The fan-beam CT imaging subsystem on the X1 machine provides eight kVCT scanning protocols with a tube peak kilovoltage (kVp) at 120 kV, the pitch ranging from 0.222 to 1.333, and the X-ray tube current ranging from 45 to 400 mA. Table 1 lists nomenclature and scan parameters of the eight kVCT scanning protocols on the X1 machine. For all scanning protocols, the reconstructed kVCT images have an axial image spacing of 2 mm, axial image resolution of 512 × 512 pixels,

Table 1. Uniformity index (UI) and low contrast visibility (LCV) test results for the CT simulation image and for kVCT images with each scanning protocol.

and axial image dimension of 50 cm × 50 cm. In this study, we evaluated kVCT image quality as compared to CT images from a clinical CT simulator (GE Discovery CT590 RT CT Simulator, GE Healthcare, Chicago, IL). A commercial image quality phantom (CatPhan 600, The Phantom Laboratory, Inc., Salem, NY) was used for evaluation. The phantom contained a uniformity test module with uniform density and an inhomogeneity module containing seven inhomogeneity inserts with relative density ranging from 0.0 to 2.16. The phantom was scanned on the X1 machine using each of the eight scanning protocols and on the clinical CT simulator with a clinical scan protocol using a kVp of 120 kV, pitch of 0.625, and X-ray tube current of 326 mA.

For the inhomogeneity inserts, the kVCT images from the X1 machine and the CT simulation images were evaluated in terms of image uniformity, low contrast visibility, and Hounsfield unit (HU) consistency. To evaluate image uniformity in the uniformity test module region, we used the uniformity index (UI) [14]:

where H U ¯ p e r i p h e r y and H U ¯ c e n t e r are average HU in peripheral and central regions of the uniformity test module, respectively. For commercial CT scanners, fluctuation of CT numbers should not exceed ± 5 HU [15], corresponding of ± 0.5 variation in UI. Visibility of low contrast inserts were evaluated with the low contrast visibility (LCV) metric, defined as [16]:

where H U ¯ P and H U ¯ L are average HU for two inserts with close HU values: polystyrene and low-density polyethylene (LDPE), respectively, and σ P and σ L are the standard deviations of HU for the polystyrene and LDPE inserts, respectively. Both the UI and LCV were calculated with in-house software that utilized a publicly available image analysis software library [17].

We generated HU-to-density tables for each kVCT scanning protocol based on the average CT numbers of the inhomogeneity inserts used for kVCT image dose calculations. To validate dose calculation results with the HU-to-density tables for kVCT images, we used a commercial anthropomorphic thoracic phantom (CIRS Thorax phantom Model 002LFC, Computerized Imaging Reference Systems, Inc., Norfolk, VA) that included bilateral low-density regions that mimic the left and right lungs and a middle high-density region that mimics the vertebral column (Figure 1). A cylindrical planning target volume (PTV) of 3 cm in diameter and 4.5 cm in length was delineated between the lung volumes and anterior to the vertebral column. This phantom was scanned on the CT simulator using a thoracic simulation protocol and on the X1 machine using all available kVCT scanning protocols. Using the CT simulation images, we generated a treatment plan with two 6 MV volumetric modulated arc therapy (VMAT) photon fields to deliver 30 Gy in 10 fractions to the target volume. We then applied the same plan to each individual kVCT image data set per kVCT scanning protocol with the same isocenter location. Three-dimensional dose distributions were calculated in all the treatment plans using the same dose calculation algorithm (the Anisotropic Analytical Algorithm, Varian Medical Systems, Palo Alto, CA), with a scanner- and protocol-specific HU-to-density table for the image used in each treatment plan. We compared the treatment plans of different CT images in terms of dose to the PTV and critical organs.

First, we performed a retrospective review of all patients treated on the X1 machine at our institution since 2021. We included the first ten prostate cancer patients who received a course of image-guided radiotherapy with at least fifteen treatment fractions between 2021 and 2023. Patient characteristics are listed in Table 2. The average age was 69 ± 9 years old (range: 54 - 80). Five of the ten patients had radical prostatectomy and received post-operative radiotherapy, while the other five had intact prostates and received definitive external beam radiotherapy on the X1 machine. Each patient had a CT simulation on the 16-slice CT simulator (Optima 560, GE Healthcare, Chicago, IL) to the pelvic region in a supine position with a slice thickness of 2.5 mm and field-of-view dimension of 50 × 50 cm. Patients were asked to maintain a full bladder during CT simulation by drinking a fixed amount of water (28 ounces) half an hour prior to CT simulation. Patients were also asked to empty their rectums through bowel movements shortly before the CT simulation appointment.

Table 2. Characteristics of prostate cancer patients in this study. STD: standard deviation.

The CT simulation images were sent to a commercial radiotherapy treatment planning system (TPS) (Eclipse v16.1, Varian Medical Systems, Inc., Palo Alto, CA), on which normal organs and target volumes were delineated. The normal organs included the bladder, rectum, penile bulb, sigmoid colon, small intestine, and femoral heads. The gross target volumes (GTVs)/clinical target volumes (CTVs) included the prostate gland and seminal vesicles for patients with intact prostate, prostate fossa for patients with prostatectomy, and pelvic lymph nodes. Margins of 3 - 5 mm were added to the CTVs to form PTVs. The target volume that included the prostate gland or prostate fossa, with or without lymph nodes, was regarded as the primary target. The images and the structure set were then sent to the X1 TPS for plan generation.

The X1 machine has been described previously [18]. Briefly, radiation was delivered with a flattening filter-free 6 MV photon beam modulated by a 64-leaf binary multi-leaf collimator (MLC) and a pair of 2 cm primary jaws to collimate the beam in the longitudinal direction. The couch remained stationary while radiation is administered with a rotational gantry, and the couch moved incrementally with a step size of 2.1 mm while the beam was off. For post-prostatectomy patients, an initial treatment plan was generated with a prescription dose of 45 Gy in 25 equal treatment fractions to the prostate fossa and pelvic lymph nodes, followed by a boost treatment plan with a prescription dose of 19.8 to 21.6 Gy in 11 to 12 equal treatment fractions to the prostate fossa. Among patients with intact prostates, one patient had a treatment plan with a prescription dose of 60 Gy, 55 Gy, and 42 Gy to the prostate, seminal vesicles, and pelvic lymph nodes, respectively, in 20 equal treatment fractions; one patient had a treatment plan with a prescription dose of 70 Gy in 28 treatment fractions to the prostate and seminal vesicles; and three patients had an initial treatment plan with a prescription plan of 45 Gy in 25 treatment fractions to the prostate, seminal vesicles, and pelvic lymph nodes, followed by either permanent prostate seed implant (one patient) or by external beam radiotherapy boost with a prescription dose of 34.2 Gy in 19 treatment fractions (two patients). In plan optimization, a smoothing penalty of 10 and a treatment time penalty of 1000 were applied to all treatment plans while structure-specific optimization parameters were individually tuned based on interim optimization results in an iterative optimization process. In all X1 treatment plans, the optimization parameters were set for a goal of covering 95% of the primary PTV with the prescription dose.

kVCT scans were performed using the integrated fan-beam kVCT system on the X1 machine after patients were aligned on the treatment couch according to skin marks. We used the medium dose/fast couch kVCT scanning protocol and adjusted the couch position translationally and rotationally according to the kVCT image registration with the planning CT images. Following this, patients were then treated with radiation.

Figure 1. Schematic diagram of the workflow for adaptive treatment plan generation and analysis in this study.

For the initial treatment plan during the first part of the treatment course, we generated ART treatment plans using kVCT images from one fraction from every five consecutive treatment fractions (corresponding to the evaluation of one fraction per week). Figure 1 shows the workflow for adaptive treatment plan generation and analysis in this study.

Pre-treatment kVCT images were retrieved from treatment archives and sent to a commercial auto-segmentation system (RT_Mind AI, MedicalMind Ltd., Beijing, China) using a convolutional neural network (CNN) algorithm for automatic delineation of structures in the pelvic region, including the bowel bag, rectum, sigmoid colon, bladder, and prostate gland [18]. The auto-segmentation results were sent to the Eclipse TPS for review and manual correction of AI-generated contours by experience planners. A radiation oncologist reviewed and approved contours used for each adaptive treatment plan. The same sets of normal organs and GTV/CTVs were delineated using pre-treatment kVCT images, and PTVs were created with the same setup margins. The kVCT images and associated structure set were transferred to the X1 treatment planning station for the creation of offline ART treatment plans. All ART treatment plans used the same jaw size of 2 cm and had the same prescription as in the original treatment plan. Similarly, the same normalization criteria were applied to the ART treatment plans. To improve the efficiency of ART treatment planning, the same optimization parameters from the original treatment plan were applied to each subsequent ART treatment plan. Due to the difference in the number of treatment fractions for each patient, a total of 48 ART treatment plans based on kVCT images were created and optimized: five ART treatment plans created for eight patients (40 total) and four ART treatment plans created for the remaining two patients (8 total). Bladder and rectum volume variations were evaluated in all ART treatment plans. Mean doses to the primary PTV and major critical organs were examined and compared with those in the original treatment plans. Pearson’s correlation coefficient was calculated between the daily bladder volume and the mean bladder dose in the adaptive plans, where the correlation was deemed strong when the absolute value of the correlation coefficient was greater than 0.7.

To determine whether kVCT images could be used for adaptive planning, we first compared kVCT image quality to that in CT simulation images. To simulate the different densities of tissues commonly imaged in the human body, we used a commercial image quality phantom with seven inhomogeneity inserts (CatPhan 600). We performed the eight kVCT scanning protocols on these phantom inserts and found that the UI ranged from −0.25 to 0.20, and the LCV ranged from 5.40 to 9.19 for all kVCT scanning protocols (Table 1). In comparison, CT simulation images acquired from the phantom inserts using a clinical scan protocol had UI and LCV values of 0.10 and 6.38, respectively. We found that all UI results were within tolerance for commercial CT scanners [14]. Next, to evaluate how well X1 kVCT images were able to provide quantitative tissue density, we compared the differences in average CT numbers and found that excluding the air insert, average CT numbers ranged from −3.5 to 5.7 HU between the CT simulator and the X1 kVCT images (Figure 2, Supplemental Table S1). Furthermore, when we evaluated the differences between scanning protocols for kVCT images, we found that CT number variations ranged between 1.8 to 3.5 HU for all inhomogeneity inserts. CT number standard deviations for the different density inserts extended from 5.7 to 7.7 HU with the CT simulator image, and from 5.8 to 8.3 HU in the kVCT images from high-dose protocols.

Based on our findings, we then used a commercial anthropomorphic thoracic phantom (CIRS) to mimic the lungs and vertebral column and validate kVCT dose calculation results. Figure 3(A) shows the dose color wash with dose calculated on the planning CT of the CIRS anthropomorphic thorax phantom with contours for the PTV and normal organs, while Figure 3(B) shows the dose color wash with dose calculated on a kVCT image using the low dose & fast couch (LF) scanning protocol. The mean PTV dose in the treatment plan based on CT simulation images was 31.14 Gy and ranged from 31.21 to 31.24 Gy based on dose calculations using kVCT images with different scan protocols from the X1 machine (Figure 3(C), Supplementary Table S2; relative difference: 0.2% - 0.3%). Compared to the mean doses for the left lung, right lung, and spine using CT simulation image dose calculation, mean doses to those organs based on kVCT dose calculation had a relative difference of 0.2% to 0.3% (Figure 3(C), Supplementary Table S2).

Figure 2. Average HU and standard deviations for inhomogeneity inserts in the CT simulation image and kVCT images with each scanning protocol ((A) relative density 0.0, (B) relative density 0.83, (C) relative density 0.92, (D) relative density 1.03, E. relative density 1.42, F. relative density 2.16). Detailed data tables are provided in the supplemental materials.

We included ten prostate cancer patients who received at least 15 treatment fractions of image-guided radiotherapy. Therapists asked these patients to drink 28 ounces of water half an hour before CT simulation and to empty their rectums through bowel movements before CT simulation, and we evaluated bladder and rectum volume variations in kVCT images (Figure 4). For these ten patients, the bladder volume in the planning CT simulation ranged from 67.5 to 609.7 cm³ whereas in kVCT images, the bladder volume fell between 27.1 to 536.6 cm³. Relative to the CT simulation images, the relative bladder volume in the kVCT images extended from 21.2% to 256.0%, and the change in bladder volume varied between −352.3 cm³ to 211.7 cm³. Seven patients had a maximum bladder volume variation of over 50% in the evaluated kVCT images. The rectum volume in planning CT simulation images exhibited a range of 37.7 to 193.0 cm³ while in the kVCT images, the rectum volume was distributed between 37.2 to 163.6 cm³.

Figure 3. Cross section of the CIRS thorax phantom and dose color wash based on (A) the CT simulation image and (B) a kVCT image using the low dose & fast couch speed scanning protocol. (C) Variation of mean dose to the target and normal organs for the CIRS thorax phantom with different X1 kVCT images relative to plan dose based on the CT simulation image.

Figure 4. Weekly (A) bladder volume and (B) rectum volume variations for each patient.

Relative to the rectum volume in CT simulation images, the rectum volume in daily kVCT images showed a variation from 28.2% to 213.2%, and the change in rectum volume ranged between −138.6 cm³ to 81.8 cm³.

Based on these findings, we created and optimized 48 ART treatment plans for the ten patients (Figure 5). When we compared the optimized ART treatment plans to the original treatment plans of each patient, we found that the mean dose variations to the primary PTV, bladder, rectum, and penile bulb ranged from 96.4% to 108.9%, 80.9% to 117.3%, 76.9% to 118.6%, and 42.6% to 144.4%, respectively. When we averaged the mean dose to each structure was over all ART treatment plans for each patient, the mean dose extended from 98.4% to 107.7% for the primary PTV, 87.7% to 109.5% for the bladder, 92.5% to 103.2% for the rectum, and 62.8% to 114.6% for the penile bulb (Figure 6).

Figure 5. Mean dose to the primary PTV and major critical organs in the original treatment plan and adaptive treatment plans. Data for the adaptive plans are plotted in sequential order of the treatment fractions for each patient.

We next compared the mean and standard deviations of representative dosimetric parameters in ART treatment plans relative to those in original treatment plans (Figure 7, Supplementary Table S3). In ART plans, the average primary PTV coverage by the prescribed dose was 96.0% ± 1.7%, compared to 95.7% ± 1.9% in original treatment plans. In contrast, the mean dose in the original treatment plans for the primary PTV, bladder, rectum, and penile bulb were 100.9% ± 2.7%, 97.8% ± 9.2%, 98.1% ± 9.2%, and 96.1% ± 24.7%, respectively. Strong correlations (Pearson correlation coefficient > 0.7 or < −0.7) between the daily bladder volume and mean bladder dose in the adaptive plans were found in only one patient, where the mean bladder dose ranged from 24.8 to 27.9 Gy in the adaptive plans.

Figure 6. Representative combined DVH for (A) a patient treated to the prostate and (B) a patient treated to the prostate bed. The solid curves are those from the treatment plan and the dotted curves are those from adaptive treatment plans based on daily kVCT images. Magenta: primary PTV; blue: penile bulb; green: bladder; yellow: rectum.

Figure 7. Mean and standard deviations of the primary PTV coverage by the prescription dose (V(Rx)) and mean dose relative to those in the original treatment plans for the primary PTV (magenta), bladder (green), rectum (yellow), and penile bulb (blue) in the ART treatment plans.