Optimizing the target detectability of cone beam CT performed in image‐guided radiation therapy for patients of different body sizes

Abstract Purpose The target detectability of cone beam computed tomography (CBCT) performed in image‐guided radiation therapy (IGRT) was investigated to achieve sufficient image quality for patient positioning over a course of treatment session while maintaining radiation exposure from CBCT imaging as low as reasonably achievable (ALARA). Methods Body CBCT scans operated in half‐fan mode were acquired with three different protocols: CBCTlowD, CBCTmidD, and CBCThighD, which resulted in weighted CT dose index (CTDIw) of 0.36, 1.43, and 2.78 cGy, respectively. An electron density phantom that is 18 cm in diameter was covered by four layers of 2.5‐cm‐thick bolus to simulate patients of different body sizes. Multivariate analysis was used to examine the impact of body size, radiation exposure, and tissue type on the target detectability of CBCT imaging, quantified as contrast‐to‐noise ratio (CNR). Results CBCTmidD allows sufficient target detection of adipose, breast, muscle, liver in a background of water for normal‐weight adults with cross‐sectional diameter less than 28 cm, while CBCThighD is suitable for adult patients with larger body sizes or body mass index over 25 kg/m2. Once the cross‐sectional diameter of adult patients is larger than 35 cm, the CTDIw of CBCT scans should be higher than 2.78 cGy to achieve required CNR. As for pediatric and adolescent patients with cross‐sectional diameter less than 25 cm, CBCTlowD is able to produce images with sufficient target detection. Conclusion The target detectability of soft tissues in default CBCT scans may not be sufficient for overweight or obese adults. Contrary, pediatric and adolescent patients would receive unnecessarily high radiation exposure from default CBCT scans. Therefore, the selection of acquisition parameters for CBCT scans optimized according to patient body size was proposed to ensure sufficient image quality for daily patient positioning in radiation therapy while achieving the ALARA principle.


| INTRODUCTION
Approximately 50% of cancer patients can benefit from radiation therapy in the management of their diseases. [1][2][3] The accuracy of radiation therapy depends on the conformal deposition of ionizing radiation to the target volume and the efforts to spare its neighboring healthy tissues. 4,5 To achieve high-precision treatments, imaging plays a crucial role in planning and delivering radiation beams. 6,7 It has been demonstrated that the use of image-guided radiation therapy (IGRT) may improve the clinical outcome of patients undergoing radiation therapy. 8,9 In our department, patients are routinely scanned by multidetector computed tomography (MDCT) scanners for planning purposes before treatment. During daily treatment, cone beam CT (CBCT) mounted on the gantry of linear accelerator is used to detect target position relative to the planned radiation beams to improve the accuracy of treatment delivery through geometric corrections. Target detectability of CBCT is a very important image quality metric to achieve a high level of patient positioning and treatment accuracy. [10][11][12][13] In spite of the increasing use of CBCT to verify and correct patient setup, the contrast resolution of CBCT in delineating soft tissue structures is lower than that of MDCT. 14,15 MDCT has approximately 3 Hounsfield units (HU) contrast resolution, while CBCT allows a contrast resolution of 10 HU. 16 CBCT imaging performed in IGRT is usually acquired by using particular imaging geometry, beam characteristics, and reconstruction method for a specific body part in clinical routine practice. However, individual body dimensions would affect the photon statistics in CBCT data, where patients of larger body sizes receive reduced radiation to the isocenter and internal organs, thus causing degradation in image quality. 17 These characteristics of CBCT imaging indicate that optimizing the scan protocols according to patient dimensions is essential to ensure sufficient image quality for daily patient positioning in radiation therapy.
Fractionated radiation treatments are usually delivered in 20 fractions to improve patient tolerance, so the total radiation dose from CBCT is a factor of 20 greater than that of a single scan. 18 Besides, CBCT doses are distributed to the entire imaging region, not only the target volume. 19 Hence, it is necessary to know what the radiation doses are from CBCT and raise awareness of using lower radiation doses. In this study, the tradeoff between target detectability and radiation dose was investigated for CBCT performed in IGRT to achieve sufficient image quality for patient positioning over a course of treatment session while maintaining radiation exposure as low as reasonably achievable (ALARA). The performance of routine MDCT in target detection was also evaluated for comparison purpose. CBCT images were acquired using the on-board imager system installed on a Varian TrueBeam STX radiation therapy machine (Varian Medical System, Palo Alto, CA, USA). Preset parameters are configured per anatomical site for imaging geometry, beam characteristics, and reconstruction method. In the CBCT system we operate, body scans can be obtained in three vendor default modes: Thorax (CBCT lowD ), Pelvis (CBCT midD ), and Pelvis Obese (CBCT highD ) ( Table 1). Once a specific CBCT mode was chosen, the corresponding weighted CT dose index (CTDI w ) was displayed on the operator's console prior to image acquisition. The CBCT lowD protocol used 125 kVp and 270 mAs, which resulted in CTDI w of 0.36 cGy. The

2.B | Image quality evaluation
The calibration phantoms consisting of an electron density phantom and additional annuluses were used to evaluate the image quality of YANG ET AL. The contrast-to-noise ratio (CNR) is an important index for the detection and diagnosis of structure and details of interest in CT, 20 so CNR was used in this study to quantify the target detectability in both MDCT and CBCT. A circular region-of-interest (ROI) of 31 pixels was placed on the target and background regions in 11 slices (the central slice AE5 slices) to calculate the mean and standard deviation of HUs within ROI. The target ROIs were located at the rod inserts simulating various tissue materials, while the background ROI was located at the rod insert made of plastic water. The CNR was defined as where CT# is the mean CT number of the target region, CT# BG and SD BG are the average and standard deviation of CT numbers of the background region, respectively. A CNR of 1.0 occurs when the image contrast (or difference) between target and background was equal to the background noise. Based on our clinical experiences, a CNR of 5 is required for target detection perceived by naked eyes to ensure sufficient geometric accuracy.

2.C | Multivariate analysis
There is a well-recognized tradeoff between image quality and radiation dose in CT imaging. [21][22][23][24][25][26][27] In our department, the tube current of routine MDCT body scans is modulated automatically by the AEC system to achieve consistent image quality for different patient sizes  28 Our model to explain the relationship between independent and dependent variables was: where B 0 to B 3 are the regression coefficient (B) to be estimated.   Figure 3 shows the axial images acquired using CBCT lowD , CBCT midD , CBCT highD (left to right) for CALphan 18 cm , CALphan 28 cm , and CALphan 38 cm (top to bottom).
The shading artifacts were not seen in CALphan 18 cm , but became more pronounced in larger phantoms. For the same calibration phantom, the pattern of shading artifacts was similar in CBCT images acquired with different protocols. The box and whisker plots shown in Fig. 4 summarize the impact of body size, CTDI w and tissue density on CNR of CBCT imaging performed in IGRT. For these box and whisker plots, any value outside the whiskers is considered to be an outlier and marked with a red cross. All data in Fig. 4 The regression model in Eq. (3) yielded an R 2 of 0.8026. Figure 5 shows the CTDI w required for CBCT images achieving CNR = 5 as a function of body size estimated based on the regression model in

| DISCUSSION
Based on naked-eye observation of Fig. 3, the discrimination of rod inserts from the background region becomes more difficult for CBCT images acquired with larger calibration phantoms. This phenomenon was also verified quantitatively in Fig. 4(a). The degradation of target detectability in larger phantoms may be owing to (1) the increase of target detectability was improved by reducing the quantum fluctuations in CBCT, which was verified qualitatively in Fig. 3 and quantitatively in Fig. 4(b). As seen in Fig. 3, the rod inserts simulating bone and lung can still be differentiated from the background region when CALphan 38 cm was imaged by CBCT lowD , but not for the rod inserts simulating soft tissues. Moreover, it was found in Fig. 4(c) that the rod inserts simulating bone and lung have higher CNR values. These  A maximum VIF value in excess of 10 is taken as an indication that multicollinearity may be unduly influencing the least square estimates. YANG ET AL.

| 315
The regression relationship has R 2 larger than 0.80, indicating a good fit to the data. According to our results in Table 2

| CONCLUSION
The tradeoff between target detectability and radiation dose was investigated for CBCT performed in IGRT to ensure sufficient image quality for daily patient positioning in radiation therapy while achieving the ALARA principle. Multivariate analysis was used to examine the impact of body size, radiation exposure and tissue type on the target detectability of CBCT imaging, quantified by CNR. Based on our results, CBCT midD is able to produce images with sufficient target detection of adipose, breast, muscle, and liver in a background of water for normal-weight adults with cross-sectional diameter less than 28 cm, while CBCT highD should be used for adult patients with larger body sizes or higher body mass index. Once the cross-sectional diameter of adult patients is larger than 35 cm, the CTDI w of CBCT scans needs to be higher than 2.78 cGy to achieve CNR of 5.
As for pediatric and adolescent patients with cross-sectional diameter less than 25 cm, CBCT lowD may allow sufficient image quality in abdominal and pelvic scans.

ACKNOWLEDG MENT
This study was supported in part by the Cathay General Hospital in Taiwan (grant number TCCTIC 1052C017).
F I G . 5. CTDI w estimated based on the regression model to achieve CNR = 5 for patients of different body sizes.