIC IMAGING SYSTEM INTEGRATED
`INTO A MEDICAL LINEAR ACCELERATOR FOR LOCALIZATION OF BONE
`AND SOFT-TISSUE TARGETS
`
`DAVID A. JAFFRAY, PH.D.,*† DOUGLAS G. DRAKE, B.S.,*† MICHEL MOREAU, PH.D.,*
`ALVARO A. MARTINEZ, M.D.,* AND JOHN W. WONG, PH.D.*†
`
`*Department of Radiation Oncology, William Beaumont Hospital, Royal Oak, MI; and †Department of Physics,
`Oakland University, Rochester, MI
`
`Purpose: Dose escalation in conformal radiation therapy requires accurate field placement. Electronic portal
`imaging devices are used to verify field placement but are limited by the low subject contrast of bony anatomy
`at megavoltage (MV) energies, the large imaging dose, and the small size of the radiation fields. In this article,
`we describe the in-house modification of a medical linear accelerator to provide radiographic and tomographic
`localization of bone and soft-tissue targets in the reference frame of the accelerator. This system separates the
`verification of beam delivery (machine settings, field shaping) from patient and target localization.
`Materials and Methods: A kilovoltage (kV) x-ray source is mounted on the drum assembly of an Elekta SL-20
`medical linear accelerator, maintaining the same isocenter as the treatment beam with the central axis at 90° to
`the treatment beam axis. The x-ray tube is powered by a high-frequency generator and can be retracted to the
`drum-face. Two CCD-based fluoroscopic imaging systems are mounted on the accelerator to collect MV and kV
`radiographic images. The system is also capable of cone-beam tomographic imaging at both MV and kV energies.
`The gain stages of the two imaging systems have been modeled to assess imaging performance. The contrast-
`resolution of the kV and MV systems was measured using a contrast-detail (C-D) phantom. The dosimetric
`advantage of using the kV imaging system over the MV system for the detection of bone-like objects is quantified
`for a specific imaging geometry using a C-D phantom. Accurate guidance of the treatment beam requires
`registration of the imaging and treatment coordinate systems. The mechanical characteristics of the treatment
`and imaging gantries are examined to determine a localizing precision assuming an unambiguous object. MV and
`kV radiographs of patients receiving radiation therapy are acquired to demonstrate the radiographic perfor-
`mance of the system. The tomographic performance is demonstrated on phantoms using both the MV and the
`kV imaging system, and the visibility of soft-tissue targets is assessed.
`Results and Discussion: Characterization of the gains in the two systems demonstrates that the MV system is
`x-ray quantum noise-limited at very low spatial frequencies; this is not the case for the kV system. The estimates
`of gain used in the model are validated by measurements of the total gain in each system. Contrast-detail
`measurements demonstrate that the MV system is capable of detecting subject contrasts of less than 0.1% (at 6
`and 18 MV). A comparison of the kV and MV contrast-detail performance indicates that equivalent bony object
`detection can be achieved with the kV system at significantly lower doses (factors of 40 and 90 lower than for 6
`and 18 MV, respectively). The tomographic performance of the system is promising; soft-tissue visibility is
`demonstrated at relatively low imaging doses (3 cGy) using four laboratory rats.
`Conclusions: We have integrated a kV radiographic and tomographic imaging system with a medical linear
`accelerator to allow localization of bone and soft-tissue structures in the reference frame of the accelerator.
`Modeling and experiments have demonstrated the feasibility of acquiring high-quality radiographic and tomo-
`graphic images at acceptable imaging doses. Full integration of the kV and MV imaging systems with the
`treatment machine will allow on-line radiographic and tomographic guidance of field placement. © 1999
`Elsevier Science Inc.
`
`Portal imaging, Conebeam computed tomography, Kilovoltage, Megavoltage.
`
`Oral Presentation at the 1997 ASTRO Meeting, Orlando, FL—
`October 1997
`Reprint requests to: D. Jaffray, Ph.D., Dept. of Radiation Oncol-
`ogy, William Beaumont Hospital, Royal Oak, MI 48073. Tel: (248)
`551-7024; Fax: (248) 551-0089; E-mail: djaffray@beaumont.edu.
`Acknowledgements: We would like to acknowledge Mr. Rob
`Cooke, Mr. Kevin Brown, and Dr. Di Yan for valuable discus-
`sions; Mr. John Musselwhite for his engineering support; and, Mr.
`Daniel Bilsky, Mr. Walter Jendhoff, Mr. Bill Porteners, Mr. Gabe
`
`Blosser, and Mr. John Kuchar for their technical assistance
`throughout the development of the dual-beam system. We thank
`Dr. Peter Munro for the long-term loan of the contrast-detail
`phantom. Initial interactions with Dr. Paul Cho and Dr. Cheng Pan
`are greatly appreciated. This development would not have been
`possible without the technical support provided by Elekta Oncol-
`ogy Systems. This project is funded in-part by the National Cancer
`Institute (CA-66074).
`Accepted for publication 25 March 1999.
`
`773
`
`Page 1 of 17
`
`Elekta Exhibit 1010
`
`

`
`774
`
`I. J. Radiation Oncology c Biology c Physics
`
`Volume 45, Number 3, 1999
`
`INTRODUCTION
`
`The escalation of tumor dose in radiotherapy promises
`increased probability of disease control (1). The proximity
`of the target to normal tissues makes dose escalation chal-
`lenging. Safe pursuit of these higher doses has spurred the
`development of conformal therapy techniques. These tech-
`niques attempt to conform the radiation field to a well-
`specified target volume. The success of the conformal ther-
`apy approach requires (i) accurate characterization of the
`uncertainties in field placement to create reasonable margins
`for use in the planning process, and (ii) verification of the
`treatment delivery.
`Numerous studies of field placement error have been
`reported in the literature over the past 10 years (2–5). These
`studies have demonstrated that on-line megavoltage portal
`imaging is useful in measuring field placement uncertainty.
`Many of these studies also report on the poor quality of the
`portal images and how the quality of the images hampers
`the detection of field placement errors (5).
`The primary reason for the poor quality of the megavolt-
`age portal images is the intrinsically low subject contrast of
`bony anatomy at megavoltage energies (6). The dramatic
`drop in contrast with increasing x-ray energy requires that
`the noise introduced by the imaging system be extremely
`low. In the case of conventional verification/localization
`film, the low contrasts are masked by film noise and the
`fixed display contrast of the film (7). Fluoroscopic systems
`suffer from poor light collection and electronics noise in the
`light sensor and readout electronics (8). While many inves-
`tigators continue to investigate methods of reducing noise in
`these systems, we have taken a more direct approach: to
`increase the subject contrast by using a kilovoltage (kV)
`x-ray source to localize the placement of the field.
`Using a kV x-ray source to determine field placement is
`not a novel concept. Johns et al. (9) report the design of a
`60Co treatment unit with a kV x-ray tube in the head to
`guide field placement. Investigators have reported on the
`addition of an x-ray tube to the head of an accelerator to
`generate kV radiographs of the patient in treatment position
`(10 –12). Other investigators have demonstrated that a low-
`energy x-ray beam can be produced by a low atomic num-
`ber, transmission target in the head of the accelerator (13).
`In 1992, Suit et al. (14) urged the community to develop
`better techniques for verifying field placement, proposing
`the addition of a kV source to the accelerator. In a previous
`report, we described the development of a prototype digital
`imaging system for kV localization on a medical linear
`accelerator (15).
`Beyond the obvious advantage of improved image quality
`for portal imaging, the development of an on-line kV im-
`aging system can provide other advantages. The dose de-
`livered in a kV exposure is significantly lower than that
`required for a megavoltage image. This reduction in dose
`allows more frequent imaging with open-fields, and the
`acquisition of images at non-treatment gantry angles, which
`may improve the precision of localization. In this article, we
`
`report on the characteristics and performance of an MV
`imaging system and a kV imaging system that have been
`integrated with a medical linear accelerator for radiographic
`localization. The dosimetric advantage of kV radiographs
`for localization is examined and the potential for tomo-
`graphic guidance is demonstrated with the system using
`cone-beam computed tomography (CB-CT). The combined
`system consisting of the medical accelerator and the two
`on-line imaging systems (kV and MV) is referred to as the
`dual-beam system (DBS).
`
`MATERIALS AND METHODS
`
`Dual-beam system
`Overview. Photographs of the DBS are shown in Fig. 1.
`Two fluoroscopic imager assemblies are attached to the
`accelerator; one detects the megavoltage (MV) treatment
`beam, the other detects the kV beam projected at 90° to the
`treatment beam axis. An Elekta SL-20 (Elekta Oncology
`Systems, Crawley, UK) linear accelerator forms the basis of
`the system. This accelerator is computer controlled and
`produces 6 and 18 MV photon beams. Field shaping is
`performed with an 80 leaf collimator. The SL-20 is a
`drum-based accelerator, making the installation of a retract-
`able kV x-ray source relatively straightforward (Fig. 1d).
`A 600,000 heat unit (HU) x-ray tube (Eureka Rad-92 in
`Varian Sapphire housing, 0.6 and 1.2 mm focal spots, 12.5°
`rotating anode; Varian X-ray Tube Products, Arlington
`Heights, IL, USA) with a manual collimation system has
`been installed on the SL-20 with the focal spot located 100
`cm from the machine isocenter. The tube is supported by 3
`hardened steel shafts (nominal 2.5” diameter) that retract
`into the accelerator’s drum structure (Fig. 1d). The shafts
`are supported by three bearing assemblies set in an alumi-
`num mount. The aluminum mount is attached to the face of
`the accelerator drum structure at the position corresponding
`to a 90° gantry angle. Adjustments are built into the assem-
`bly for focal spot alignment. The total weight of the x-ray
`tube and mechanical components is approximately 90 kg.
`Under this load, and at full extension, the predicted deflec-
`tion in focal spot position was approximately 0.2 cm. The
`x-ray tube is powered by a 40 kW high-frequency, radio-
`graphic generator (Innerscan Inc., Chicago, IL, USA). The
`exposure-to-charge ratio in air for this tube is measured
`to be 5.7 mR/mAs at isocenter (100 cm) for a 90 kVp
`spectrum. The tube spectrum has been modeled based on
`measured first and second HVLs of aluminum (Al) (HVL1
`5 9.5 mm) (16).
`5 4.0 mm; HVL2
`The MV radiographic imager assembly is shown sche-
`matically in Fig. 2 and is similar in design to that reported
`by Atari et al. (17). Briefly, a phosphor screen (Gd2O2S:Tb,
`Fuji HR-16 [back], 165 mg/cm2) mounted on a 1.5-mm
`stainless steel plate is used for primary x-ray detection. The
`light emitted by the phosphor screen is reflected by three
`front-surface mirrors and focussed by a lens (f/0.95 operat-
`ing at f/1.4, 50 mm) on to a 512 3 512 back-thinned
`charge-coupled device (CCD) (Sensor SI502A, Scientific
`
`Page 2 of 17
`
`

`
`Imaging systems integrated into a linear accelerator c D. A. JAFFRAY et al.
`
`775
`
`Fig. 1. Photographs of the dual-beam system. The dual-beam system was constructed on a Elekta SL-20 medical linear
`accelerator. The design allows normal operation of the accelerator with the imagers removed and the kV x-ray tube
`retracted (a). The imager can be easily attached for kV localization. The kV tube is extended manually to reach the
`source plane (b). The focal spot of the x-ray tube is at 100 cm SAD providing the same imaging geometry as that of
`the MV source. This provides a large aperture for conebeam CT with the system (c). The drum structure of the SL-20
`simplifies the addition of the retractable arm of the x-ray tube. Three hardened-steel shafts support the x-ray tube (d).
`
`Imaging Technologies [SITE] Inc., Beaverton, OR, USA).
`Back-thinning provides a quantum efficiency of 78% at 550
`nm (peak in the Gd2O2S:Tb spectrum). The CCD sensor is
`housed in a Photometrics CH-250 camera housing (Photo-
`metrics Inc., Tucson, AZ, USA). The sensor is cooled to
`240°C to minimize the dark current collected in the sensor
`during read-out. The sensor has a full-well capacity of
`280,000 electrons, corresponding to the full-range of the
`12-bit digitizer. A shutter is installed to minimize the effects
`of light leaks in the imager housing. The camera is read out
`under computer control at 500 kpixels/s. The long read-out
`period (0.5 s) of the camera prevents operation in a fluoro-
`scopic mode.
`The kV imager assembly is identical to the MV assembly
`with the exception of one less mirror in the optical path and
`the replacement of the 1.5 mm stainless steel plate with a
`thin Al plate (1/32”) for support of the phosphor screen. The
`kV imager assembly is a refined design of the original MV
`assembly; the third mirror has been eliminated through
`modifications to the mounting assembly. The presence of
`the third mirror does not significantly alter the performance
`
`of the MV system. In the case of the megavoltage detector,
`the stainless steel plate increases the interaction probability
`for the MV X rays and reduces the signal generated by
`scattered X rays exiting the patient. The imager housing
`(Fig. 1b) for the MV system has been modified to allow
`retraction by ;34 cm. The kV imager housing is non-
`retractable and is attached and removed before and after an
`imaging procedure.
`An in-depth characterization of the imaging performance
`of the kV and MV radiographic systems is on-going and is
`not the focus of this article.
`Control system. The DBS is slaved to the operation of the
`linear accelerator. A personal computer (Pentium-based)
`controls the x-ray generator and the acquisition of images
`and also monitors the gantry angle reported by the acceler-
`ator. A schematic of the control system is shown in Fig. 3.
`The generator is controlled via an RS-232 port, allowing
`full-control of the generator’s parameters and control of the
`exposure. The cameras are read via two ISA control cards in
`the PC. The gantry angle is monitored via the accelerator’s
`analog monitoring channel; the voltage reading is digitized
`
`Page 3 of 17
`
`

`
`776
`
`I. J. Radiation Oncology c Biology c Physics
`
`Volume 45, Number 3, 1999
`
`Fig. 2. Schematic of the imaging systems used to acquire the MV and kV radiographs. Both systems are based on a
`cooled CCD camera for collecting light emitted by a 165 mg/cm2 Gd2O2S:Tb screen. The kV and MV systems are
`identical with the exception of an additional mirror in the MV assembly and the use of different metal plates. The kV
`assembly is a newer generation of the MV design. The housing for the MV assembly has been altered to permit
`retraction of the screen and large mirror allowing the couch to lower to the floor completely.
`
`to 12 bits by an analog-to-digital board in the control PC.
`The gantry angle can be determined to within 60.1 degrees
`with this mechanism.
`
`System performance
`Mechanical characteristics of the system. The DBS was
`constructed to measure the patient position with respect to
`the reference frame of the medical linear accelerator. The
`localization accuracy and precision that can be achieved
`with this system will depend on the mechanical stability or
`
`Fig. 3. The kV and MV imagers are controlled via a PC. This
`figure shows the mechanisms used to control and monitor the
`different components of the system. The gantry angle of the
`accelerator is monitored via an analog voltage line provided
`through the accelerator’s interface. The images acquired by the
`system are stored in a distributed database. The same system is
`used to acquire the many projections used for the conebeam
`reconstructions.
`
`rigidity of the system’s components. To estimate the preci-
`sion with which high-contrast objects can be located, mea-
`surements were made of the range of flex for each of the
`four components of the DBS as the gantry was rotated
`through 360°. These four components include: the kV x-ray
`source, the MV x-ray source, the MV imager assembly, and
`kV imager assembly.
`The magnitude of accelerator drum wobble and eccen-
`tricity was determined by directing a small (,1 mm diam-
`eter) laser beam onto the face of the drum. The gantry was
`rotated through 360° and the path of the laser spot on the
`drum was plotted. The laser was then adjusted to the geo-
`metric centroid of the path and the process was repeated.
`After a few iterations, the maximum extent of the path was
`determined to be the magnitude of eccentricity at the face of
`the drum. The measurements were repeated with a rigid
`mechanical extension securely fastened to the face of the
`drum. The extension moved the plane of measurement 1 m
`out from the face of the drum (isocenter lies 1.25 m out from
`the face of the drum). This experiment, in conjunction with
`the results from the face of the gantry, allowed us to
`approximate the magnitude of accelerator (drum) wobble at
`the plane of the x-ray source orbit (isocenter).
`The run-out of the gantry (motion of the entire drum
`assembly along the axis of rotation) was also measured.
`This was done with the use of a dial gauge attached to the
`front face of the accelerator (near the axis of rotation). The
`tip of the dial gauge was placed in contact with the end of
`the accelerator’s couch. The tip of the gauge was positioned
`on the gantry’s axis of rotation through an iterative proce-
`dure similar to that described in the previous paragraph. The
`gantry was then rotated through 360° and the relative move-
`ment between the end of the couch and the drum face was
`
`Page 4 of 17
`
`

`
`Imaging systems integrated into a linear accelerator c D. A. JAFFRAY et al.
`
`777
`
`Table 1. Mechanical characteristics of the DBS components
`
`Table 2. Stages in imaging system and their corresponding gains
`
`Component
`
`Range
`(mm)
`
`Tangential
`Axial
`Radial
`Drum
`
`MV
`source
`
`MV
`imager
`
`kV
`source
`
`kV
`imager
`
`1.5
`1.1
`2.0
`0.6
`
`4.0
`1.1
`5.0
`0.6
`
`4.0
`1.1
`5.0
`0.6
`
`1.5
`1.1
`1.5
`0.6
`
`The measured range of motion of each of the system’s four
`components. To establish a conservative estimate for localizing
`precision, it is assumed that each of these motions are completely
`independent. The ability to correct for these motions depends on
`their reproducibility as a function of gantry angle; preliminary
`investigations suggest that such corrections could be applied reli-
`ably.
`
`recorded. Care was taken to verify that the couch was rigid
`and immobile as the gantry rotated.
`To estimate the range of flex of the two x-ray source
`assemblies and the two imager assemblies, a solid state laser
`was mounted rigidly on the face of the drum and its beam
`(after passing through a cylindrical lens) was directed onto
`each of the four system components in-turn. As the gantry
`was rotated, the intersection of the line projected by the
`laser with the component was recorded on a small strip of
`paper attached securely to the surface of the component.
`The range of relative movement between the projected laser
`beam and the structure was taken to be due to flex in the
`component. The laser (and projected line) was then rotated
`and the measurement was repeated in the orthogonal dimen-
`sion (Table 1). It is important to note that the results of this
`measurement present a combination of the component’s flex
`and any shearing of the assembly. In addition, they do not
`determine the path of the component but provide a reason-
`able estimate for the range of the non-rigid motion. A more
`complete characterization of component flex is on-going, in
`which the movements of each component are characterized
`in three separate dimensions (radial, longitudinal, and an-
`gular) as a function of the nominal gantry angle (18).
`With these estimates of component flex, the geometric
`precision of the DBS for three-dimensional localization was
`estimated assuming an orthogonal-pair imaging application,
`which uses either the MV or kV system. The calculations
`were performed assuming the three-dimensional point of
`interest is near machine isocenter. Using the uncertainties in
`the viewing geometry introduced by the mechanical non-
`rigidity of the system,
`it was possible to calculate the
`localizing precision using the following method. At a given
`gantry angle, the range of angular motion of the source (kV
`or MV) and the range of angular motion of its correspond-
`ing detector assembly define a truncated sector in the iso-
`center plane. In an orthogonal view, the same ranges of
`motion define a second truncated sector in the isocenter
`plane. The extent of the area represented by the intersection
`of these two truncated sectors is taken as a measure of the
`localizing precision. Small ranges of radial motion (toward
`
`Stage in
`imaging
`system
`
`Gain
`(quanta exiting
`stage/quanta
`entering stage)
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`g1 (interacting X ray/
`incident X ray)
`g2 (optical quanta
`produced/interacting
`X ray)
`g3 (fraction of optical
`quanta escaping
`screen)
`g4 (fraction of optical
`quanta reflected by
`mirrors while en
`route to lens)
`g5 (fraction of optical
`quanta collected by
`lens)
`g6 (quantum efficiency
`of CCD)
`
`Detector and X ray energy
`
`Megavoltage
`imager
`(6 MV)
`
`Kilovoltage
`imager
`(90 kVp)
`
`0.025
`
`0.67
`
`1.96 3 104
`
`3.05 3 103
`
`0.7
`
`0.7
`
`0.83
`
`0.88
`
`1.36 3 10-4
`
`1.42 3 10-4
`
`0.78
`
`0.78
`
`Listing of the gain stages in the imaging system. Scattering
`stages have not been included in this analysis. The values used for
`each stage were determined directly or taken from the literature.
`
`or away from isocenter) of the components have negligible
`effects on the shape and location of the intersection defined
`here, and, therefore, have been ignored in this analysis.
`These calculations are performed to indicate the worst-case
`condition in which no corrections for mechanical flex have
`been applied.
`
`Radiographic imager performance: kilovoltage and
`megavoltage
`The kV and MV imaging systems operate on the same
`principle: the interaction of an x-ray within the phosphor
`screen (or metal plate/phosphor screen) produces light,
`which is collected by a high-speed lens and focussed onto
`an efficient, low-noise CCD. Each stage in the two systems
`has been characterized to examine the transfer of signal
`through the imaging chain. Table 2 lists the six stages
`included in the analysis. The gain or interaction probability
`for each stage has been estimated or taken from the litera-
`ture. From these values, it can be determined whether the
`two systems are operating near their theoretical limit of
`performance, as dictated by the noise associated with x-ray
`quanta used to form the image. This analysis is restricted to
`low-spatial frequencies and does not consider the small
`amount of additive noise introduced by the CCD camera.
`The gain estimates for the first stage (g1) were determined
`using Monte Carlo techniques (19) for the MV detector at 6
`MV (20) and using a spectral model (90 kVp) for the
`kilovoltage detector (16). The conversion of x-ray energy to
`light was determined from the Monte Carlo generated (MV)
`or analytical (kV) estimates of energy deposition in the
`phosphor layer. For the kV detector, K-fluorescence and
`re-absorption was approximated using the results of Chan et
`
`Page 5 of 17
`
`

`
`778
`
`I. J. Radiation Oncology c Biology c Physics
`
`Volume 45, Number 3, 1999
`
`Table 3. Measured and calculated gains for kV and MV imaging
`systems
`
`Detector and X-ray Energy
`
`Kilovoltage
`imager with
`90 kVp spectrum
`
`Megavoltage
`imager with
`6 MV spectrum
`
`0.125
`
`0.139
`
`1.11
`0.19
`
`0.036
`
`0.030
`
`0.83
`1.43
`
`Gain
`
`Measured (e2 detected in
`CCD/incident X ray)
`Calculated e2 detected in
`CCD/incident X ray)
`Calculated/measured
`Measured* (e2 detected in
`CCD/interacting X ray)
`
`The total gain of the two imaging systems for 6 MV and 90 kVp
`x-ray beams. The total gain of the system has also been calculated
`and comparison is made to measurement. Excellent agreement
`between the measured and calculated values was found.
`* Note: This estimate is based on the measured gain and the
`estimates of g1 presented in Table 2.
`
`al. (21) The number of optical quanta was calculated as-
`suming a conversion efficiency of 15% for the Gd2O2S:Tb
`screen and an average optical photon energy of 2.3 eV (550
`nm, green light, peak in emission spectrum). The probabil-
`ity of light escape from the phosphor was taken to be 0.7;
`this is the average of the results reported by Bissonnette et
`al. (22) and Ginzburg and Dick (23) for a Lanex Fast (back)
`screen (133 mg/cm2). The fraction of light reaching the
`CCD was determined by the reflectance of the front-surface
`mirrors (BV-2, Optical Coating Laboratories Inc., Santa
`Rosa, CA, USA) (g4) and the collection efficiency of the
`lens (g5) (8). The quantum absorption efficiency of the CCD
`at 550 nm was measured to be 0.78 (Calibration performed
`by Photometrics Inc., Tuscon, AZ, USA). The electronic
`gain of the camera electronics was measured to be 75.7 e2
`and 82.0 e2 per analog-to-digital converter unit (ADCU) at
`a gain setting of 1X for the MV imager and kV imager,
`respectively. From these gains, the quantum accounting
`diagrams (QADs) for the two systems are generated (24).
`These diagrams indicate the relative magnitude of quantum
`noise introduced at each stage in the system. Ideally, the
`great majority of the quantum noise should be introduced by
`the x-ray fluence itself. If this is the case, the system is
`considered “x-ray quantum-noise limited.”
`To verify that the estimates for gain are correct, the total
`gain of each system (electrons/incident x-ray) was measured
`under controlled exposure conditions. The fluence was es-
`timated using Rogers’ (25) dose-to-fluence conversion ta-
`bles (for 6 MV) and the Tucker and Barnes (16) model for
`simulating kV spectra (for 90 kVp). The total gains for both
`systems are plotted at the final stage in the QAD and are
`included in Table 3.
`
`Contrast-detail performance
`The contrast-detail (C-D) performance of both the kV and
`the MV radiographic imaging systems were measured (26).
`
`Transmission images of an aluminum C-D phantom (1.29
`cm thick) in 27 cm of water (lucite container) were acquired
`at a fixed imaging geometry (87 cm SSD, 160 cm SDD, 123
`cm from source to Al plate). The 27 cm water bath was
`included to simulate the scatter and absorption in a clinical
`imaging condition. Two MV images were acquired with 10
`monitor units at both 6 MV and 18 MV x-ray energies.
`Single kV images were acquired at 90 kVp using 10, 20, 30,
`40, 60, and 120 mAs exposures. The maximum dose for the
`5 1.5 cm) for
`megavoltage exposures were 13.6 cGy (dmax
`5 3 cm) for 18 MV. The
`6 MV and 13.7 cGy (dmax
`maximum dose for the kilovoltage exposures (correspond-
`ing to the entrance dose) determined from in-air measure-
`ments was 0.097, 0.195, 0.29, 0.39, 0.58, and 1.17 cGy for
`the six exposure levels. The in-air exposures were converted
`to dose assuming an approximate value of 0.9 for the
`exposure-to-dose conversion factor (fmed) and a measured
`back-scatter factor of 1.42 for 90 kVp (16 cm 3 18 cm field
`size at isocenter). Six observers were asked to locate the
`minimum observable diameter for each contrast level (depth
`of water-filled aluminum well). Each observer was allowed
`to vary the ambient room light level and was allowed to
`vary the display contrast to their liking. The display mag-
`nification was fixed. The average minimum detectable hole
`diameter was calculated across all six observers. At low
`contrasts and low doses, some observers could not detect a
`minimum hole diameter (i.e., no hole was large enough); if
`2 or more of the 6 observers were unable to detect anything
`at a given contrast level, that contrast level was removed
`from the analysis.
`For the 6 and 18 MV beams, the subject contrasts were
`determined from measured attenuation coefficients for wa-
`ter, aluminum, and lucite. For the 90 kVp images, the model
`of the x-ray spectrum was used to calculate the subject
`contrasts for each well in the phantom. The subject contrasts
`are calculated using the ‘difference over the mean’ tech-
`nique (27).
`The contrast-detail performance of the megavoltage and
`kilovoltage systems were then compared to demonstrate the
`dosimetric advantage of kV radiography over MV radiog-
`raphy in detecting bony structures. The contrast-detail
`curves for the MV and kV systems were compared on a
`common scale: depth of water-filled aluminum well. The
`MV and kV doses that produced equivalent object detection
`were determined for both 6 and 18 MV using intermediate
`contrasts and hole diameters. The use of the aluminum
`phantom (as opposed to bone) for this measurement is
`reasonable; the ratio of linear attenuation coefficients for Al
`to bone is constant within 610% over x-ray energies rang-
`ing from 10 keV to 20 MeV (28). From this comparison, the
`increase in dose (at depth of maximum dose) required at 6
`and 18 MV to achieve the same contrast visibility of a
`bone-like structure at 90 kVp was determined. It should be
`kept in mind that, in this study, three factors have a signif-
`icant influence on visibility: (i) the change in subject con-
`trast of the Al (nominally bone-equivalent) phantom with
`x-ray energy; (ii) the change in transmitted x-ray fluence;
`
`Page 6 of 17
`
`

`
`Imaging systems integrated into a linear accelerator c D. A. JAFFRAY et al.
`
`779
`
`and (iii) the difference in detector performance between the
`kV and MV imaging systems. Strictly speaking, the dosi-
`metric advantage determined here only applies to these two
`imaging systems and this specific imaging task.
`The clinical operation of the two imaging systems (mega-
`voltage and kilovoltage) was demonstrated on patients re-
`ceiving radiation therapy. The number of monitor units used
`to acquire each MV image is reported with each image (see
`captions). The exposure (in-air) for each kV radiographic
`exposure was measured using a PMX-3 electrometer (RTI
`Electronics AB, Molndal, Sweden) and an energy compen-
`sated radiation diode (R-25). (The dose is reported in the
`figure caption for each image.) For some fields, both kV and
`MV images were acquired. These have been fused to dem-
`onstrate a potential use of the DBS for producing kV open-
`field double exposures.
`Conebeam tomographic imaging. In addition to radio-
`graphic localization, the DBS has been designed to allow
`conebeam tomographic localization of soft-tissue structures
`in the reference frame of the accelerator. The high-fre-
`quency x-ray generator has a high duty cycle (20%) that
`allows multiple (;200) radiographs to be acquired in rapid
`succession (1 per second). To generate a CB-CT dataset, a
`series of radiographic exposures are acquired at regular
`angular intervals as the accelerator gantry is rotated through
`a specified range (typically 180° or 360°). The control
`system operates the camera’s shutter and read-out mecha-
`nisms in synchrony with the firing of the x-ray generator;
`these operations are slaved to the accelerator’s gantry angle.
`The accelerator gantry rotates continuously at a relatively
`low speed (;30°/min) with the control system monitoring
`the gantry angle. When the gantry angle is within a pre-set
`tolerance (60.1°) of the desired angle, the camera shutter is
`opened, the generator delivers a short exposure (50 ms), and
`the image is read out and stored to disk. Minor variations in
`gantry speed do not affect the acquisition, and the short
`exposure period (50 ms) introduces negligible blurring in
`the radiographs at these gantry speeds.
`The energy of the x-ray spectrum used for the conebeam
`acquisitions is typically 120 kVp and each image is pro-
`duced with approximately 3 mAs of charge. A 180 projec-
`tion scan would deliver a total charge of 540 mAs to the
`anode. In addition to the projection images, 20 bright-field
`images and 20 dark-field images are taken following acqui-
`sition. Each projection image is corrected by subtracting the
`dark-field signal (light leaks, digital offset, and dark-current
`in the CCD) and by dividing out the non-uniformities in the
`bright-field. No corrections for uncertainties in the scanning
`geometry due to flex or warping were applied. No scatter
`compensation algorithms were used in the reconstruction of
`these images.
`The 180 images are transferred via network to a work-
`station for reconstruction. A volumetric CB-CT dataset is
`reconstructed on a 70 MHz SparcStation 20 using a filtered
`back-projection algorithm based on that first described by
`Feldkamp et al. (29,30). The filtering process takes app