Deformable Scintillation Dosimeter I: Challenges and Implementation using Computer Vision Techniques
DDeformable Scintillation Dosimeter I: Challengesand Implementation using Computer VisionTechniques
E Cloutier , , L Archambault , , L Beaulieu , ‡ Physics, physical engineering and optics department and CancerResearch Center, Universite Laval, Quebec, Canada. CHU de Quebec – Universit´e Laval, CHU de Quebec, Quebec,Canada
Abstract.
Plastic scintillation detectors are increasingly used to measure dosedistributions in the context of radiotherapy treatments. Their water-equivalence, real-time response and high spatial resolution distinguish them from traditional detectors,especially in complex irradiation geometries. Their range of applications couldbe further extended by embedding scintillators in a deformable matrix mimickinganatomical changes. In this work, we characterized signal variations arising from thetranslation and rotation of scintillating fibers with respect to a camera. Correctionsare proposed using stereo vision techniques and two sCMOS complementing a CCDcamera. The study was extended to the case of a prototype real-time deformabledosimeter comprising an array of 19 scintillating. The signal to angle relationshipfollows a gaussian distribution (FWHM = 52 ◦ ) whereas the intensity variation fromradial displacement follows the inverse square law. Tracking the position and angle ofthe fibers enabled the correction of these spatial dependencies. The detecting systemprovides an accuracy and precision of respectively 0.008 cm and 0.03 cm on the positiondetection. This resulted in an uncertainty of 2 ◦ on the angle measurement. Displacingthe dosimeter by ± ±
10% (mean ± standard deviation) to the reference position. Applying corrections reduced thevariations thus resulting in relative intensities of 100 ± ± ±
3% to 100 ±
1% after the correction.Therefore, accurate correction of the signal collected by a camera imaging the outputof scintillating elements in a 3D volume is possible. This work paves the way to thedevelopment of real-time scintillator-based deformable dosimeters. ‡ Present address: Department of Physics, Universit´e Laval, Quebec, QC BS8 1TS, Canada. a r X i v : . [ phy s i c s . m e d - ph ] J a n eformable Scintillation Dosimeter I : Challenges and correction techniques
1. Introduction
Over the last decade, water-equivalent radio-luminescent materials have been used in avariety of setups to quantify delivered dose distributions of radiotherapy treatments.From plastic scintillating fiber detectors to volumetric scintillation dosimeters andCherenkov imaging, such systems enable real-time measurements with high spatialresolution over a wide range of energies [1–3], without the need for energy-dependentcorrection factors. Moreover, the advent of complex personalized treatment plans using agreater number of small fields, more modulated beams and magnetic fields [4–6] highlightthe advantages of plastic scintillation detectors making them well suited tools for therising challenges of advanced radiation therapy techniques [7].Over the same period, a growing clinical interest to consider anatomical variationsin treatment planning and delivery has developed. Inter-fractional and intra-fractionalorgan motion, as well as anatomical deformations, have been shown to result in clinicallysignificant dose variations that need to be accounted for [8, 9]. This adds another layerof complexity for dose measurements. Therefore, there is an increasing need for newdosimeters capable of measuring dose in a deformable matrix mimicking anatomicalvariations [10]. Scintillators have been used for 3D dosimetry and may be an ideal choicefor measurement in the presence of deformations. Volumetric scintillation dosimetershave demonstrated the ability to perform millimeter resolution, real-time and water-equivalent dosimetry of dynamic treatment plan over 2D [11] and 3D volumes [12–17].A scintillator-based deformable dosimeter would be suited to the challenges imposed byboth motion management and advanced radiotherapy modalities. Furthermore, giventhe rapidly increasing role of artificial intelligence [18] in radiation oncology, the needfor accurate experimental validation will likely increase in the future. However, goingfrom static to deformable geometries entails new difficulties. Applying a deformationto a radioluminescent-based phantom will lead to translations and rotations of theradioluminescent elements resulting in variations of the signal collected, even if nochange in deposited dose is expected.This work is the first to investigate the signal variations arising from thedisplacement and rotations of a point-like scintillator directly imaged by a camera (i.e.not coupled to a clear optical fiber). Using computer vision techniques, the positionof the tip of a scintillating fiber and it’s angulation in regards to the photo-detectoris tracked, and signal variations are corrected. Then, those correction techniques areapplied to the case of a deformable phantom comprising an array of 19 scintillatingfibers measuring the dose from a linac. The dosimeter and correction method weresubsequently applied to the simultaneous deformation vector fields and dose distributionmeasurements, which is presented in the companion paper [19]. eformable Scintillation Dosimeter I : Challenges and correction techniques
Signal caracterization(section 2.1)
Angular correction sCMOS mountedon robot M One scintillatorat isocenter FDistal correction
Signal correction(section 2.2)
3D positionning accuracy sCMOS1 + CCD F One scintillatormounted on robot MAngular measurement sCMOS2 + CCD
Correction validation(section 2.3)
Single scintillator sCMOS mountedon robot M One scintillatorat isocenter F19 scintillators dosimeter sCMOS1 + sCMOS2+ CCD F Deformabledosimeter M
2. Methods
Measurements were conducted with different detection setups which are summarized intable 1. Throughout this work, green scintillators (length: 1.2 cm, diameter: 0.1 cm,BCF-60; Saint-Gobain Crystals, Hiram, OH, USA) are used. All irradiations wereperformed with a 6 MV photon beam (Clinac iX, Varian, Palo Alto, USA).
Signal variations caused by the displacement and rotation of scintillating fibers wereseparately characterized using a sCMOS camera (Quantalux, Thorlabs, Newton, USA)mounted on a Meca500 small industrial robot arm (Mecademic, Montreal, Canada)(figure 1). From different viewpoints, the sCMOS acquired the scintillating signalfrom the tip of a scintillator positioned at the isocenter of a 6 MV photon beam.All measurements were compared to the signal obtained at a reference position setto (r, θ , φ ) = (35, 0, 0). The relation between the collected light and the orientation ofthe camera with respect to the scintillator was characterized by moving the camera withthe robot around the scintillator, within the robot’s limits ( ± ◦ ), keeping a constantradial distance (r = 35 cm, θ , φ = 0). Then, the signal to radial distance ( r ) relationshipwas measured by moving the camera towards to scintillating fiber, from 30 to 43 cm,keeping the orientation fixed (r, θ = 0, φ = 0). Acquisitions from a uniform white emitter eformable Scintillation Dosimeter I : Challenges and correction techniques r Scintillating fiber (0,0,0) sCMOS (r, θ , φ )Lateral displacement(a) (b) ✓ Figure 1: Setup used for the θ and r calibration and validation using lateraldisplacements : (a) pictures the camera mounted on the robot while (b) presents thecoordinate system.screen were also performed to quantify the impact of vignetting in the resulting images.The vignetting for each pixel (i, j) was calculated using a cos ( θ ( i,j ) ) fit as suggested byRobertson et al [20]. A setup of 3 cameras was designed to measure the signal, orientation and 3D position ofirradiated scintillation fibers (figure 2). The setup comprises two sCMOS and one cooledCCD (Alta U2000, Andor Technology, Belfast, United Kingdom). The CCD camerawas chosen for its capacity to provide stable measurements, whereas the sCMOS wereselected for their high spatial resolution (1920 x 1080 pixels). The resulting detectionassembly aims at correcting the signal from moving scintillators measured with staticcameras.
To account for the rotation of a scintillating fibers, asCMOS camera was positioned in front of a CCD. Angles were calculated from themeasured vertical ( dy i ) and lateral ( dx i ) displacement shifts by the facing cameras :sin θ m = dx + dx l ; sin φ m = dy − dy l . (1)The accuracy of tilt measurements was assessed by mounting a 1.2 cm length (0.1 cmdiameter) scintillating fiber on the robot arm. θ m and φ m were simultaneously measuredwhile rotating the fiber with the robot in the θ r and φ r direction from 0 ◦ to 30 ◦ . Distance corrections rely on the 3D distance( r = x + y + z ) of the fiber in the object space with respect to the cam-era’s sensor center. Hence, a stereoscopic pair of camera was used to project the 2Dimage position of each scintillating fiber onto the 3D object space and correct varia-tions resulting from changes in their optical coupling with the cameras. Using computer eformable Scintillation Dosimeter I : Challenges and correction techniques (cid:101) P ( x, y, z ) on a 2D image plane (cid:101) p ( x (cid:48) , y (cid:48) ) using a projective transformation as: (cid:101) p = s (cid:102) m = K × [ R t − R t t ] × (cid:101) P . (2) K and [ R t − R t t ] respectively refer to the intrinsic and extrinsic parameters of the camera,which can be extracted from calibration [21]. The intrinsic parameters matrix dependson the properties of the detector used whereas the extrinsic parameters matrix dependson the position (rotation, translation) of the detectors with regards to the imaged scene.Once known, it is possible to reconstruct the (x,y) position of an image point in theobject space. However, using only one camera limits the projection to (x,y) coordinatesas the z position (depth) is degenerate. The use of an additional camera imaging theobject from a different perspective removes the degeneracy along z and enables the 3Dpositioning of the object.In this work, we paired the Alta U2000 cooled CCD to a sCMOS to locate the tipscintillating fibers in the object space ( x, y, z ). With this location, it was possible toapply the distance correction to signal variations arising from the movement of the fibers.Cameras were calibrated using a (15 ×
10) chessboard pattern and a calibration algorithminspired by Zhang from the OpenCV Python library version 3.4.2 [22, 23]. Imageswere rectified and corrected for distortion before performing the triangulation. Therectification eases triangulation calculations whereas the distortion correction increasesits accuracy. The position of the left camera in relation to the first one, as obtained fromcalibration, is presented in table 2. The accuracy of the 3D tracking from stereo-visionwas assessed by mounting a scintillating fiber on the robot arm. Displacements in thex, y and z axis were subsequently performed by the robot in increments of 1 cm. eformable Scintillation Dosimeter I : Challenges and correction techniques and the CCD coordinate systemsas obtained from the calibrationTranslation [cm] X : 10.67 Y : -0.33 Z : 3.45Angle [ ◦ ] Pitch : -1.3 Yaw : -15.62 Roll : -0.26 The signal resulting from the lateral displacement was acquired to validate the proposedcorrection technique. The case of a single scintillating fiber was first assessed bymounting a sCMOS on the robot imaging a fixed scintillating fiber. Measurement weretaken from -7 to 7 cm in increments of 1 mm and a correction was performed usingknown distance ( r ) and orientation ( θ , φ ). The method was extended to the case of a deformable scintillator-based dosimetercomprising as array of 19 BCF-60 scintillating fibers (figure 2) and a complete correctionwas carried out without prior knowledge on the distance and orientation of the fibers.
The deformable dosimeter prototype consists of a clear,flexible cylindrical elastomer in which 19 scintillating fibers were embedded (figure 2).The cylinder is made from a commercial urethane liquid elastomer compound (Smooth-On, Macongie, USA) cast in a silicone cylindrical mold (diameter: 6 cm, thickness:1.2 cm). The compound was degassed in vacuum prior to pouring in order to removetrapped air bubbles which would have reduced the final transparency of the elastomer.Nineteen scintillations fibers were inserted in the cylindrical elastomer guided by a3D-printed template. Each scintillating fiber was covered by a heat-shrinking opaquecladding to isolate the scintillation light from its surrounding and, more importantly,limit the collected signal to the one emerging from its ends. The scintillating fibers wereembedded in the phantom forming a 1x1x1 cm triangular grid array.
The density (in g/cm ) of the detector was extractedfrom a CT-scan (Siemens Somatom Definition AS Open 64, Siemens Healthcare,Forchheim, Germany). CT-scans of the bulk elastomer (i.e. no fibers embedded) and areference water volume were also acquired for comparison. The pitch, current and energyof the scanner were respectively set to 0.35, 60 mA and 120 kVp. The detector was alsoirradiated with a 6 MV, 600 cGy/min photon beam (Clinac iX, Varian, Palo Alto, USA)while being imaged. The center of the detector was aligned with the isocenter of thelinac. Dose linearity was studied while varying the dose deposited or the dose rate.Different dose rates were achieved by varying the distance between the detector and eformable Scintillation Dosimeter I : Challenges and correction techniques The dosimeter was displaced laterally from -3to 3 cm relative to it’s initial position relative to the camera. Radial displacementwere also conducted moving the dosimeter from 32 to 38 cm from the CCD camera.Displacements were achieved by translating the treatment couch in 1 cm incrementsand repositionning the center of the dosimeter at the isocenter of the linac, to keep thedose constant. The irradiations were of 100 monitor units (MU). The radiometry, i.e.quantitative measurement of scintillating signal related to the dose, was carried by theCCD camera, while both sCMOS measured the angle and 3D position of the fibers. TheCCD camera was positioned 35 cm from the dosimeter and coupled to a 12 mm focallength lens (F/ ×
3. Results
20 0 20Rotation [°]0.850.900.951.00 I n t e n s i t y [ - ] MeasuredGaussian fit (FWHM = 52.0°, r =0.97) (a)
30 35 40Distance [cm]0.40.60.81.0 I n t e n s i t y [ - ] MeasuredFit : 1/ r ( r = 0.99) (b) Y P i x e l [ - ] (c) Figure 3: Characterization of the angular θ (a) and radial r (b) dependencies of thesignal collected by the camera. Vignetting (c) was also characterized to correct signalvariations of the sensor. Rotating the camera’s optical axis with respect to the scintillating fibers axis resultsin a decrease of the collected signal. This decrease can be modeled according to agaussian distribution with a full width at half max (FWHM) of 52 ◦ (see figure 3a).For comparison, the scintillating fiber has a numerical aperture of 0.583 which results eformable Scintillation Dosimeter I : Challenges and correction techniques ◦ in air. Figure 3b presents the distance to signalrelationship obtained while varying the distance between the camera and the scintillatingfiber from 30 to 43 cm. Increasing the distance results in a decrease of the collectedsignal following the inverse square law ( R > . Figure 4 presents θ m and φ m measured while moving a scintillating fiber in the θ (a)and φ (b) direction. Rotating the scintillating fiber from 0 to 30 ◦ resulted in differencesup to 2.3 ◦ between the measured and predicted tilts. M e a s u r e d [ ° ] mm D i ff . [ ° ] (a) M e a s u r e d [ ° ] mm D i ff . [ ° ] (b) Figure 4: θ m and φ m measured while rotating a scintillating fiber either in the θ (a) or φ (b) axis. Figure 5 presents the measured displacement by the stereoscopic pair in the x, y andz axis while moving the fibers in 1 cm increments in each directions, successively.Throughout all the measurements, the system provided a mean accuracy and precisionof 0.008 cm and 0.03 cm respectively.
Figure 6 presents the signal variations measured from a lateral displacement of thecamera imaging one scintillating fiber and the resulting signal after angular, radial andvignetting corrections are applied. A gaussian fit was applied to raw data to accountfor uncertainties arising from wobbling movements of the camera through the robot’sdisplacement. Figure 6(b) shows the corrected signal and the contribution of vignetting, eformable Scintillation Dosimeter I : Challenges and correction techniques M e a s . d i s p l a c e m e n t [ c m ] xyz E rr o r [ c m ] (a) M e a s . d i s p l a c e m e n t [ c m ] xyz E rr o r [ c m ] (b) M e a s . d i s p l a c e m e n t [ c m ] xyz
34 35 36 37 38 39Position [cm]0.050.000.05 E rr o r [ c m ] (c) Figure 5: 3D measured displacement, while moving the scintillating fiber in the X(a), Y (b) or (c) direction. The errors represent the difference between the expecteddisplacement and the one measured with the stereo pair of cameras.angle and distance to the magnitude of the correction. The combined correction resultedin signal variations lesser than 0.5% for lateral displacements ranging from -7 to 7 cm. I n t e n s i t y [ - ] MeasuredGaussian fit (FWHM=12.3 cm, r =0.90) (a) I n t e n s i t y [ - ] Gaussian fit (FWHM=12.3 cm)Corr:VignettingCorr:Vignetting+AngleCorr:Vignetting+Angle+Distance (b)
Figure 6: Signal variation arising from lateral displacement of the camera (a) and itscorrection using angular, radial and vignetting corrections (b). Horizontal lines in (b)represents a ± Evaluation of the voxel density values from CT-scans yielded (mean ± standard deviation) densities of 1.002 ± ± ± respectively for water, the urethane elastomer, and the elastomerwith the scintillating fibers inside. Figure 7 further presents a slice acquired from the CTand a profile drawn across a region of interest. Even if the region of interest interceptsfour scintillating fibers, those are indistinguishable from the bulk elastomer.As could be expected, the detector exhibited a linear dose-light relationship (R > eformable Scintillation Dosimeter I : Challenges and correction techniques (a) H o un s f i e l d un i t [ - ] (b) Figure 7: Image of a CT slice of the dosimeter (a) and a Housfield Unit profile extractedfrom a region of interest (b). Dashed red lines on (a) correspond to the selected regionof interest. I n t e n s i t y [ - ] (a) I n t e n s i t y [ - ] (b) Figure 8: Linearity of the scintillation signal as function of the dose (a) and dose rate(b) for all of the 19 scintillating fibers. Solid lines represent linear fits : R > .
999 forall measurements.proportionality remained linear (R > The signal obtained for the 19 scintillating fibersas a result of varying the distance between the gel and the CCD camera from 32 to 38 cmis presented of figure 9. The raw signal varies from 84.7% to 117.9% of the one obtainedat 35 cm (used as reference). Applying the inverse square law using the 3D positioning ofthe fibers provided by the stereo matching cameras to the resulting signal reduced thosevariation to 97.4% up to 101.9%. Figures 9(c) et 9(d) present the distribution from allgathered data from the 19 scintillating fibers for each distance prior and after correctionsare applied. Radial distance variations resulted in mean ± standard deviation intensitiesof 100 ±
10 % and 100 ±
1% before and after corrections, respectively.Similarly, the raw signal variations caused by the lateral displacement of thedeformable dosimeter are presented in figure 10. Displacing the dosimeter from -3cm to 3 cm relative to it’s initial position caused signal variations between 88.3% and eformable Scintillation Dosimeter I : Challenges and correction techniques
32 34 36 38Distance [cm]0.80.91.01.11.2 I n t e n s i t y [ - ] (a)
32 34 36 38Distance [cm]0.80.91.01.11.2 I n t e n s i t y [ - ] (b) F r e q u e n c y [ - ] (c) F r e q u e n c y [ - ] (d) Figure 9: Measured signal variation resulting from moving the dosimeter from a distance32 to 38 cm in depth before (a) and after applying the signal corrections (b). (c) and(d) present the distribution of raw and corrected data, respectively.103.7%. Once corrected for angular and distal variations, signal variations ranged from95.8% to 104.2%. The signal drop observed at 3 cm on figure 10 (a) results from a smallangulation of the prototype after its re-positioning at the isocenter. The angulation wasdetected by the system and corrected as seen on 10 (b). The data distributions presentedof figures 10(c) et 10(d) reveal mean ± standard deviation intensities of 98% ±
3% and100% ±
1% before and after corrections, respectively. I n t e n s i t y [ - ] (a) I n t e n s i t y [ - ] (b) F r e q u e n c y [ - ] (c) F r e q u e n c y [ - ] (d) Figure 10: Measured signal variation resulting from lateral displacement before (a) andafter applying the signal corrections (b). (c) and (d) present the distribution of raw andcorrected data, respectively. eformable Scintillation Dosimeter I : Challenges and correction techniques
4. Discussion
The signal produced by a scintillating fiber that is measured with a camera depends onthe distance between the camera and the scintillating fiber as well as the angle betweentheir respective axes. The resulting decrease of the measured signal as function of the tiltof the fiber arises from the combined gaussian output of the guided scintillation signalin the fiber and the non-guided isotropic signal emitted at the tip of the fiber. Signalvariations related to the distance between the camera and the fibers follows the inversesquare-law, as previously demonstrated [25, 26]. As a consequence, if not corrected, adeformation of 1 cm in the z axis, captured by a camera distant of 35 cm, would leadto signal variations of 5.48%. Thus, tracking position and angulation of the fibers isessential for adequate dose measurements.This work proposes the use of computer vision techniques to track the position ofscintillating fibers. The 3D optical position tracking enabled a precision of 0.03 cm.This is slightly larger than the tolerance on the robot’s positioning of 0.01 cm andthe camera’s pixel resolution of 0.02 cm. The discrepancy on figure 5(a) happeningwhen the fiber passes the robot’s wrist center highlights the robot’s singularity point(i.e. a configuration where the robot is blocked in certain direction, thus modifyingits path). Overall, our prototype dosimeter constitutes an application well suited tostereo vision. Indeed, the accuracy of 3D reconstruction in stereo vision relies on 1)the feature detection, and 2) the feature matching. Solving the correspondence problemin the image pairs is one of the main challenges of stereo vision and many strategieshave been proposed to solve it [27]. Having 19 well-defined and organized points tomatch significantly eases that challenge. As a result, our uncertainties are limited tothe feature detection, i.e. centroids, accuracy. Keeping the dosimetry application inmind, an uncertainty of 0.03 cm would lead to a 0.17% dose uncertainty, at a distanceof 35 cm. As for the angle measurement, the uncertainty of 2 ◦ is consistent with thespatial resolution of the camera’s limited to 0.02 cm. Compromising the field of view,with a longer focal length objective for example, would improve the spatial and resultingangular resolution of the system. The system could also be improved with the additionof another sCMOS forming a stereoscopic pair with the facing camera (sCMOS1) tocompletely position the fibers on both sides [19].Correction functions (distance, angle and vignetting) were validated before theirapplication to the case of a deformable dosimeter. To do so, signal variations resultingfrom the motion of a sCMOS imaging an irradiated scintillating fiber, from -7 to 7 cm,were corrected using expected distances ( r ) and angles ( θ ). Thus, a signal decreasedown to 85% was reduced to 0.5%, after corrections. Using known r and θ enabled avalidation that minimized uncertainties related to the distance and angle measurement.Density of the deformable detector presented no significant difference with water,meaning the detector can simultaneously act as a water-equivalent detector andphantom. Overall, displacement of the dosimeter radially generated higher signalvariations than lateral displacement. However, radial signal variations were more eformable Scintillation Dosimeter I : Challenges and correction techniques ≈
10% as the field of view covered a patient’s whole body. Inthis work, the stereo matching provides an uncertainty margin of 0.03 cm, thus reducingthe potential dose uncertainty or even allowing the positioning of the scintillator closerto the cameras. Therefore, precise 3D positioning and the associated corrections canincrease the signal-to-noise ratio (SNR) of dose measurements using radioluminescentelements. Distance and angular dependencies corrections will be essential to accuratedose measurements when extending scintillator’s range of application to deformablecases: deforming a phantom comprising scintillators will lead to translations and
EFERENCES
5. Conclusion
The novel use of scintillating fibers in a varying geometry phantom presented newdifficulties that were characterized and corrected. Hence, measurements of the angularand distal variations of the fibers from the detector reduced the signal dependencieson the varying geometry of the gel. Pairing a cooled CCD to two sCMOS enabled the3D positionning and angular tracking of 19 moving scintillating fibers. All together,the setup enabled a correction workflow accounting for distal and angular variations ofmoving scintillating elements. Moreover, we prototyped a novel deformable scintillationdetector measuring the dose at 19 points in a flexible phantom. This works is a steptoward the use of plastic scintillators in moving and varying geometries.
6. Acknowledgement
The authors thank Serge Groleau for his help manufacturing the elastomer. Thiswork was financed by the Natural Sciences and Engineering Research Council ofCanada (NSERC) Discovery grants
References [1] Brian W. Pogue, Adam K. Glaser, Rongxiao Zhang, and David J. Gladstone.“Cherenkov radiation dosimetry in water tanks – video rate imaging, tomographyand IMRT & VMAT plan verification”. In:
Journal of Physics: Conference Series
EFERENCES
Physics in Medicine andBiology
Radiation Protection Dosimetry
Physics inMedicine & Biology
Radiation Measurements
116 (Sept. 2018), pp. 10–13.[7] A. S. Beddar and L Beaulieu.
Scintillation Dosimetry . CRC Press, Apr. 1, 2016.[8] B. Schaly, J. A. Kempe, G. S. Bauman, J. J. Battista, and J. Van Dyk. “Trackingthe dose distribution in radiation therapy by accounting for variable anatomy”.In:
Physics in Medicine and Biology
MedicalPhysics
Medicalphysics
Medical Physics
Applied Optics (), p. 11.[13] Madison Rilling, Guillaume Allain, Simon Thibault, and Louis Archambault.“Tomographic-based 3D scintillation dosimetry using a three-view plenopticimaging system”. en. In:
Medical Physics (May 2020), mp.14213.[14] Mathieu Guillot, Luc Gingras, Louis Archambault, Sam Beddar, and Luc Beaulieu.“Performance assessment of a 2D array of plastic scintillation detectors for IMRTquality assurance”. en. In:
Physics in Medicine and Biology
EFERENCES
Medical Physics
Physics in Medicine andBiology
Medical Physics
Radiotherapy and Oncology
Physics in Medicine & Biology
Multiple view geometry in computervision . OCLC: 171123855. 2004.[22] Z. Zhang. “A flexible new technique for camera calibration”. In:
IEEETransactions on Pattern Analysis and Machine Intelligence
Dr. Dobb’s Journal of Software Tools (2000).[24] Louis Archambault, Tina Marie Briere, and Sam Beddar. “Transient noisecharacterization and filtration in CCD cameras exposed to stray radiation froma medical linear accelerator”. In:
Medical Physics
Physics in Medicine & Biology
EFERENCES
MedicalPhysics
Proceedings ELMAR-2010 . 2010, pp. 57–64.[28] Irwin I Tendler, Petr Bruza, Mike Jermyn, Xu Cao, Benjamin B Williams, Lesley AJarvis, Brian W Pogue, and David J Gladstone. “Characterization of a non-contactimaging scintillator-based dosimetry system for total skin electron therapy”. en.In:
Physics in Medicine & Biology
Physics inMedicine and Biology
Medical Physics