The effect of respiratory motion on electronic portal imaging device dosimetry

Abstract There is an increasing need to develop methods for in vivo verification of the delivery of radiotherapy treatments. Electronic portal imaging devices (EPID's) have been demonstrated to be of use for this application. The basic principle is relatively straightforward, the EPID is used to measure a two‐dimensional (2D) planar exit or portal dose map behind the patient during the treatment delivery that can provide information on any errors in linear accelerator output or changes in the patient anatomy. In this paper we focused on the effect of intra‐fraction motion, particularly respiratory motion, on the measured 2D EPID dose–response. Measurements were made with a breast phantom undergoing one‐dimensional (1D) sinusoidal motion with a range of amplitudes (0.5, 1.0, and 1.5 cm) and frequencies (12, 15, and 20 cycles/min). Further measurements were made with the phantom undergoing breathing sequences measured during patient planning computed tomography simulation. We made use of the quadratic calibration method that converts the EPID images to a surrogate for dose, equivalent thickness of Plastic Water®. Comparisons were made of the 2D thickness maps derived for the different motions compared to the static phantom case and the resulting dose difference analyzed over the “breast” region of interest. A 2D gamma analysis within the same region of interest was performed of the motion images compared to static reference image. Comparisons were made of 1D thickness profiles for the moving and static phantom. The 1D and 2D analyses show the method to be sensitive to the smallest motion amplitude of 0.5 cm tested in the phantom measurements. The results using the phantom demonstrate the method to be a potentially useful tool for monitoring intra‐fraction motion during the delivery of patient radiotherapy treatments as well as more generally providing information on the effects of motion on EPID based in vivo dosimetric verification.

portal dose map behind the patient during the treatment delivery that can provide information on any errors in linear accelerator output or changes in the patient anatomy. In this paper we focused on the effect of intra-fraction motion, particularly respiratory motion, on the measured 2D EPID dose-response. Measurements were made with a breast phantom undergoing one-dimensional (1D) sinusoidal motion with a range of amplitudes (0.5, 1.0, and 1.5 cm) and frequencies (12, 15, and 20 cycles/min). Further measurements were made with the phantom undergoing breathing sequences measured during patient planning computed tomography simulation.
We made use of the quadratic calibration method that converts the EPID images to a surrogate for dose, equivalent thickness of Plastic Water ® . Comparisons were made of the 2D thickness maps derived for the different motions compared to the static phantom case and the resulting dose difference analyzed over the "breast" region of interest. A 2D gamma analysis within the same region of interest was performed of the motion images compared to static reference image. Comparisons were made of 1D thickness profiles for the moving and static phantom. The 1D and 2D analyses show the method to be sensitive to the smallest motion amplitude of 0.5 cm tested in the phantom measurements. The results using the phantom demonstrate the method to be a potentially useful tool for monitoring intra-fraction motion during the delivery of patient radiotherapy treatments as well as more generally providing information on the effects of motion on EPID based in vivo dosimetric verification.
Technological advances in the planning and delivery of radiation therapy has led to an increase in complexity of patient treatments. 1 This increase in complexity has led to an increased sensitivity to uncertainties and errors in the treatment process and a subsequent need for verification of the delivered treatment. 2 Electronic portal imaging devices (EPIDs) have been shown to be suitable for in vivo dosimetric verification of treatments. [2][3][4][5][6] The concept for using twodimensional (2D) flat panel EPIDs for pretreatment and treatment time verification is conceptually straightforward and can be performed in one of two ways. A 2D integrated measurement of the delivered dose is made during radiation delivery and then compared with a reference 2D dose map calculated (or measured) at the EPID plane. Alternatively, the 2D EPID dose measurement can be backprojected into a CT-based model of the patient and compared with a reference dose distribution in the same CT-based patient model. 7 The CT model can be derived from the treatment planning CT or for a more accurate prediction be based on a treatment time cone beam CT. EPIDs have been shown in a number of studies to be valuable for dosimetric verification of breast radiotherapy treatments. [8][9][10][11][12][13] As already stated although this appears to be conceptually straightforward, the implementation of a solution with high accuracy is nontrivial such that the technique is still not in widespread routine clinical use.
A technique for using EPID-based measurements of radiological thickness for verifying the delivery of radiotherapy treatments has been previously demonstrated. 3,14,15 This was an extension of the quadratic calibration method that has been used for designing compensators and IMRT fields for tangential breast radiotherapy treatment fields. [16][17][18] The technique was shown to be a reliable means of directly relating measured EPID images to a reference Monte-Carlo EPID simulation, using the radiological or poly(methyl)methacrylate equivalent thickness as a surrogate for dose. It was shown to be suitable for verifying treatment delivery and identifying changes in the treatment field, patient position, and target location as well as patient physical thickness.
Differences between delivered and planned dose distribution can have a number of sources including radiation delivery errors, changes in detector response, and changes in patient anatomy which would change the transmitted dose reaching the portal dosimeter. Potential radiation delivery errors and intra-and inter-fractional changes in the patient anatomy will be detrimental to the treatment outcome and are the reason in vivo treatment verification is highly desirable. 19,20 EPID dosimetry is considered to be the favored modality for in vivo dosimetry going into the future and it has been shown to be effective for detecting the various types of errors encountered in the radiotherapy treatment. [20][21][22] If EPID dosimetry is going to be more widely used in the clinical setting it is important that the sources of potential differences between measured and reference dose to the EPID are understood and characterized. One of these sources is intra-fraction respiratory motion that would be expected to cause a difference between the predicted and measured signal in the EPID. This paper investigates the effect of respiratory motion on measured EPID images converted to radiological thickness maps. An emphasis is placed on the larger motion amplitudes that the literature suggests can occur in patients undergoing thoracic radiotherapy. [23][24][25] The study aims to investigate how different magnitudes of regular and irregular patient motion sequences manifest in the EPID images when calibrated for radiological thickness and to determine if the method could be suitable for simple and efficient monitoring of intra-fraction patient motion and compliance with breathhold techniques. The work will also provide important information on the effects of motion on EPID dosimetry techniques.

2.A | Calibration
Two sets of measurements were performed using an Elekta Precise and an Elekta Agility linear accelerator in 6 MV photon mode operating at a nominal dose rate of 600 MU/min. The method described in this paper is independent of the accelerator and therefore when presenting and discussing this work the accelerator will not be speci- The quadratic calibration method, described in more detail in Ref. [3,26], was applied to the seven calibration transmission images, and a set of calibration coefficients, α(x,y) and β(x,y), was derived for each pixel in the detector. The calibration coefficients, obtained using a least squares polynomial fit, relate the intensity signal, I(x,y), measured in each detector pixel to equivalent Plastic Water ® thickness, t PW (x,y) through the expression, The quadratic term (t 2 ) in Equation (2) is introduced to account for spectral variations in the linear accelerator photon beam and beam hardening in the object which cause the signal recorded by a detector pixel (x,y) to deviate from a simple exponential function of object thickness. Equation (2) can then be inverted in order to convert subsequent images of any object to equivalent Plastic Water ® thickness maps, t PW (x,y). An iterative algorithm is used to incorporate corrections for differences in field size between the calibration field size and the treatment field size, and scatter. 3,[16][17][18] The first estimate of the thickness map t PW (x,y) is calculated using A new estimate for the image pixel intensity, corrected for field size and scatter, is then calculated using at the treatment field size to the calibration field size I calib (x, y).

SPR ref (x, y)
is the scatter-to-primary ratio for the calibration measurements and SPR treat (x, y) is the scatter-to-primary ratio for the breast phantom measurements (treatment field size). The first order approximation used for calculating the SPR is given by Ref. [27] SPRðx; yÞ ¼ k 0 A tðx; yÞ (5) where A is the field size (area) at the isocentre plane, and t(x, y) is the thickness of the object (calibration phantom or breast phantom) and k 0 is a parameter with a value of 1.93 × 10 −5 /cm 3 to account for system geometry and the Plastic Water ® calibration phantom electron density. 3,27 An updated t PW (x, y) (using Equation (3)] is then calculated using the new I i+1 (x, y) of Equation (4).This correction algorithm for calculating the thickness image t PW (x, y) was repeated and found to converge after five iterations.

2.B | Treatment fields
The treatment fields were planned using the Pinnacle treatment

2.C | Motion simulation
The anterior oblique wedged field was delivered to the CIRS thorax phantom (including a breast attachment) that was fixed on the CIRS programmable motion platform. Figure 1 shows a photograph of the experimental setup of the linear accelerator, EPID, phantom and motion platform. Measurements were performed with a static phantom and with the same phantom undergoing different motion sequences. The static phantom measurement was repeated three times to test reproducibility.

2.D | Image analysis
A threshold technique was first used to create a mask representing the radiation field in the images. The boundary pixels (field edge) of the mask were determined using an implementation of a "chaincode" algorithm. 28,29 EPID images of the breast phantom were converted to 2D thickness maps using the quadratic calibration technique described in Section 2.A. The "breast" outline in the static thickness image was manually contoured to create a region of interest (ROI) and the image pixels within the ROI determined using the F I G . 1. The experimental setup for simulating the effects of motion on breast radiotherapy. The direction of one-dimensional motion was along the longitudinal couch direction. A comparison of the thickness maps with the phantom undergoing motion to the thickness map obtained for the static case was performed. Comparison was performed in 2D and for 1D profiles taken through the 2D thickness maps. The 2D thickness maps were also converted back into Intensity maps I(x,y), using Equation (2), and α and β values of 0.05/cm and 1 × 10 −4 /cm respectively so that the dose difference, I motion (x, y) − I static (x, y), could be analyzed. Dose differences between the static and motion images were analyzed for individual pixels within the ROI. The spatial accuracy of dose delivery is also important such that small changes in high-dose gradient regions (or high thickness gradients) could lead to significant dose differences. Therefore, the dose differences in the ROI between static and moving images were also quantitatively assessed using a gamma analysis that combines the distance to agreement criteria with dosimetric difference (local dose difference was used in this analysis), for a range of different pass criteria including 5%/5 mm, 3%/3 mm, 2%/2 mm, and 1%/1 mm. 30

| DISCUSSION
In this paper we have investigated the effect of respiratory type motion on EPIDs calibrated for thickness using the quadratic calibration technique. The method used in this work was originally Further quantitative analysis of the dose differences, shown in Table 2 indicates differences becoming more significant as the amplitude is increased. The most significant motion, with an amplitude of 1.5 cm, results in 85% of the pixels within the "breast" ROI having a dose difference of 5% or less. For motion with an amplitude of 1 cm approximately 80% of the pixels within the same ROI have a dose difference of 2% or less. The agreement improves significantly for show the regions where dose differences are greater than 3%/3 mm. Table 3 shows that all six patient motion sequences would result in EPID data where less than 90% of the points within the ROI pass the 3%/3 mm gamma criterion, a benchmark commonly used when comparing dose distributions in a clinical setting. Patient 1 has the most significant differences which is evident from the thickness map in Fig. 7(a) and 2D gamma map in Fig. 7(d). There would appear to be a baseline shift in the phantom, which would probably be corrected through daily image guidance prior to treatment delivery.
However, this does show the method to be able to detect baseline shifts and motion that could occur during the treatment delivery and after pretreatment image guided setup verification.
The use of other motion tracking techniques such as external optical markers or surface monitoring 31,32 could also be used to provide further frequency information on the intra-fraction motion to complement the EPID-based equivalent thickness information.
We propose that this method will be of use for in vivo quantita-  would especially like to thank the staff of the PAH for technical assistance in obtaining the experimental measurements described in this paper.

CONFLI CTS OF INTEREST
The authors declare no conflict of interest.