Photon counting detector CT (PCD-CT) has the potential to sort detected x-ray photons into more than two energy bins to perform multi-material decomposition using K-edge materials without additional mathematical constraints/assumptions. The availability of multiple energy thresholds in PCD-CT opens the possibility for unique clinical applications such as simultaneous multi-contrast-agent imaging. To evaluate the technical performance of a PCD-CT system with four energy thresholds for conducting multi-material decomposition of iodine, gadolinium, and water after a comprehensive system calibration procedure. Experiments were performed on a commercial PCD-CT scanner (NAEOTOM Alpha, Siemens Healthineers) using a research mode with four energy thresholds. Material calibration was performed by scanning contrast materials with known concentration placed in various sizes of cylindrical water-equivalent phantoms at various tube currents and positions. From these data, per-voxel contrast models were derived for each material by multiple linear regressions up to second order, where the predictor variables were combinations of tube current, phantom diameter, and distance to isocenter. After calibration, small (20 cm diameter) and large (30cm×40cm) uniform multi-energy CT phantoms and an anthropomorphic thorax phantom (31cm×39cm), each containing bone, soft tissue, and several quantitative inserts of iodine (0.2-5.0mg/ml), gadolinium (0.3-5.0mg/ml), and a mixture of both (5.0mg/ml each) were scanned using the four-energy threshold mode at 140 kV with energy thresholds set at 20, 52, 75, and 82 keV. Images for each energy threshold were reconstructed with a filtered back-projection algorithm and a quantitative kernel (Qr44). Subsequent image-based material decomposition was performed on individual voxels of the energy threshold images using the models derived from the calibration process to generate water, iodine, and gadolinium basis images. Quantitative material concentrations and image noise magnitudes, and noise power spectra (NPS) were measured. Iodine and gadolinium concentrations above 1.25 mg/ml were visible in the material basis images in all phantoms. Averaged quantitative accuracy for iodine and gadolinium in the small, large, and thorax phantoms were 0.39±0.30 mg/ml, 0.96±0.53 mg/ml, and 0.53±0.35 mg/ml, respectively. In the water basis images, accuracy was within 24 HU in all phantoms. Noise magnitudes in iodine and gadolinium basis images were 2.91 and 3.52 times that of the lowest energy threshold image for the small phantom, and were 6.92 and 7.17 times that of the lowest threshold image for the large phantom. Noise texture did not change between the energy threshold and material basis images for either size phantom as measured by the NPS. After a comprehensive system calibration, a commercial PCD-CT with four energy thresholds performed multiple material decomposition of iodine, gadolinium, and water for concentrations above 1.25 mg/ml in human-sized phantoms. Noise amplification and quantitative accuracy from the decomposition greatly depended on object size.

BackgroundCarbon ion radiotherapy (CIRT) offers higher linear energy transfer (LET) and superior relative biological effectiveness, making it a promising option for treating hypoxic, radioresistant tumors. Spot‐scanning Hadron Arc (SHArc) therapy enables dose‐averaged LET (LETd) escalation in the tumor but increases planning and setup complexity. Dynamic delivery remains impractical for gantry‐based carbon ion arc therapy due to the system's large size and complex control requirements. Step‐and‐shoot delivery, while less efficient, provides a feasible alternative and represents a key step towards clinical SHArc implementation.PurposeThis work establishes the first step‐and‐shoot planning and delivery technique for carbon ion arc therapy (static SHArc) at the Heidelberg Ion‐beam Therapy Center using the gantry system. Static SHArc therapy is evaluated in terms of plan quality and delivery feasibility for pancreatic cancer.MethodsStatic SHArc plans were optimized for seven pancreatic cancer cases, considering 20 gantry angles. Two distinct energy layer (EL) selection techniques were investigated: (1) Central EL, selecting the seven central ELs per beam, and (2) MU‐Based EL, prioritizing ELs contributing the highest monitor units (MU). LETd optimization was performed to escalate the minimum LETd to ∼50–80 keV/µm within the gross tumor volume. Static SHArc plans were compared against conventional IMPT using two single‐field optimized posterior oblique beams (2‐SFO), in terms of dose conformality, LETd, and robustness against setup (5 mm) and range (1.5%) uncertainties. Inter‐fractional robustness was assessed via forward dose calculation on daily control CT scans. Dosimetric validation and delivery time verification for static SHArc using the heavy ion gantry system were conducted via end‐to‐end testing with ion chamber and film measurements in a cylindrical PMMA phantom.ResultsStatic SHArc plans improved LETd distributions in the tumor without compromising target coverage and clinical OAR constraints. The MU‐based EL method increased minimum target dose, whereas Central EL enabled higher LETd concentration in the tumor center. Both energy selection methods for static SHArc exhibited reduced inter‐fractional robustness compared to the two‐SFO. Dosimetric verification showed deviations < 3% and total delivery time was ∼27 min.ConclusionsThis study investigates static SHArc, a step‐and‐shoot approach for delivering carbon ion arc therapy. While static SHArc can provide dosimetric advantages, particularly in terms of LETd distribution, EL selection plays a key role. Improving inter‐fraction robustness remains crucial for clinical implementation.

Monte Carlo (MC)-based estimation of dose-averaged linear energy transfer (LETd) in proton therapy is accurate but computationally intensive. Although deep learning (DL) models offer efficient alternatives, few have been optimized for multi-source inputs or validated on anatomically complex tumors, such as nasopharyngeal carcinoma (NPC). This study developed and evaluated a multi-source DL framework for voxel-level LETd prediction in proton therapy for NPC, aiming to identify optimal input configurations and assess both prediction accuracy and clinical reliability. This study analyzed 100 patients with NPC, yielding 500 proton therapy fields. Ground-truth LETd maps were generated using GPU-based Monte Carlo simulations. A deep learning model was trained with different combinations of four input features: dose distribution, CT image, beam mask, and radiological depth, resulting in eight configurations (Mdose, Mdose+ct, Mdose+beam, Mdose+depth, Mdose+beam+depth, Mdose+ct+depth, Mdose+ct+beam, and Mdose+ct+beam+depth). The dataset was randomly divided into training (300 fields, 60 patients), validation (100 fields, 20 patients), and testing (100 fields, 20 patients) sets. Model performance was evaluated using mean absolute error (MAE), peak signal-to-noise ratio (PSNR), normalized cross-correlation (NCC), structural similarity index measure (SSIM), and 3D Gamma analysis. A 3D U-Net baseline was retrained under identical conditions for comparison. Model interpretability was examined with Gradient-weighted Class Activation Mapping (Grad-CAM), and prediction confidence was assessed using uncertainty maps. The Mdose+ct model achieved the best performance (MAE 0.21±0.03keV/µm, PSNR 28.99±1.33, NCC 0.98±0.01, SSIM 0.971±0.007, Gamma 3%/3mm: 94.69±3.69%), outperforming the 3D U-Net baseline (MAE 0.38±0.52keV/µm). Grad-CAM visualizations confirmed that the model focused on physically relevant regions, while uncertainty maps highlighted high-variability zones at LETd transitions. The model generated voxel-level LETd predictions in approximately 7 s per case, which was substantially faster than GPU-based MC simulations (30 s per case). The proposed multi-source deep learning framework enables rapid and accurate voxel-level LETd prediction for NPC. Combining dose with CT yielded the most reliable input configuration, and incorporating interpretability and uncertainty analyses enhances confidence in its suitability for biologically guided proton therapy planning.

Denoising low dose computed tomography (CT) images can have great advantages for the aim of minimizing the radiation risk of the patients, as it can help lower the effective dose to the patient while providing constant image quality. In recent years, deep denoising methods became a popular way to accomplish this task. Conventional deep denoising algorithms, however, cannot handle the correlation between neighboring pixels or voxels very well, because the noise structure in CT is a resultant of the global attenuation properties of the patient and because the receptive field of most denoising approaches is rathersmall. The purpose of this study is to improve existing denoising networks, by providing them additional information about the imagenoise. We here propose to generate additional noise realizations by simulation, reconstruct them, and use these noise images as additional input into existing denoising networks. This noise augmentation is intended to guide the denoising process. The additional noise realizations are not only required during training, but also during inference. The rationale behind this noise-augmented deep denoising (NADD) is that CT image noise is strongly patient-specific and it is non-local since it depends on the attenuation of X-ray beams. NADD is architecture-agnostic and can thus be used to improve any previously proposed method. We demonstrate NADD using existing denoising networks that we slightly modified in their input layer in order to take the CT image that is to be denoised plus additional noise images as input. To do so, we modified three popular denoising networks, the CNN10, the ResNet, and the WGAN-VGG and apply them to clinical cases with 90% dose reduction. In all cases tested, the denoising networks strongly benefit from the noise augmentation. Noise artifacts that are being misinterpreted by the original networks as being anatomical structures, are correctly removed by the NADD version of the same networks. The more noise images are provided, the better the performance. Providing additional simulated noise realizations helps to significantly improve the performance of CT image denoisingnetworks.

BackgroundTumor heterogeneity presents significant challenges in breast cancer diagnosis and treatment. Four‐dimensional dynamic contrast‐enhanced dedicated breast CT (4D DCE‐bCT) is a novel imaging technique designed to capture contrast agent kinetics with high spatial and temporal resolution, enabling detailed assessment of tumor perfusion and heterogeneity.PurposeTo develop modular breast tumor phantoms capable of simulating a range of physiologically relevant perfusion patterns and structural heterogeneities for validation of 4D DCE‐bCT.MethodsTumor phantoms (1.5 cm diameter) were 3D‐printed using clear resin in several designs: small‐channel phantoms (0.8 and 1.0 mm diameter), a leaking vessel model with permeable outer walls, gyroid structures (1.3 and 1.5 mm pores) mimicking microvascularization, and a dual‐input/output model to replicate heterogeneous perfusion. The phantoms were integrated into a programmable flow system, enabling iodinated contrast (up to 5 mg I/mL) delivery with custom profiles: full wash‐in/wash‐out, persistent, plateau, and partial wash‐out. 4D DCE‐bCT acquisitions with a 65 kV + 0.25 mm Cu filter involved 1 pre‐contrast scan (360 pulses over a 10‐second revolution at 80 mA) followed by three post‐contrast phases (400 pulses over 10 revolutions at 32 mA). Images were reconstructed using 40 projections at 5‐second intervals using prior image constrained compressed sensing (PICCS). Time–intensity curves (TICs) were analyzed in volumes of interest in various sections of the tumor phantoms. A gamma variate function was fitted to each TIC, and the corresponding fit parameters and coefficients of determination (R2) were extracted.ResultsDynamic imaging demonstrated successful capture of expected contrast kinetics. Larger channels (1.0 mm) produced 2.5 times greater enhancement compared to smaller ones (0.8 mm). The R2 values for the 0.8 mm and leaking channel were lower, 0.54 and 0.67 versus 0.85 in the 1 mm channel, due to a higher noise level in the signal. The leaking vessel phantom exhibited delayed wash‐out by ∼30 s. Distinct flow patterns were evident in the dual‐input/output model. The gamma variate parameters showed the expected trends due to the programmed flow patterns.ConclusionThe developed 3D‐printed tumor phantoms effectively simulate key perfusion features and point towards the feasibility of using 4D DCE‐bCT to image detailed perfusion patterns.

A growing number of cancer patients with brain metastases can benefit from stereotactic radiosurgery (SRS) thanks to recent advances in systemic therapies which have led to improved survival. Meanwhile, selection criteria for SRS treatments are evolving to include patients with increasingly many metastases. With an increasing patient load, single-isocenter treatments on widely available C-arm linear accelerators are an attractive option. However, the planning of such treatments is challenging for multi-target cases due to the island blocking problem, which occurs when the multi-leaf collimator cannot conform to all targetssimultaneously. We propose a multi-target partitioning algorithm that mitigates excessive exposure of normal tissue caused by the island blockingproblem. The proposed algorithm considers an initial set of arc trajectories and divides (partitions) the set of targets per trajectory into smaller subsets to treat with separate back-and-forth arc passes, simultaneously optimizing both the target subsets and collimator angles to minimize island blocking. We incorporated this algorithm into a fully automated treatment planning script and evaluated it on 20 simulated patient cases, each with 10 brain metastases and 21Gy prescriptions. For each case, the script generated a series of volumetric modulated arc therapy (VMAT) plans with increasingly many arcs along the three trajectories. Each such plan was compared to four baseline plans generated with alternative heuristics for distributing targets across arcs. We also evaluated the algorithm retrospectively on six clinicalcases. Partitioning significantly improved the gradient index (GI), global efficiency index ( ) and brain compared to simultaneous treatment of all metastases. For example, the average GI improved from 5.9 to 3.3, from 0.32 to 0.46, and normal brain from 49 to 26 between 3 and 9 arcs. The baseline plans improved similarly, but the proposed algorithm was significantly better at utilizing a limited budget of arcs. All target partitioning strategies increased the total number of monitor units (MUs). The dose gradient in single-isocenter VMAT plans can be substantially improved by treating a smaller subset of metastases at a time. This requires more MUs and arcs, implying a trade-off between delivery time and plan quality, which can be explored using the algorithm proposed in thispaper.

Deep learning-based segmentation methods for optical coherence tomography (OCT) have demonstrated outstanding performance. However, the stochastic distribution of training data and the inherent limitations of deep neural networks introduce uncertainty into the segmentation process. Accurately estimating this uncertainty is essential for generating reliable confidence assessments and improving model predictions. To address these challenges, we propose a novel uncertainty-guided cross-layer fusion network (UGCFNet) for retinal OCT segmentation. UGCFNet integrates uncertainty quantification into the training process of deep neural networks and leverages this uncertainty to enhance segmentation accuracy. Our model employs an encoder-decoder architecture that quantitatively assesses uncertainty at multiple stages, directing the network's focus toward regions with higher uncertainty. By facilitating cross-layer feature fusion, UGCFNet enhances the comprehensive understanding of both semantic information and morphological details. Additionally, we incorporate an improved Bayesian neural network loss function alongside an uncertainty-aware loss function, enabling the network to effectively utilize these mechanisms for better uncertainty modeling. We conducted extensive experiments on the publicly available AI-Challenger and OIMHS OCT segmentation datasets. The training, validation, and testing sets of the AI-Challenger dataset are comprised of 32, 8, and 43 OCT volumes, yielding a total of 4096, 1024, and 5504 B-scans, respectively. The training, validation, and testing sets of the OIMHS dataset consist of 100, 25, and 25 OCT volumes, resulting in 2,310, 798, and 751 B-scans, respectively. The results demonstrate that UGCFNet achieves state-of-the-art performance, with average Dice similarity coefficients of 79.47% and 93.22% on the respective datasets. Our proposed UGCFNet significantly advances retinal OCT segmentation by integrating uncertainty guidance and cross-level feature fusion, offering more reliable and accurate segmentation outcomes.

BackgroundRadiotherapy workflows conventionally deliver one treatment plan multiple times throughout the treatment course. Non‐coplanar techniques with beam angle optimization or dosimetrically optimized pathfinding (DOP) exploit additional degrees of freedom to improve spatial conformality of the dose distribution compared to widely used techniques like volumetric‐modulated arc therapy (VMAT). The temporal dimension of dose delivery can be exploited using multiple plans (sub‐plans) within one treatment course. For instance, temporally feathered radiation therapy (TFRT) uses iso‐curative sub‐plans to deliver an alternance of higher and lower doses compared to a single plan to selected organs‐at‐risk (OARs), facilitating the dynamic recovery process of healthy tissues.PurposeThis study presents a simultaneous optimization framework based on direct aperture optimization with or without DOP to optimize multiple coplanar or non‐coplanar sub‐plans within one treatment course and demonstrates its use for TFRT planning.MethodsThe goal of the framework was to minimize an objective function consisting of weighted upper or lower dose‐volume, generalized equivalent uniform dose, and normal tissue objectives set on the dose distribution of each sub‐plan or the combined total plan. Reference VMAT and dynamic collimator trajectory radiotherapy (colli‐DTRT) single plans were created and used to derive the objectives for TFRT planning. The TFRT high‐to‐low dose modulation was integrated into the objective list and systematically investigated for a digital academic phantom using three variations (“soft”, “medium”, “hard”). Additionally, a “super‐soft” variation used the same objectives for all five sub‐plans (i.e., no high‐to‐low dose modulation). Furthermore, “medium” TFRT sub‐plans with colli‐DTRT were created for three more complex clinically motivated head and neck cases.ResultsFor the phantom, the sub‐plans were iso‐curative with target D98% within 1.6% of the reference plans. High‐to‐low OAR dose modulation was achieved with median Dmean differences between high and low dose fractions of 2.7% of the prescription dose (soft), 3.3% (medium), and 4.4% (hard) for VMAT. Median OAR Dmean differences were 2.8% of the prescription dose (soft), 6.3% (medium), and 6.1% (hard) for colli‐DTRT. The dose distributions of the total plans had higher homogeneity indices (HI = D98%/D2%) compared to the reference plans. Lower OAR Dmean were achieved for the soft, medium, and hard TFRT variation in the total plans compared to the reference plan. However, in the super‐soft variation only five of the 10 feathered OARs showed this reduction.For the three clinically motivated cases with colli‐DTRT, median OAR Dmean differences between the high and low dose fractions were 3.5%, 4.2%, and 7.5% of the prescription dose. The total plans’ dose distributions had higher HIs compared to their respective reference plans and lower or equal Dmean for all feathered OARs except for pharynx in one case.ConclusionA framework for the simultaneous optimization of multiple coplanar or non‐coplanar sub‐plans within one treatment course was developed. Simultaneous optimization was investigated with a phantom and three clinically motivated cases for TFRT planning. OAR dose modulation in the sub‐plans was achieved while increasing target homogeneity and reducing OAR doses in the dose distribution of the total plans compared to the reference plans.

Online adaptive radiation therapy (ART) offers a paradigm shift in radiotherapy by enabling adjustments to the planned dose based on daily anatomical variation. In the context of cone-beam computed tomography (CBCT) for online ART on a standard linac, thoracic and abdominal treatment sites in particular present unique challenges due to the typically large treatment volumes, mobile anatomy, scatter-induced image quality degradation, and hounsfield unit (HU) limitations. A recent hardware and software upgrade for a standard linac, Varian TrueBeam (TB) v4.1 HyperSight, seeks to overcome these challenges through implementation of a larger kV imager panel (i.e., 43×43 cm), increased gantry speed (i.e., from 6 to 9°/s), and improved HU accuracy. However, investigation of the new upgrade is essential to harness the full potential of these advancements. We report on physical characterization and a digital Monte Carlo (MC) model of the new imaging system hardware. The open-source GEANT4 Application for Tomographic Emission (GATE) MC toolkit, which allows scintillation systems, including CBCT, to be accurately modeled, was utilized. All physical components of the new TB upgrade were modeled from vendor-provided geometry and material specifications. The model was validated using physical measurements acquired on the upgraded system. Specifically, the modulation transfer function (MTF), noise power spectrum (NPS), profiles across the physically larger detector, scatter-to-primary ratio (SPR), and loss in spatial resolution as a function of the increased gantry speed and an object's distance from isocenter. The latter was quantified using the pixel distance between the 15% and 85% intensities of the over-sampled edge spread function (ESF) for source-to-edge-phantom distances (SEPDs) of 80, 100, and 120 cm. Focal spot motion was also characterized by the MTF at SEPD of 100 cm. The MTF50 was 0.901 and 0.889 mm-1 for the measurement and simulation, respectively, for a 125 kVp beam. The normalized root mean square error (nRMSE) was 0.013. While small, the model displayed degraded spatial resolution accuracy for other beam qualities. The general trend of the physically measured normalized noise power spectrum (nNPS) curves was reproduced with the model at all beam energies; however, a small systematic offset was observed. Excellent agreement was observed between central-image x- and y-profiles of measured and MC-generated projections, indicating correct modeling of the larger imaging detector. Simulated SPRs closely agreed with those of measurement. The measured ESF widths for the 6 and 9°/s acquisitions were both 2.5 pixels, indicating no reduction in spatial resolution in the projection space as a result of the increased acquisition speed. The effect of focal spot blurring due to increased gantry speed was accurately modeled, considering ESF width differences no larger than 0.5 pixels were observed for all SEPDs. Decreased spatial resolution in projection images was observed for SEPDs of 80 and 120 cm compared to 100 cm. The MC model of the novel TB 4.1 HyperSight upgrade accurately reproduced physical measurements acquired on the new system. The model will be used alongside physical testing on the new TB platform, working towards an online CBCT-based ART protocol for thoracic and abdominal treatments.

Understanding respiratory motions of liver and its surrogate organs is crucial for precise dose delivery in liver cancer radiotherapy. Although these motions have been studied for respiratory motion management in the supine posture, few studies have quantified them and evaluated their correlations in the upright posture. This study quantified the respiratory motions of liver and surrogate organs and evaluated the correlations between the liver motions and surrogate signals for respiratory motion monitoring in both the supine and upright postures. 4D dynamic volume computed tomography (CT) images were acquired from 13 asymptomatic volunteers in both supine and upright postures while free breathing. The 4D liver motions were calculated by performing deformable image registration. Subsequently, superior-inferior (SI) motion at the right diaphragm apex and anterior-posterior (AP) motion at the abdominal skin surface were obtained as surrogate signals from the 4D CT images. The average displacements and maximum magnitudes of liver motions and surrogate signals were compared between the postures. Moreover, the correlations between the liver motions and surrogate signals were evaluated based on correlation models. The average liver displacements from the end-exhalation along the SI, AP, and left-right directions were -6.1, 2.1, and -0.4 mm in the supine posture and -4.6, 2.5, and 0.1 mm in the upright posture, respectively. The maximum motion range along the SI direction significantly decreased from 16.3 ± 6.6 mm in the supine posture to 11.0 ± 2.7 mm in the upright posture. Similar to the supine posture, the correlation models also reproduced the liver motions with high fitting accuracy in the upright posture. Compared to the conventional supine posture, the liver SI motion became smaller in the upright posture while keeping evident correlations with the surrogate signals, which suggests the upright posture will allow precise dose delivery with reduced internal target margins.