IntroductionIn the preoperative planning of periacetabular osteotomy (PAO), anterior acetabular rotation as well as lateral rotation should be considered in patients with inadequate anterior acetabular coverage. False profile (FP) radiographs are used to measure the anterior center edge angle (ACEA) to assess anterior acetabular coverage, but pelvic rotation can affect the measurements. The aim of this study was to (1) set a valid reference value for ACEA for preoperative planning of PAO, (2) investigate the effects of pelvic rotation from FP radiographs on the measured ACEA, and (3) determine the range of “correct positioning.”Materials and MethodsWe analyzed 61 patients (61 hips) who underwent PAO. ACEA was measured in each digitally reconstructed radiography (DRR) image of the FP radiograph reconstructed in different degrees of pelvic rotation. Detailed simulations were performed to determine the range of “correct positioning” (0.67 < ratio of the distance between the femoral heads to the diameter of the femoral head < 1.0). The vertical-center-anterior (VCA) angle was measured on the CT sagittal plane, and its correlation with the ACEA was investigated. The reference value of ACEA was determined by receiver operating characteristic (ROC) curve analysis.ResultsThe ACEA measurement increased by 0.35° for every 1° pelvic rotation approaching the true lateral view. The pelvic rotation that met the “correct positioning” criteria was found at 5.0° (range: 63.3°-68.3°). The ACEA on the FP radiographs showed a good correlation (r=0.71, P<0.001) with the VCA angle. The ROC curve revealed that an ACEA <13.6° was associated with inadequate anterior coverage (VCA <32°).ConclusionsDuring preoperative PAO planning, anterior acetabular rotation should be considered in addition to lateral rotation in patients with an ACEA <13.6° on FP radiographs. Images with "correct positioning" can also have a measurement error of 1.7°due to the pelvic rotation.

Two-dimensional to three-dimensional (2D/3D) registration is critical in image-guided interventions, particularly in vascular procedures such as endovascular therapy (EVT), where accurate alignment between preoperative 3D images and intraoperative 2D x-ray angiography can improve procedural safety and precision. However, achieving both real-time and precise 2D/3D registration remains challenging due to high computational demands. This study aims to develop a fast 2D/3D registration method that directly optimizes the in-plane alignment between digitally reconstructed radiographs (DRRs) and x-ray angiography images, thereby eliminating the need for forward projection at each objective function evaluation and improving computational efficiency without compromising registration accuracy. We propose a registration method that solves an alternative 2D/2D registration problem to estimate in-plane transformation parameters and iteratively updates the underlying 2D/3D registration parameters. The method incorporates a skull region-of-interest (ROI) map to reduce anatomical mismatches between DRRs and angiography images, and maximizes a similarity metric, normalized cross-correlation (NCC), in an iterative manner. The algorithm was evaluated using both numerical simulations (XCAT phantom) and clinical angiography data, comparing performance to an intensity-based 2D/3D registration and a version of our method without skull ROI guidance. The success rate was evaluated differently depending on the data source: based on the NCC in simulations, and on parameter accuracy compared to manual estimations by two experts in clinical data. Also, registration parameter accuracy, runtime, number of forward projections, and convergence behavior were assessed. In the numerical simulation, the proposed method reduced runtime by more than 10 compared to conventional 2D/3D registration, while maintaining an equivalent success rate. In the clinical study, it achieved runtime reduction of approximately 6 and a 100% success rate in non-truncated cases, where the head was mostly captured within the x-ray angiography field of view, and demonstrated strong visual agreement with the manual estimation results. In truncated cases, although the registration performance decreased across all methods, the proposed method outperformed the 2D/3D registration. We proposed a fast 2D/3D registration method that achieved significantly reduced runtime and improved accuracy. Despite being limited to rigid registration, it shows promise for real-time clinical use with further optimization.

Fiducial marker implantation for tumor localization in radiotherapy is effective but invasive and carries complication risks. To address this, we propose a marker-less tumor tracking framework to explore the feasibility of a cross-patient deep learning model, aiming to eliminate the need for per-patient retraining. A novel degradation model generates realistic simulated data from digitally reconstructed radiographs (DRRs) to train a Restormer network, which transforms clinical fluoroscopic images into clean, DRR-like images. Subsequently, a DUCK-Net model, trained on DRRs, performs tumor segmentation. We conducted a feasibility study using a clinical dataset from 7 lung cancer patients, comprising 100 distinct treatment fields. The framework achieved an average processing time of 179.8 ms per image and demonstrated high accuracy: the median 3D Euclidean tumor center tracking error was mm, with directional errors of mm (LR), mm (SI), and mm (AP). These promising results validate our approach as a proof-of-concept for a cross-patient marker-less tumor tracking solution, though further large-scale validation is required to confirm broad clinical applicability.

DRR dosimetry; Introducing a new application of Digitally Reconstructed Radiographs for evaluation of dose distribution in radiation therapy.

Limited-angle cone-beam CT (LA-CBCT) reduces imaging time and dose but suffers from severe under-sampling artifacts and distortions. 2D-3D deformable registration mitigates this issue by estimating LA-CBCTs through the deformation of a prior, fully-sampled CT/CBCT, using deformation-vector-fields (DVFs) optimized by limited-angle cone-beam projections. Population-trained 2D-3D registration networks enable fast inference but face accuracy challenges, particularly under varying limited-angle scan directions. On the other hand, patient-specific models are more adaptable but typically require considerable runtimes to optimize model parameters from scratch for each case. To improve the accuracy and efficiency of 2D-3D registration-driven LA-CBCT estimation, a hybrid 2D-3D deformable registration framework was proposed. The hybrid population-based and patient-specific 2D-3D deformable registration framework (HB-2D3DReg) synergized the advantages of both population-based and patient-specific approaches while mitigating their limitations. It integrated the fast inference of population-trained models with the test-time adaptability of patient-specific models through a two-stage approach. First, a population-based 2D-3D registration network, 2D3D-RegNet, was trained on a cohort dataset in an unsupervised manner, with a similarity loss defined between digitally reconstructed radiographs (DRRs) of the estimated LA-CBCTs and limited-angle 2D projections. Then, a 2D-3D registration network based on implicit neural representation (INR), 2D3D-INR, refined the DVFs solved by the population-based model during test time for each independent testing case. The population-based 2D3D-RegNet accelerated the optimization of the patient-specific 2D3D-INR and reduced the latter's chance of being trapped at a local optimum, while the patient-specific network, in turn, enhanced the accuracy of the population-based model. HB-2D3DReg was evaluated using a dataset of 48 4D-CTs, 26 of which were used to train the population-based model and 22 for testing. Different limited-angle scan scenarios, featuring varying scan directions and angles, were assessed. HB-2D3DReg attained superior LA-CBCT estimation and registration accuracy. Under an orthogonal-view 90° scan (45° each) with varying scan directions, HB-2D3DReg achieved mean (±S.D.) image relative error of 7.99±2.16% and target registration error of 3.70±1.94mm, compared to 15.40±2.41% and 8.52±3.31mm (no registration), 9.82±2.12% and 6.38±2.46mm (2D3D-RegNet only), and 9.71±2.33% and 5.01±2.77mm (2D3D-INR only) on the DIR-lab dataset. HB-2D3DReg took ∼3min to optimize at test time, compared to 13min for the 2D3D-INR method. HB-2D3DReg achieved accurate and robust 2D-3D deformation registration for LA-CBCT estimation, enabling efficient anatomy monitoring to guide radiotherapy. The code will be released at: https://github.com/sanny1226/HB-2D3DReg.

PurposeDigital energy modulation is a novel framework with the potential to enhance projectional x‐ray imaging by enabling translation between different x‐ray energy domains. We evaluate the feasibility of integrating machine learning methods into this approach by leveraging digitally reconstructed radiographs (DRRs) generated from dual‐energy CT datasets.MethodsDRRs were created in 15° increments from 0° to 90°, producing 3500 images per energy domain (2 polyenergetic, 4 monoenergetic). A supervised deep‐learning approach was used to train models for energy translation, focusing on conversions between polyenergetic domains and from polyenergetic to monoenergetic images. Model performance was assessed using peak signal‐to‐noise ratio (PSNR), structural similarity index (SSIM), mean squared error (MSE), and mean absolute percentage error (MAPE). Cross‐validation and projection‐specific dataset splits were used for evaluation.ResultsThe models trained using cross‐validation on the various energy translations achieved the following results: PSNR: 29.1 ± 2.0, SSIM: 0.947 ± 0.017, MSE: 169.1 ± 68.3, MAPE: 8.2% ± 1.8%. When translating between polyenergetic high‐energy and low‐energy domains in projection‐specific datasets (anterior‐posterior [0°] and lateral [90°] views), models achieved the following results: PSNR: 27.4 ± 0.5, SSIM: 0.909 ± 0.003, MSE: 195.9 ± 39.7, MAPE: 10.4% ± 2.1%.ConclusionThese findings demonstrate the feasibility of a digital energy modulation framework for projectional x‐ray imaging using machine learning for energy translation. The results support the potential of this approach to enhance projectional x‐ray imaging, though future work is needed to refine the models and further explore clinical applications.

Purpose: In markerless tumor tracking (MTT) with deep learning, model performance suffers from domain shifts due to noise and anatomical changes. This study aimed to develop a convolutional neural network (CNN) model for real-time MTT segmentation. Methods: An Uncertain Feature-refinement Attention Unet (UFA-Unet), designed based on insights into CNN behavior under domain distribution shifts that occur between digitally reconstructed radiographs (DRRs) and kV X-ray fluoroscopic (XF) images, is proposed. A qualitative ablation study was performed to examine the contribution of each UFA-Unet component to segmentation accuracy. The model feasibility of UFA-Unet was evaluated through quantitative and phantom studies. The quantitative study included ten lung cancer cases, each containing two datasets (1 st -plan and 2 nd -plan), with a mean interval of 28 days between four-dimensional computed tomography (4DCT) acquisitions. Patient-specific models were trained on 1 st -plan DRRs and validated using noise-injected 1 st -and 2 nd -plan DRRs. In the phantom study, UFA-Unet was trained with only a single exhalation phase (T50) of 4DCT data and evaluated using dynamic phantom XF images with 25-mm amplitude motion. UFA-Unet was compared against U-Net, Attention-Unet, and Swin-Unet. Results: The ablation study confirmed that each component suppressed over-activation to improve segmentation accuracy. In the quantitative study, UFA-Unet maintained superior performance compared with conventional models on both 1 st - and 2 nd -plan DRRs with noise injection. Furthermore, in the phantom study, UFA-Unet demonstrated robust tracking under previously unseen respiratory phases, achieving a 95 th percentile 3D error of 0.61–3.13 mm and consistently outperforming conventional models. Conclusion: UFA-Unet provides accurate, robust, and real-time segmentation, thus demonstrating its suitability for clinical MTT.

BackgroundLower limb rotation at the time of imaging may affect the measurement of joint line angles using plain radiographs, potentially compromising measurement accuracy. Accurate joint line angle assessment is important for orthopaedic surgical planning and limb alignment. This study aimed to investigate changes in distal femoral and proximal tibial joint line angles in response to limb rotation and to evaluate the correlation between these changes using digitally reconstructed radiographs (DRRs) generated from computed tomography (CT) images.MethodsPreoperative CT data from 50 knees scheduled for total knee arthroplasty (TKA) or unicompartmental knee arthroplasty (UKA) at our institution were analysed using TKA planning software. The femur and tibia were aligned perpendicular to their mechanical axes in the coronal and sagittal planes. The surgical epicondylar axis (SEA) and Akagi’s line were used as reference axes for femoral and tibial rotation, respectively, with 0° defined as neutral rotation. Each bone was rotated from 20° external rotation (ER) to 20° internal rotation (IR) in 5° increments using the software, and DRR images were generated at each position. The lateral distal femoral angle (LDFA), medial proximal tibial angle (MPTA), and posterior tibial slope (PTS) were measured at each rotational angle. The absolute values, their variations with rotation, and the correlations between each angle and its respective changes were analysed.ResultsAt 0°, the mean ± standard deviation values were 87.5 ± 2.4°, 83.9 ± 2.6°, and 8.4 ± 3.2° for LDFA, MPTA, and PTS, respectively. The mean changes across the range of 20° ER to 20° IR were 0.3° (1.8°–0.6°), 1.1° (1.0°–1.5°), and 6.2° (1.9°–2.4°) for LDFA, MPTA, and PTS, respectively. MPTA was negatively correlated with ΔPTS (r = -0.7). Based on MPTA values, patients were categorised into three groups: <82.5°, 82.5–84.5°, and > 84.5°. ΔPTS was significantly different between the < 82.5° (8.6 ± 2.3°) and 82.5–84.5° (6.9 ± 1.6°) groups, and between the 82.5–84.5° and > 84.5° (4.0 ± 1.9°) groups (P < 0.05 and P < 0.001, respectively).ConclusionsLimb rotation significantly affects PTS, particularly in cases with greater medial inclination of the tibial plateau. Caution should be exercised when using plain radiographs for preoperative planning or postoperative evaluation.

Medical image registration is an indispensable preprocessing step to align medical images to a common coordinate system before in-depth analysis. The registration precision is critical to the following analysis. In addition to representative image features, the initial pose settings and multiple poses in images will significantly affect the registration precision, which is largely neglected in state-of-the-art works. To address this, the paper proposes a dual-pose medical image registration algorithm based on improved differential evolution. More specifically, the proposed algorithm defines a composite similarity measurement based on contour points and utilizes this measurement to calculate the similarity between frontal-lateral positional DRR (Digitally Reconstructed Radiograph) images and X-ray images. In order to ensure the accuracy of the registration algorithm in particular dimensions, the algorithm implements a dual-pose registration strategy. A PDE (Phased Differential Evolution) algorithm is proposed for iterative optimization, enhancing the optimization algorithm's ability to globally search in low-dimensional space, aiding in the discovery of global optimal solutions. Extensive experimental results demonstrate that the proposed algorithm provides more accurate similarity metrics compared to conventional registration algorithms; the dual-pose registration strategy largely reduces errors in specific dimensions, resulting in reductions of 67.04% and 71.84%, respectively, in rotation and translation errors. Additionally, the algorithm is more suitable for clinical applications due to its lower complexity.

This study compared two deep learning architectures-generative adversarial networks (GANs) and convolutional neural networks combined with implicit neural representations (CNN-INRs)-for generating synthetic CT (sCT) images of the spine from biplanar radiographs. The aim of the study was to identify the most robust and clinically viable approach for this potential intraoperative imaging technique. A spine CT dataset of 216 training and 54 validation cases was used. Digitally reconstructed radiographs (DRRs) served as 2D inputs for training both models under identical conditions for 170 epochs. Evaluation metrics included the Structural Similarity Index Measure (SSIM), peak signal-to-noise ratio (PSNR), and cosine similarity (CS), complemented by qualitative assessments of anatomical fidelity. The GAN model achieved a mean SSIM of 0.932 ± 0.015, PSNR of 19.85 ± 1.40 dB, and CS of 0.671 ± 0.177. The CNN-INR model demonstrated a mean SSIM of 0.921 ± 0.015, PSNR of 21.96 ± 1.20 dB, and CS of 0.707 ± 0.114. Statistical analysis revealed significant differences for SSIM (p = 0.001) and PSNR (p < 0.001), while CS differences were not statistically significant (p = 0.667). Qualitative evaluations consistently favored the GAN model, which produced more anatomically detailed and visually realistic sCT images. This study demonstrated the feasibility of generating spine sCT images from biplanar radiographs using GAN and CNN-INR models. While neither model achieved clinical-grade outputs, the GAN architecture showed greater potential for generating anatomically accurate and visually realistic images. These findings highlight the promise of sCT image generation from biplanar radiographs as an innovative approach to reducing radiation exposure and improving imaging accessibility, with GANs emerging as the more promising avenue for further research and clinical integration.

Patient-specific deep learning tracking for real-time 2D pancreas localisation in kV-guided radiotherapy

Abstract Introduction: The focus of this study was to determine the set-up errors so as to estimate the margin between the Clinical Target Volume (CTV) and the Planning Target Volume (PTV) and to suggest optimum margins for planning target volume (PTV) coverage in thorax cancers. Methods: In the present study data from 51 patients was incorporated. A total of 1308 portal images were examined. Set up errors were estimated by superimposing a digitally reconstructed radiograph (DRR), using an electronic portal image device (EPID) as a reference image. The Medio-Lateral (ML), Cranio-Caudal (CC), and Antero-Posterior (AP) directions were subsequently evaluated. According to the shifts obtained, systematic and random errors were computed. The van Herk formula was employed to determine the values for the clinical-to-planning target volume (CTV-PTV) margins. Results: The systematic error was found to be between 1.0 mm and 1.7 mm, 1.0 mm and 1.8 mm, and 2.1 mm and 3.1 mm along the x, y, and z axis. In the x, y, and z axis, the random error varied from 0.5 mm to 0.7 mm, 0.4 mm to 0.8 mm, and 0.7 mm to 1.7 mm, respectively. Based on the Van Herk equation, the PTV margin following our findings was estimated to be 4.7 mm, 3.3 mm, 8.8 mm for lung, 3.6 mm, 2.7 mm, and 5.7 mm for oesophagus, and 3.0 mm, 4.9 mm, and 8.6 mm for breast in the x, y, and z dimensions respectively. Conclusion: This study demonstrates that an 8.8 mm extension of CTV to PTV margin for the lung, 5.7 mm for the oesophagus, and 8.6 mm for the breast, serving as an upper limit, is sufficient to guarantee that 90% of patients diagnosed with thoracic cancers will receive a cumulative CTV dose that is at least 95% of the prescribed dose.

Soft tissue pathologies and bone defects are not easily visible in intra-operative fluoroscopic images; therefore, we develop an end-to-end MRI-to-fluoroscopic image registration framework, aiming to enhance intra-operative visualization for surgeons during orthopedic procedures. The proposed framework utilizes deep learning to segment MRI scans and generate synthetic CT (sCT) volumes. These sCT volumes are then used to produce digitally reconstructed radiographs (DRRs), enabling 2D/3D registration with intra-operative fluoroscopic images. The framework's performance was validated through simulation and cadaver studies for core decompression (CD) surgery, focusing on the registration accuracy of femur and pelvic regions. The framework achieved a mean translational registration accuracy of 2.4 ± 1.0 mm and rotational accuracy of 1.6 ± for the femoral region in cadaver studies. The method successfully enabled intra-operative visualization of necrotic lesions that were not visible on conventional fluoroscopic images, marking a significant advancement in image guidance for femur and pelvic surgeries. The MRI-to-fluoroscopic registration framework offers a novel approach to image guidance in orthopedic surgeries, exclusively using MRI without the need for CT scans. This approach enhances the visualization of soft tissues and bone defects, reduces radiation exposure, and provides a safer, more effective alternative for intra-operative surgical guidance.

Cone-beam CT (CBCT) is crucial for patient alignment and target verification in radiation therapy (RT). However, for non-coplanar beams, potential collisions between the treatment couch and the on-board imaging system limit the range that the gantry can be rotated. Limited-angle measurements are often insufficient to generate high-quality volumetric images for image-domain registration, therefore limiting the use of CBCT for position verification. An alternative to image-domain registration is to use a few 2D projections acquired by the onboard kV imager to register with the 3D planning CT for patient position verification, which is referred to as 2D-3D registration. The 2D-3D registration involves converting the 3D volume into a set of digitally reconstructed radiographs (DRRs) expected to be comparable to the acquired 2D projections. The domain gap between the generated DRRs and the acquired projections can happen due to the inaccurate geometry modeling in DRR generation and artifacts in the actual acquisitions. We aim to improve the efficiency and accuracy of the challenging 2D-3D registration problem in non-coplanar RT with limited-angle CBCT scans. We designed an accelerated, dataset-free, and patient-specific 2D-3D registration framework based on an implicit neural representation (INR) network and a composite similarity measure. The INR network consists of a lightweight three-layer multilayer perception followed by average pooling to calculate rigid motion parameters, which are used to transform the original 3D volume to the moving position. The Radon transform and imaging specifications at the moving position are used to generate DRRs with higher accuracy. We designed a composite similarity measure consisting of pixel-wise intensity difference and gradient differences between the generated DRRs and acquired projections to further reduce the impact of their domain gap on registration accuracy. We evaluated the proposed method on both simulation data and real phantom data acquired from a Varian TrueBeam machine. Comparisons with a conventional non-deep-learning registration approach and ablation studies on the composite similarity measure were conducted to demonstrate the efficacy of the proposed method. In the simulation data experiments, two X-ray projections of a head-and-neck image with discrepancy were used for the registration. The accuracy of the registration results was evaluated on experiments set up at four different moving positions with ground-truth moving parameters. The proposed method achieved sub-millimeter accuracy in translations and sub-degree accuracy in rotations. In the phantom experiments, a head-and-neck phantom was scanned at three different positions involving couch translations and rotations. We achieved translation errors of and subdegree accuracy for pitch and roll. Experiments on registration using different numbers of projections with varying angle discrepancies demonstrate the improved accuracy and robustness of the proposed method, compared to both the conventional registration approach and the proposed approach without certain components of the composite similarity measure. We proposed a dataset-free lightweight INR-based registration with a composite similarity measure for the challenging 2D-3D registration problem with limited-angle CBCT scans. Comprehensive evaluations of both simulation data and experimental phantom data demonstrated the efficiency, accuracy, and robustness of the proposedmethod.

Background and objectiveAccurate beam data acquisition using three‐dimensional (3D) water tanks is essential for beam commissioning and quality control (QC) in clinical radiation therapy. This study introduces a novel method for quantitative QC of the system, utilizing MV images and webcam videos. The stability of the motor drive speed and the positional accuracy of the fixture were evaluated under two measurement modes: “continuous mode” and “step‐by‐step mode.”MethodsA TRUFIX mounting system (PTW Freiburg Inc., Germany) was used to attach the center of the steel ball to its top, ensuring alignment with the water surface of the tank. To assess deviations from the radiation isocenter, MV images were acquired and compared with digitally reconstructed radiographs (DRRs). These evaluations were performed at different speed settings (slow, medium, and fast) using ET CT Body Marker (BRAINLAB Inc., USA) mounted on the drive unit. A webcam was utilized to capture the images, and custom‐developed tracking software was employed to analyze deviations in driving speed and positional errors.ResultsThe mean error of the radiation isocenter was 0.37 ± 0.09 mm. As the motor drive speed increased, the discrepancy between the set speed and the actual speed observed in the analysis also became larger. In “continuous mode,” the deviation from the displayed value was greater than that observed in “step‐by‐step mode.”ConclusionIt is demonstrated that the proposed analysis method can quantitatively evaluate radiation isocenter misalignment, tank setup position deviation, and both the indicated drive speed values and their stability. At higher drive speeds, the “step‐by‐step mode” showed smaller deviations from the indicated values.

Abstract Objectives: Hip dysplasia is characterized with multiple radiographic measurements which quantify acetabular coverage of the femoral head. The purpose of this study was to use an automated measurement process implemented on digitally reconstructed radiographs (DRRs) to quantify the relationship between acetabular rotations within the coronal, sagittal, and axial planes during a virtual periacetabular osteotomy (PAO) and traditional radiographic metrics of hip coverage. Methods: Virtual PAOs were performed on 3D surface models of 20 patients indicated for PAO. The simulated acetabular PAO fragment was digitally rotated in 2° increments up to 22° in the coronal/sagittal planes to increase lateral and anterior coverage respectively, and 10° in the axial plane to increase acetabular anteversion. AP and false profile DRRs were generated for the preoperative condition and each rotation (n=864 DRRs per hip). A 2D-to-3D projection method was used to project 2D sourcil landmark(s) defined on preoperative DRRs onto the 3D model, track that 3D location during simulated PAO fragment rotation, and then re-project the rotated 3D location onto an updated DRR. The updated sourcil locations combined with landmarks not dependent on the acetabulum were used to automatically calculate LCEA, Tönnis angle, and ACEA associated with the acetabular reorientation. To measure the AWI/PWI, the femoral neck axis (FNA) and a best-fit femoral head circle were manually defined on the preoperative DRR. Intersections between the anterior and posterior acetabular rim on the 3D model and a plane defined by the FNA axis and the DRR source were detected for each simulated PAO rotation. These intersections were then projected back onto the DRR and used with the femoral head radius to calculate the AWI/PWI. Results: One-degree changes in the coronal plane resulted in a 1.04° increase in LCEA and a 0.96° decrease in Tönnis angle. One-degree changes in the sagittal plane resulted in a 0.015 increase in AWI and a 1.05° increase in ACEA, and one-degree changes in the axial plane resulted in a 0.022 decrease in PWI. Conclusion: Knowledge of how rotations in specific anatomic planes affect certain radiographic coverage measurements may assist clinicians with pre-operative planning to correct specific coverage deficiencies.

Introduction.Real-time 2D-kV-triggered images used to evaluate intra-fraction motion during abdominal radiotherapy only provides 2D information with poor soft-tissue contrast. The main goal of this research is to evaluate a novel method that generates synthetic 3D-MRI from single 2D-kV images for online motion monitoring in abdominal radiotherapy.Methods.Deformable image registration (DIR) is performed between one 4D-MRI reference phase and all other phases, and principal-component-analysis (PCA) is implemented on their respective deformation vectors. By sampling 1000 times the PCA eigenvalues and applying the new deformations over a reference CT, 1000 digital reconstructed radiographs (DRRs) were generated to train a convolutional neural network to predict their respective eigenvalues. The method was implemented and tested using a digital phantom (XCAT) and an MRI-compatible phantom (ZEUS) with five DRR angles (0°, 45°, 90°, 135°, 180°). Seven motion scenarios were tested. For model performance, mean absolute error (MAE) and root mean square error (RMSE) were reported. Image quality was evaluated with structure similarity index (SSIM) and normalized RMSE (nRMSE), and target-volume variations were evaluated with volumetric dice coefficient (VDC) and Hausdorff-distance (HD).Results.The model performance across the evaluated angles were MAE(XCAT, ZEUS)= (0.053 ± 0.003, 0.094 ± 0.003), and RMSE(XCAT, ZEUS)= (0.054 ± 0.007, 0.103 ± 0.002). Similarly, SSIM(XCAT, ZEUS)= (0.994 ± 0.001, 0.96 ± 0.02), and nRMSE(XCAT, ZEUS)= (0.13 ± 0.01, 0.17 ± 0.03). For all motion scenarios for XCAT and ZEUS, SSIM were 0.98 ± 0.01 and 0.84 ± 0.02, nRMSE were 0.14 ± 0.01 and 0.27 ± 0.02, VDC were 0.98 ± 0.01 and 0.90 ± 0.01, and HD were 0.24 ± 0.02 mm and 2.3 ± 0.8 mm, respectively, averaged across all angles. Finally, SSIM, nRMSE, VDC and HU values for ZEUS using thedeformedimages as ground truth, presented an improvement of 13%, 28%, 4%, and 76%, respectively.Conclusions. Results from a digital and physical phantom demonstrate a novel approach to generate real-time 3D synthetic MRI from onboard kV images on a conventional LINAC for intra-fraction monitoring in abdominal radiotherapy.

Chest X-ray (CXR) is crucial for diagnosing lung diseases, especially lung nodules. Recent studies indicate that bones, such as ribs and clavicles, obscure 82 to 95% of undiagnosed lung cancers. The development of computer-aided detection (CAD) systems with automated bone suppression is vital to improve detection rates and support early clinical decision-making. Current bone suppression methods face challenges: they often depend on manual subtraction of bone-only images from CXRs, leading to inefficiency and poor generalization; there is significant information loss in data compression within deep convolutional end-to-end architectures; and a balance between model efficiency and accuracy has not been sufficiently achieved in existing research. We introduce a novel end-to-end architecture, the mask-guided model, to address these challenges. Leveraging the Pix2Pix framework, our model enhances computational efficiency by reducing parameter count by 92.5%. It features a rib mask-guided module with a mask encoder and cross-attention mechanism, which provides spatial constraints, reduces information loss during encoder compression, and preserves non-relevant areas. An ablation study evaluates the impact of various factors. The model undergoes initial training on digitally reconstructed radiographs (DRRs) derived from CT projections for bone suppression and is fine-tuned on the JSRT dataset to accelerate convergence. The mask-guided model surpasses previous state-of-the-art methods, showing superior bone suppression performance in terms of structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and processing speed. It achieves an SSIM of 0.99 ± 0.002 and a PSNR of 36.14 ± 1.13 on the JSRT dataset. This study underscores the proposed model's effectiveness compared to existing methods, showcasing its capability to reduce model size and increase accuracy. This makes it well-suited for deployment in affordable, low-power hardware devices across various clinical settings.

The potential use of deep learning in performing autocorrection of setup errors in patients receiving radiotherapy.

Fiducial markers (FM) are essential in prostate robotic stereotactic body radiotherapy (SBRT). Accuray® (Madison, WI, USA) provides an implantation guideline for reliable detection. We report on complication rates and analyze how the geometrical implantation quality correlated with subsequent detection rates. We also investigated whether factors such as single vs. double FM, body mass index (BMI), prostatic gland volume, and implantation-to-treatment interval were predictive for geometry and detection quality. Aretrospective analysis of 64patients receiving transrectal ultrasound (TRUS)-guided transperineal implantation of ≥ 3 prostate FM and robotic SBRT between January2011 and May2021 was performed. Adverse events (AE) were classified according to the Society of Interventional Radiology (SIR) classification system. Digitally reconstructed radiographs (DRR) and the planning CT constituted the basis for implant geometry calculations. Marker detection rates were obtained from the Synchrony® (Accuray®) log. Complication rates were low, with mostly mild AE. Double FM significantly improved the rate of obtaining an optimal implantation geometry. High FM detection rates during treatment could be achieved independent of implantation geometry and type of FM. BMI and prostatic gland volume did not correlate with geometry and detection quality. An implantation-to-treatment interval of > 42days was predictive for lower detection rates. Transperineal intraprostatic FM implantation is asafe procedure. We recommend the use of double markers for reduction of trauma (two punctures instead of four) and, hence, increased patient comfort. Double FM were significantly predictive for achieving an optimal implantation geometry, which was borderline significant for improved marker detection rates over the course of the five-fraction treatment.