Megavoltage CT enhancement for cervical cancer tomotherapy using a generative adversarial network with deformable convolution and self-attention.

Megavoltage Computed Tomography Imaging: A Potential Tool to Guide and Improve the Delivery of Thoracic Radiation Therapy

Purpose: Intensity modulated radiation therapy (IMRT) planning using pre‐treatment megavoltage computed tomography (MVCT) provides a viable solution to adaptive radiotherapy. A strategy was developed to assess the impact of MVCT image quality on treatment plans using Helical TomoTherapy. Methods: The Octavius phantom (PTW) was scanned with a Philips CT‐scanner and the Helical TomoTherapy unit for kilovoltage CT (kVCT) and MVCT, respectively. MVCT images were acquired with three modes (slice thickness): coarse (6‐mm), normal (4‐mm) and fine (2‐mm). Contours from existing head‐and‐neck plans were transferred from the Pinnacle treatment planning system to phantom images and IMRT plans were generated using the TomoTherapy planning station. Comparison between the kVCT‐based plan (benchmark plan) and MVCT‐based plans were conducted to evaluate dosimetric variations. Dose volume histograms (DVH) were examined for plan quality. Plans using MVCTs acquired with three modes were also evaluated for dosimetric differences. Additional plan evaluations were performed with retrospective dose reconstruction based on the delivered leaf sequences. Results: Isodose distributions of four plans are comparable despite the intrinsic difference between MVCT and kVCT images. No significant variations on the target coverage were found from DVHs. However, plan using coarse MVCT images gave largest dose deviations (> 50%) for critical structures from that of benchmark plan while the best agreement was observed between fine MVCT and kVCT. Plan parameters were found to be consistent among four plans. Comparisons between planning doses and reconstructed treatment doses revealed high fidelity delivery. Conclusions: IMRT planning using MVCTs proved to be equivalent with that using standard kVCT when fine imaging mode was selected. Target coverage on MVCT plans is comparable to that on the standard kVCT plan. MVCT thicknesses were shown to have impact on plan quality. Caution is needed when performing treatment planning with MVCTs.

Helical tomotherapy is a unique approach to image-guided IMRT that combines features of a linear accelerator and a CT scanner. This design allows generation of megavoltage CT (MVCT) images, which among other uses, are used to verify daily setup. In this study, we assessed the image-quality, absorbed radiation doses, and clinical practicality of MVCT from our helical tomotherapy prototype unit. Phantom studies were first performed to assess the capabilities of MVCT. Next, MVCT images from human patients prospectively enrolled on institutional review board-approved imaging and treatment protocols were analyzed. MVCT was obtained using a 4-MV beam from the University of Wisconsin helical tomotherapy prototype device. These scans were compared with conventional kilovoltage (kVCT) images from a diagnostic CT scanner. MVCT images in phantoms demonstrate an ability to detect contrast differences as small as 3%. Small objects, 1.2 to 1.6 mm, were seen with good resolution. In human subjects, MVCT imaging of tumor targets and normal anatomy revealed sufficient detail for patient repositioning. MVCT imaging of metallic objects showed minimal artifact in comparison with kVCT. Patient scans were obtained in about 1 to 5 minutes and resulted in absorbed radiation doses of 1.5 to 3 cGy. MVCT is an elegant pretreatment position and setup verification tool. MVCT images of human subjects obtained from the helical tomotherapy unit showed good resolution and contrast. The high-quality three-dimensional information permits its use in day-to-day setup verification. The unique properties of MVCT also provide the potential for primary imaging of anatomic regions near metal prostheses as well as nonmedical applications. Additional investigations are underway to improve image quality, further reduce patient dose, and aid adaptive radiotherapy and dose reconstruction.

BackgroundAccurate Hounsfield units (HU) are critical for dose calculation and anatomical visualization, but are often affected by dental artifacts in head and neck (H&N) cancer patients. The HyperSight cone‐beam computed tomography (CBCT) platform provides improved image quality over previous CBCT platforms and offers metal artifact reduction (iCBCT MAR) reconstruction.PurposeThis study evaluates the quality of HyperSight CBCT images compared to current clinical standards: TrueBeam CBCT for image guidance and fan‐beam CT (FBCT) from a CT simulator for treatment planning, using images captured during H&N cancer treatment.MethodsImages for 30 H&N cancer patients were acquired on a HyperSight CBCT, conventional TrueBeam and FBCT, with 24 patients exhibiting metal dental artifacts. The HyperSight images were reconstructed using iCBCT MAR and iterative (iCBCT Acuros) algorithms. The four image sets were rigidly registered and compared using the artifact index (AI) measured in the oral cavity and the percentile range (PR) measured in the oral cavity, brain, brainstem and eyes to assess image non‐uniformity. The HU accuracy was calculated relative to FBCT (baseline) for soft tissues (oral cavity, brainstem, submandibular and parotid glands), and bone (mandible). The contrast relative to baseline was evaluated between the oral cavity and nearby structures. Image‐based metrics were computed relative to FBCT including structural similarity index measure (SSIM), mean‐square error (MSE) and peak signal‐to‐noise ratio (PSNR).ResultsThe HyperSight iCBCT MAR images showed a significant reduction in AI values compared to the other images (p < 0.0004), but higher PR values indicating decreased HU uniformity compared to HyperSight iCBCT Acuros and FBCT (p < 0.0002). The soft‐tissue HU and contrast values were significantly closer to baseline in both HyperSight images compared to TrueBeam (HU: p < 0.001, contrast: p < 0.001). For soft‐tissue the HU mean absolute deviation (MAD) from baseline was 16 ± 10 HU for HyperSight iCBCT Acuros, 15 ± 10 HU for HyperSight iCBCT MAR, and 35 ± 22 HU for TrueBeam. For bone, the HU MAD from baseline was 153 ± 233 HU, 185 ± 268 HU, and 214 ± 212 HU, respectively. The HyperSight iCBCT Acuros algorithm achieved significantly superior SSIM, MSE, and PSNR metrics compared to TrueBeam and HyperSight iCBCT MAR in regions with large amounts of bone and air.ConclusionsHyperSight iCBCT MAR significantly reduced artifacts compared to HyperSight iCBCT Acuros, TrueBeam and FBCT, making it particularly beneficial for patients with metal implants. Both HyperSight reconstructions demonstrated improved soft‐tissue HU accuracy and contrast compared to TrueBeam, however the iCBCT Acuros algorithm may be preferred when metal‐induced artifacts are not a concern. These results support the suitability of HyperSight images in adaptive treatment workflows requiring accurate image quality, even with severe metal artifacts.

Objective. In helical tomotherapy, image-guided radiotherapy employs megavoltage computed tomography (MVCT) for precise targeting. However, the high voltage of megavoltage radiation introduces substantial noise, significantly compromising MVCT image clarity. This study aims to enhance MVCT image quality using a deep learning-based denoising method. Approach. We propose an unpaired MVCT denoising network using a coupled generative adversarial network framework (DeCoGAN). Our approach assumes that a universal latent code within a shared latent space can reconstruct any given pair of images. By employing an encoder, we enforce this shared-latent space constraint, facilitating the conversion of low-quality (noisy) MVCT images into high-quality (denoised) counterparts. The network learns the joint distribution of images from both domains by leveraging samples from their respective marginal distributions, enhanced by adversarial training for effective denoising. Main Results. Compared to an analytical algorithm (BM3D) and three deep learning-based methods (RED-CNN, WGAN-VGG and CycleGAN), the proposed method excels in preserving image details and enhancing human visual perception by removing most noise and retaining structural features. Quantitative analysis demonstrates that our method achieves the highest peak signal-to-noise ratio and Structural Similarity Index Measurement values, indicating superior denoising performance. Significance. The proposed DeCoGAN method shows remarkable MVCT denoising performance, making it a promising tool in the field of radiation therapy.

PurposeModern computed tomography (CT) scanners have an extended field‐of‐view (eFoV) for reconstructing images up to the bore size, which is relevant for patients with higher BMI or non‐isocentric positioning due to fixation devices. However, the accuracy of the image reconstruction in eFoV is not well known since truncated data are used. This study introduces a new deep learning‐based algorithm for extended field‐of‐view reconstruction and evaluates the accuracy of the eFoV reconstruction focusing on aspects relevant for radiotherapy.MethodsA life‐size three‐dimensional (3D) printed thorax phantom, based on a patient CT for which eFoV was necessary, was manufactured and used as reference. The phantom has holes allowing the placement of tissue mimicking inserts used to evaluate the Hounsfield unit (HU) accuracy. CT images of the phantom were acquired using different configurations aiming to evaluate geometric and HU accuracy in the eFoV. Image reconstruction was performed using a state‐of‐the‐art reconstruction algorithm (HDFoV), commercially available, and the novel deep learning‐based approach (HDeepFoV). Five patient cases were selected to evaluate the performance of both algorithms on patient data. There is no ground truth for patients so the reconstructions were qualitatively evaluated by five physicians and five medical physicists.ResultsThe phantom geometry reconstructed with HDFoV showed boundary deviations from 1.0 to 2.5 cm depending on the volume of the phantom outside the regular scan field of view. HDeepFoV showed a superior performance regardless of the volume of the phantom within eFOV with a maximum boundary deviation below 1.0 cm. The maximum HU (absolute) difference for soft issue inserts is below 79 and 41 HU for HDFoV and HDeepFoV, respectively. HDeepFoV has a maximum deviation of −18 HU for an inhaled lung insert while HDFoV reached a 229 HU difference. The qualitative evaluation of patient cases shows that the novel deep learning approach produces images that look more realistic and have fewer artifacts.ConclusionTo be able to reconstruct images outside the sFoV of the CT scanner there is no alternative than to use some kind of extrapolated data. In our study, we proposed and investigated a new deep learning‐based algorithm and compared it to a commercial solution for eFoV reconstruction. The deep learning‐based algorithm showed superior performance in quantitative evaluations based on phantom data and in qualitative assessments of patient data.

Javanese script is an important part of Indonesia’s cultural heritage, representing cultural values from the past. However, recognizing and classifying Javanese characters within manuscripts is challenging due to the limited availability of data and uneven distribution of character classes. The decline in formal use of Javanese script has drastically reduced the pool of manuscript samples, causing certain characters to appear rarely and skewing class frequencies. Existing methods that utilize Generative Adversarial Networks (GANs) attempt to address this problem. However, they often struggle to generate characters that are both consistent and visually accurate in terms of structural details. To address these issues, this study introduces a skeleton-based balancing GAN (SkelBAGAN), which improves the structural details of the previous method for generating characters. The proposed method introduces three main enhancements: (i) a layer for extracting the character skeleton structure, (ii) an optimized pretrained network using an autoencoder for learning the skeleton distribution, and (iii) refinement of the evaluation function, preserving both the distribution and structural fidelity in the adversarial process. The performance of the proposed model is evaluated against previous methods using the Fréchet Inception Distance (FID) to assess distribution quality and the Structural Similarity Index Measure (SSIM) to evaluate structural fidelity. The results indicate that the proposed methods outperform previous methods in balancing the FID and SSIM metrics. The integration of all enhancements in SkelBAGAN achieves the lowest FID, indicating improved generative quality while maintaining competitive SSIM values. The qualitative study indicates that SkelBAGAN outperforms previous methods in character generation. These results highlight how the skeleton-based improvement of the quality of generated characters enhances the recognition performance for underrepresented Javanese characters in imbalanced datasets. Ultimately, this work contributes to the broader effort to preserve the Javanese script as a vital element of Indonesia’s cultural identity.

Purpose: Verify the accuracy of the dose calculation algorithm in the presence of a hip prosthesis when planning with megavoltage CT (MVCT) images of the TomoTherapy Hi‐Art II planning system. Methods and Materials: Artifacts from imaging high‐Z materials are greatly reduced in MVCT compared to kilovoltage CT (kVCT). This may allow more accurate treatment planning for patients with metallic implants. The Radiological Physics Center's (RPC) pelvic phantom was modified to accommodate a commercial Co‐Cr‐Mo hip prosthesis and imaged with both kVCT and MVCT. On the TomoTherapy Hi‐Art II planning system, the MVCT‐to‐density table was extended to high‐Z materials by including stainless steel and lead in the calibration. Contours were drawn on kVCT images without the prosthesis present and fused to the MVCT image set. Helical tomotherapy plans were created using the MVCT images with no constraints on beam entry or exit location. The delivered dose distribution was measured using TLDs in the PTV and EBT radiochromic film in the coronal and sagittal planes. Criteria based on TG‐53 recommendations were used for comparison between measurements and calculations. Results: Dose calculation with MVCT images resulted in minor variations between calculated and measured dose. It was found that on average 97% of the dose distribution measured with film agreed with the treatment planning system to within +/−5% or 3mm. Profiles across the PTV also showed good agreement between measurement and calculation. The average TLD dose at the center of the PTV was 96.0% of the expected dose. Conclusion: Using dose calculations based on MVCT images, it is possible to accurately calculate the dose distribution in the pelvic region while treating through a high‐Z prosthetic implant. Work was supported by PHS grant CA10953 and CA081647 awarded by NCI.

Cone-beam CT (CBCT) is widely used in dental clinics but exhibits limitations in assessing soft tissue pathology because of its lack of contrast resolution and low Hounsfield Units (HU) quantification accuracy. We aimed to increase the image quality and HU accuracy of CBCTs while preserving anatomical structures. We generated CT-like images from CBCT images using a patchwise contrastive learning-based GAN model. Our model was trained on unpaired CT and CBCT datasets with the novel combination of losses and the feature extractor pretrained on our training dataset. We evaluated the quality of the images generated by our model in terms of Fréchet inception distance (FID), peak signal-to-noise ratio (PSNR), mean absolute error (MAE), and root mean square error (RMSE). Additionally, the structure preservation performance was assessed by the structure score. As a result, the generated CT-like images by our model were significantly superior to those generated by various baseline models in terms of FID, PSNR, MAE, RMSE, and structure score. Therefore, we demonstrated that our model provided the complementary benefits of preserving the anatomical structures of the input CBCT images and improving the image quality to be similar to those of CT images.

To evaluate pix2pix and CycleGAN and to assess the effects of multiple combination strategies on accuracy for patch-based synthetic computed tomography (sCT) generation for magnetic resonance (MR)-only treatment planning in head and neck (HN) cancer patients. Twenty-three deformably registered pairs of CT and mDixon FFE MR datasets from HN cancer patients treated at our institution were retrospectively analyzed to evaluate patch-based sCT accuracy via the pix2pix and CycleGAN models. To test effects of overlapping sCT patches on estimations, we (a) trained the models for three orthogonal views to observe the effects of spatial context, (b) we increased effective set size by using per-epoch data augmentation, and (c) we evaluated the performance of three different approaches for combining overlapping Hounsfield unit (HU) estimations for varied patch overlap parameters. Twelve of twenty-three cases corresponded to a curated dataset previously used for atlas-based sCT generation and were used for training with leave-two-out cross-validation. Eight cases were used for independent testing and included previously unseen image features such as fused vertebrae, a small protruding bone, and tumors large enough to deform normal body contours. We analyzed the impact of MR image preprocessing including histogram standardization and intensity clipping on sCT generation accuracy. Effects of mDixon contrast (in-phase vs water) differences were tested with three additional cases. The sCT generation accuracy was evaluated using mean absolute error (MAE) and mean error (ME) in HU between the plan CT and sCT images. Dosimetric accuracy was evaluated for all clinically relevant structures in the independent testing set and digitally reconstructed radiographs (DRRs) were evaluated with respect to the plan CT images. The cross-validated MAEs for the whole-HN region using pix2pix and CycleGAN were 66.9±7.3 vs 82.3±6.4HU, respectively. On the independent testing set with additional artifacts and previously unseen image features, whole-HN region MAEs were 94.0±10.6 and 102.9±14.7HU for pix2pix and CycleGAN, respectively. For patients with different tissue contrast (water mDixon MR images), the MAEs increased to 122.1±6.3 and 132.8±5.5HU for pix2pix and CycleGAN, respectively. Our results suggest that combining overlapping sCT estimations at each voxel reduced both MAE and ME compared to single-view non-overlapping patch results. Absolute percent mean/max dose errors were 2% or less for the PTV and all clinically relevant structures in our independent testing set, including structures with image artifacts. Quantitative DRR comparison between planning CTs and sCTs showed agreement of bony region positions to <1mm. The dosimetric and MAE based accuracy, along with the similarity between DRRs from sCTs, indicate that pix2pix and CycleGAN are promising methods for MR-only treatment planning for HN cancer. Our methods investigated for overlapping patch-based HU estimations also indicate that combining transformation estimations of overlapping patches is a potential method to reduce generation errors while also providing a tool to potentially estimate the MR to CT aleatoric model transformation uncertainty. However, because of small patient sample sizes, further studies are required.

MRI has been increasingly implemented for the treatment planning process and the MRI-only workflow in both photon and proton therapy. This study aims to investigate and explore the accuracy of dose calculations in head and neck region based on synthetic CT (sCT) images, using the commercial software methods for both volumetric modulated arc therapy (VMAT) and intensity modulated proton therapy (IMPT) plans. The sCT images were generated from MRI images for five head and neck cancer cases that underwent VMAT and IMPT by MRI Planner commercial software. The mean absolute error (MAE) and mean error (ME) within the body, bone and brainstem contour and dice scores in body, and bone were used for evaluation. The treatment plans of VMAT (3 arcs) and IMPT (3-6 beam direction) were generated and optimized in the Eclipse treatment planning system (Version 16.1). The treatment plans were originally optimized and calculated on reference CT images and subsequently recalculated on synthetic CT images. The dose distributions were evaluated at D98%, D95% and D2% of PTV high risk and low risk for VMAT plans and CTV high risk and low risk for IMPT plans, Dmean and Dmax of organs at risk. Moreover, the dose comparisons between reference CT (rCT) and synthetic CT (sCT) images were analyzed at 3%/2mm gamma criteria. The MAE within the body, bone, and brainstem contour was 137.7±40.68, 387.27±114.84, and 33.07±2.78 Hounsfield Units (HU). Dice scores of body, and bone were 0.92±0.05 and 0.82±0.09. Comparing the average dose, the dose difference between using rCT and sCT images for VMAT was less than 2% in the PTV and less than 3% in the CTV. Additionally, in organs at risk, the highest dose difference between using rCT and sCT images was 30.09% for VMAT and 22.56% for IMPT. The average gamma passing rate at 3%/2mm was 98.8% for VMAT plan in PTV HR and 98.3% for PTV LR, respectively. The average gamma passing rate at 3%/2mm was 91.0% for IMPT plan in CTV HR and 93.8% for CTV LR, respectively. The dose difference between using rCT and sCT images has a greater impact for IMPT compared to VMAT. The commercial software used for generating sCTs has demonstrated significant dose differences for IMPT over VMAT plans for the head and neck region. The dose calculations based on these sCTs lead to more inaccurate dosimetric results; therefore, it should be considered when dealing with proton treatment plans.

Modeling set-up error by daily MVCT for prostate adjuvant treatment delivered in 20 fractions: Implications for the assessment of the optimal correction strategies

Purpose: To investigate the effectiveness of highly conformal radiation modalities in conjunction with a patient setup correction method, Helical Tomotherapy (HT) was employed using the Helical Megavoltage CT (MVCT) scanner of the tomotherapy unit. Any discrepancy between the planned and delivered dose distributions would likely affect the clinical outcome. Method and Materials: The case of a head and neck cancer was used to evaluate the accuracy of dose delivery. The right parotid, left parotid and spinal cord are the primary organs at risk proximal to the target volume. MVCT images were taken before and after the patient setup correction. MVCT images of the patient were acquired daily. The three dose distributions (planned, with and without patient setup correction) were compared by using the complication‐free tumor control probability (P +) and the biologically effective uniform dose (BEUD) concepts. Results: For the computerized treatment plan, the maximum value of P + is 88.3%, the total control probability, P B is 97.3% and the total risk for complications, P I is 9.3%. For the delivered dose distribution without patient setup correction, the value of P + is 79.8% and the probabilities P B and P I are 83.6% and 3.8%, respectively. Finally, for the delivered dose distribution with patient setup correction in every fraction the value of P + is 87.2% and the probabilities P B and P I are 92.7% and 5.5%, respectively. The expected clinical effectiveness dropped by a ΔP + of 1.1% and 8.5% in the cases with and without patient setup correction. Conclusion: Highly conformal treatment plans can deliver higher doses to the target allowing for increased control rates for the same or a reduced risk for complications. The results show that the effectiveness of the applied dose distribution drops between the treatment plan and the dose delivery with and without patient setup correction.

The purpose of this investigation was the creation of a new protocol allowing more precise dose calculations on megavoltage CT (MVCT) images for tomotherapy, both for adaptive and StatRT planning. Daily MVCT images offer, next to positioning purposes, the possibility for daily dose check and adaptive planning. Dose calculations use the image value to density table (IVDT) to calculate physical densities from Hounsfield Units (HUs). These measured HUs change over time, leading to a dose calculation error. We noticed dose calculation variations due to IVDT changes of: 0.2% dose during a day, up to 1.6% dose from long‐term variations, and up to 1.5% dose due to technical interventions. An analysis was performed applying the general methodology of a calibration problem. A model HU=bρc‐1020 was obtained using a weighted least squares inverse prediction method (HU as function of density) taking into account the heteroscedasticity. The b parameter is the major variable and depends also on the dose rate (DR). We demonstrate the correction for DR variations and the constance of the c parameter. Instead of scanning the whole tissue characterization phantom daily, we propose a simplified daily protocol: (a) morning airscan‐like procedure with only two inserts on the table (defining b and thus the IVDT curve), (b) DR variations throughout the day can be corrected for using the DR model. A patient‐specific protocol for which two inserts next to the patient are scanned could also be used, but results in equal uncertainties and is less practical. Therefore we recommend the morning procedure with dose rate variation correction. Applying the proposed transformations and the model, the correct IVDT of the moment can be reconstructed, with a simple measurement in the morning, and corrected with DR changes during the day. This corresponds with a linear mapping in time of the proposed IVDT function. The dosimetric variation is hereby reduced from up to 3% to 0.4 % for the tested pelvic and head‐and‐neck cases. In practice, several IVDT curves corresponding to “b” values can be entered. The correct IVDT curve of that moment can then be chosen from the list. Instead of the two high‐density inserts on table, any calibrated single density phantom could be used in order to create the IVDT curve of the day, but it should have a larger size than the current inserts.PACS number: 87.55.Gh

Purpose: Correcting anterior‐posterior, lateral, superior‐inferior, pitch, roll, and yaw with Image‐Guided Radiation Therapy (IGRT) will not completely correct for anatomical deformation. The purpose of this work was to identify anatomical deformation error and develop multiple strategies for addressing this error using adaptive IGRT. Method and Materials: Ten H&amp;N patients have been treated with helical tomotherapy. Megavoltage CT (MVCT) images were acquired prior to treatment The daily setup were then used to calculate the random positional setup errors between the setup marks on the patient's mask and the internal anatomy. Level I lymph nodes (LV1) and the cord were contoured on each MVCT image. DVHs and LV1 EUDs were then calculated for each fraction. If the 2 Gy per fraction isodose line did not cover the LV1 nodes, then the treatment delivery was re‐planned and adapted to account for anatomical deformation. Results: The total shifts were measured for over 400 treatment fractions. The standard deviation was ±2.8 mm in the AP direction, ±2.9 mm in the lateral direction, and ±3.2 mm in SI direction. Dose reconstruction results indicated that the LV 1 nodes were frequently being under‐dosed (even with IGRT). Using IGRT to align the patient relative to the spinal cord resulted in a reduction in the LV1 nodal EUD by 2 to 6 Gy. This is of particular concern because the risk (5 to 50 percent) of LV1 involvement for node‐positive patients. By utilizing adaptive therapy, the LV1 were completely covered by the prescribed dose (50 Gy) in all cases. In addition, dose to the cord and mandibles were minimized. Conclusion: Adapting the treatment delivery by modifying the anterior border of the field can correct for the decrease in coverage due to anatomical deformation. This technique is an alternative to increasing PTV margins that allows increased sparing of normal tissues.