Med Phys [JOURNAL]
Sun
200 papers
Image quality restoration in 15-s breath-hold PET using a diffusion-based neural network.
Hu J, Li L, Zhang Q … +8 more , Wei W, Huang G, Liu J, Liang D, Zheng H, Hu Z, He X, Zhang N
Med Phys · 2026 Mar · PMID 41761600 · Publisher ↗
BACKGROUND: Breath-hold PET imaging helps reduce respiratory motion artifacts in thoracoabdominal scans. However, its clinical application is limited by the short acquisition time, which results in significant image nois...
BACKGROUND: Breath-hold PET imaging helps reduce respiratory motion artifacts in thoracoabdominal scans. However, its clinical application is limited by the short acquisition time, which results in significant image noise and poor lesion detectability. Enhancing image quality under such conditions remains a technical challenge. PURPOSE: To improve the image quality of 15-s breath-hold PET scans, we investigated a deep learning-based framework using a diffusion probabilistic model. The goal was to suppress noise and enhance lesion visibility while maintaining quantitative accuracy under severely limited acquisition durations. METHODS: We propose TAM-DiffPET, a denoising diffusion probabilistic model (DDPM) augmented with Temporal Attention Modulation (TAM) to refine intermediate feature representations by injecting diffusion time-step embeddings and temporal contextual cues. The model was trained on paired PET datasets comprising 15-s breath-hold scans and 5-min free-breathing scans from 230 patients at Ren Ji Hospital; 180 cases were used for training and 50 for quantitative and qualitative evaluation. Performance was assessed using PSNR, SSIM, and voxel-wise SUV distributions within lesion ROIs. Visual and statistical comparisons were conducted against U-Net, CycleGAN, and a vanilla DDPM. RESULTS: The proposed method demonstrated superior performance compared to existing deep learning-based approaches. Quantitatively, it achieved the highest PSNR (40.2 ± 1.2 dB) and SSIM (0.995 ± 0.004), significantly outperforming previous deep learning based methods such as U-Net, CycleGAN, and DDPM. Voxel-wise SUV error distributions showed lower standard deviation and mean absolute error within lesion ROIs. Visual assessments revealed enhanced lesion contrast, sharper anatomical boundaries, and reduced background noise. Difference maps confirmed minimal deviation from the 5-min reference scans. Furthermore, SUV distribution analysis across representative patients confirmed that the proposed method preserves tracer uptake consistency, offering improved fidelity in clinical lesion quantification. CONCLUSION: Our diffusion-based framework effectively denoises breath-hold PET images acquired under ultrashort durations, offering improved visual clarity and quantitative fidelity. These results support its clinical utility in motion-prone scenarios, such as thoracic or abdominal imaging, and suggest its potential for enhancing diagnostic accuracy while reducing scan time and radiation burden on patients.
Monte Carlo study on fetal dose assessment for carbon beam craniospinal irradiation during pregnancy.
Choi JW, Yun Y, Lee SH … +6 more , Kim S, Lee C, Han H, Han MC, Yeom YS, Kim JS
Med Phys · 2026 Mar · PMID 41761599 · Publisher ↗
BACKGROUND: Radiotherapy may be suggested for pregnant cancer patients when treatment cannot be delayed, with careful targeting to minimize fetal dose. Considering that a successful case for a pregnant patient treated wi...
BACKGROUND: Radiotherapy may be suggested for pregnant cancer patients when treatment cannot be delayed, with careful targeting to minimize fetal dose. Considering that a successful case for a pregnant patient treated with proton beam craniospinal irradiation (CSI) was reported, carbon therapy can be also considered applicable for the CSI treatment to pregnant patients. PURPOSE: We investigated fetal organ doses from carbon beam CSI during pregnancy by performing Monte Carlo dose calculations. METHODS: We employed the high-quality pregnant female phantom series developed by University of Florida (UF) for eight gestational ages (8, 10, 15, 20, 25, 30, 35, and 38 weeks). The phantoms were converted into DICOM-RT CT images and implemented in a treatment planning system (TPS) with the prescribed relative biological effectiveness (RBE)-weighted dose of 36 Gy. A carbon CSI plan created from TPS was used to perform TOPAS MC dose calculations after commissioned to the carbon beam measurement data. Fetal organ absorbed doses for 28 organs considered radiosensitive were calculated and compared with those estimated for proton therapy in a previous study. RESULTS: The organ/tissue doses for 35 weeks showed the largest variation, ranging from 13.2 mGy (esophagus) to 88.3 mGy (lens), whereas those for 8 weeks showed the smallest variation, ranging from 21.0 mGy (gall bladder) to 30.3 mGy (brain). The fetal whole-body doses decreased from early to mid-gestational ages, but increased again during the later stage as the fetus grew predominantly in the superior direction, reducing the overall distance to the beam field. CONCLUSIONS: Considering the significant attention to carbon therapy, as a first dedicated effort, the result would be clinically informative to estimate the fetal dose from carbon therapy during pregnancy.
Harmonic-mapping-based design of gradient coils on irregular MRI surfaces.
Liu Z, Zhang Q, Du H … +5 more , Liu Q, Chen L, Huang X, Qiu B, Zhou Y
Med Phys · 2026 Mar · PMID 41761598 · Publisher ↗
BACKGROUND: Head-mounted gradient coils for brain imaging need to be designed on irregular surfaces. However, current gradient coil design methodologies often struggle to produce optimal magnetic field solutions for thes...
BACKGROUND: Head-mounted gradient coils for brain imaging need to be designed on irregular surfaces. However, current gradient coil design methodologies often struggle to produce optimal magnetic field solutions for these specialized geometries. The stream function method is typically not applicable to irregular two-dimensional surfaces, while the boundary element method involves excessive variables that are difficult to optimize. PURPOSE: To develop a new gradient coil design method with fewer variables suitable for irregular surfaces, and to complete the design of gradient coils for brain imaging on a semi-ellipsoidal surface. METHODS: Triangular meshes are constructed on the target surface, which is then harmonically mapped to a rectangle. After obtaining the mapping relationship, the connection between the stream function on the rectangle and the coil shape on the original surface is established via the inverse of the harmonic map. Finally, the particle swarm algorithm is used to optimize the basis function coefficients and complete the coil design. RESULTS: The method was used to design and coils on a semi-ellipsoid surface. The resulting coils were manufactured and measured to verify the effectiveness of the method. Both simulation and measurement results show that the and coils designed with this method can generate more linear magnetic fields with roughly the same inductance. The maximum deviation of the magnetic field generated by the Gy coil is 4.02% (simulated) and 15.78% (measured), while that by the compared method is 33.3%. The maximum deviation of the magnetic field generated by the Gz coil is 2.57% (simulated) and 7.20% (measured), while that by the compared method is 27.8%. CONCLUSIONS: This paper provides a new method for coil design on irregular two-dimensional single connected bounded surface, to design gradient coil with better magnetic field with only a few basis functions. The linearity and uniformity of the obtained magnetic field have significant advantages over the boundary element method. Even with manufacturing and measurement errors amplifying the deviation, the measured deviation remains far lower than that of the reference design.
A novel system for micron-scale analysis of energy deposition and response to low-dose radiation.
Pasricha P, McNairn C, Mansour IR … +9 more , Milligan K, Muir BR, Andrews JL, Cassol E, Chauhan V, Subedi S, Jirasek A, Murugkar S, Thomson RM
Med Phys · 2026 Mar · PMID 41757432 · Full text
BACKGROUND: Micrometer-scale evaluation of energy deposition is important for radiation protection and therapy as well as for advancing knowledge of responses to radiation in materials and biological systems. Due to the...
BACKGROUND: Micrometer-scale evaluation of energy deposition is important for radiation protection and therapy as well as for advancing knowledge of responses to radiation in materials and biological systems. Due to the stochastic nature of radiation interactions, there is significant variation in energy deposition in micrometer-sized targets, especially at low doses. This variability underscores the need for a framework for microdosimetry, particularly in low-dose scenarios. PURPOSE: The goal of this work is to develop a novel system for micron-scale characterization of energy deposition and response to radiation that is applicable at low doses, using a combination of Monte Carlo (MC) simulations and experimental techniques. METHODS: EBT3 radiochromic film samples are irradiated to absorbed doses of 0.003-0.5 Gy using the 6-MV beam from a clinical linear accelerator. To quantify energy deposition, MC simulations of the experimental irradiations are conducted to evaluate specific energy deposited within micron-scale voxels in the active layer of the film. To investigate the dose response of the film, the following two methods are employed: (i) flatbed scanner measurement of changes in optical density (OD) of the film, and (ii) Raman spectroscopy (RS) to measure response intensity across doses with micron-scale resolution. Experimental film responses are compared to predictions from the microdosimetric one-hit model. RESULTS: Specific energy distributions obtained from MC simulations show large variation in energy deposition at low doses and within small targets; the "microdosimetric spread" (relative standard deviation) is significantly higher ( 10 times) at 0.003 Gy than at 0.5 Gy, and is observed to decrease with increases in dose and target size. Both RS and OD measurements exhibit a near linear dose-response relationship, reflecting the film's sensitivity across micro- and macroscopic spatial scales. Overall, the OD and RS values determined using the one-hit model with MC-obtained specific energy distributions fit well to experimental measurements, with percentage differences up to 15 and 9.8%, respectively. An initial comparison of the relative standard deviation of RS and OD measurements (corrected for offset signal) shows qualitative agreement with the trends observed for MC-determined microdosimetric spread. CONCLUSION: This study provides first results of a system that combines simulations with experimental techniques to investigate radiation response in micron-scale targets, with a focus on low-dose radiation exposure. The system shows promise in enabling future investigations of energy deposition within small volumes at low doses, where biological responses may be heterogeneous as some cells may receive high energy deposits and incur damage, while others may experience minimal or no deposition.
Complementary fusion network of scaling attention and global attention for volumetric medical image segmentation.
Chen Y, Lu X, Chen H … +2 more , Chen T, Xie Q
Med Phys · 2026 Mar · PMID 41757430 · Publisher ↗
BACKGROUND: Medical image segmentation is crucial in the diagnosis and treatment of diseases. Attention mechanisms have been widely adopted to highlight salient regions while suppressing irrelevant information. Although...
BACKGROUND: Medical image segmentation is crucial in the diagnosis and treatment of diseases. Attention mechanisms have been widely adopted to highlight salient regions while suppressing irrelevant information. Although channel and spatial attention can emphasize a few key features within a limited local range, they struggle to model the global contextual dependencies. In contrast, self-attention in the transformer is capable of modeling such global relationships. Integrating these methods can therefore leverage their complementary strengths for more comprehensive feature representation. PURPOSE: Existing hybrid works integrate the two types of attention mechanisms in a cascading manner, which may disrupt both local key features and global contextual dependencies. This study aims to design a parallel fusion strategy that jointly exploits local and global information to achieve superior segmentation performance. METHODS: We propose a complementary fusion network (CFNet) for volumetric medical image segmentation. To learn local representations, we introduce a novel scaling attention mechanism that redefines both channel and spatial attention. The channel attention employs global average and max pooling to simultaneously capture tissue texture and fine-grained responses. Adaptive convolution is then introduced to efficiently facilitate channel interaction within a local receptive field. The spatial attention uses atrous convolution to enlarge the receptive field and capture rich spatial details. To jointly model both local and global dependencies, we design a parallel mixed module consisting of the proposed scaling attention and transformer-based global attention, achieving continuous and complementary feature learning. RESULTS: We comprehensively evaluated CFNet on four benchmark segmentation tasks, including abdominal multi-organ, cardiac, brain tumor, and left atrium segmentation. Our method achieved Dice coefficients of 87.15%, 92.31%, 86.4%, and 93.91% for the respective tasks. CONCLUSIONS: Experimental results demonstrate that our method outperforms state-of-the-art methods. These superior results highlight the potential of CFNet to support clinical decision-making and treatment planning.
A voxel-wise uncertainty-guided framework for glioma segmentation using spherical projection-based U-Net and localized refinement.
Yang Z, Yang C, Zhang R … +3 more , Wang C, Chen M, Yin FF
Med Phys · 2026 Mar · PMID 41757418 · Full text
BACKGROUND: Accurate segmentation of glioma subregions from multi-parametric MRI (MP-MRI) is critical for clinical management but remains challenging due to tumor heterogeneity and ambiguous tissue boundaries. PURPOSE: T...
BACKGROUND: Accurate segmentation of glioma subregions from multi-parametric MRI (MP-MRI) is critical for clinical management but remains challenging due to tumor heterogeneity and ambiguous tissue boundaries. PURPOSE: This study proposes an uncertainty-guided hybrid segmentation framework that integrates spherical projection-based 2D modeling with localized 3D refinement to improve segmentation fidelity. METHODS: The framework was validated on the BraTS 2020 dataset (N = 369). First, a 2D nnU-Net with spherical projection deformation was employed to generate initial slice-wise predictions. Crucially, prediction variance across multiple spherical projections was utilized to quantify voxel-level uncertainty, highlighting regions of low model confidence. A kernel-based sliding window algorithm then spatially localized 3D subvolumes with high cumulative uncertainty. These targeted regions were subsequently fed into a dedicated 3D nnU-Net for volumetric refinement. Finally, the global 2D predictions and local 3D refinements were adaptively fused using weights optimized via Particle Swarm Optimization. The proposed method was implemented to segment the enhancing tumor (ET), tumor core (TC) and whole tumor (WT). Performance was evaluated against standalone 2D and 3D nnU-Net baselines using the Dice Similarity Coefficient (DSC), HD95, sensitivity, and specificity. RESULTS: The proposed method significantly outperformed 2D and 3D baselines across ET, TC, and WT targets. Notably, it achieved a DSC of 0.8124 for ET (vs. 0.7527 for 2D and 0.7530 for 3D), 0.7499 for TC (vs. 0.7002 for 2D and 0.7027 for 3D), 0.9055 for WT (vs. 0.8989 for 2D and 0.9038 for 3D) and demonstrated consistent gains in HD95 and sensitivity. Quantitative metrics and visualizations confirmed improved spatial coherence and boundary preservation in structurally complex regions. CONCLUSION: By utilizing interpretable uncertainty maps as a spatial attention mechanism, this approach dynamically allocates computational resources to anatomically ambiguous regions. The resulting hybrid framework successfully combines 2D efficiency with 3D contextual accuracy, offering a robust solution for automated glioma segmentation.
Small dose monitor based on silicon-carbide diodes for FLASH radiotherapy.
Lopez Paz I, Fleta C, Henao Á … +2 more , Heinrich S, Guardiola C
Med Phys · 2026 Mar · PMID 41757409 · Full text
BACKGROUND: The FLASH biological effect in radiotherapy has been observed to appear at ultra-high dose rates UHDR ( 40 Gy/s), where the accurate dosimetry at such high rates is still a challenge. PURPOSE: A new 4 4 arr...
BACKGROUND: The FLASH biological effect in radiotherapy has been observed to appear at ultra-high dose rates UHDR ( 40 Gy/s), where the accurate dosimetry at such high rates is still a challenge. PURPOSE: A new 4 4 array of SiC-based detectors (1 mm diameter, 2.2 mm pitch) is proposed for dosimetry in UHDR, as well as the feasibility of a position sensitive technology demonstrator covering 7 7 placed on a movable micro-stage to cover larger surfaces. METHODS: In the ElectronFlash LINAC at the Institute Curie, two silicon carbide prototypes (a 2.2 mm diameter single diode and a 4 4-array of 1 mm diameter with a pitch of 2.2 mm), biased at 0 V, are exposed to a 0.5-5 pulsed electron beam of 7 MeV alongside a flashDiamond PTW as reference dosimeter to characterize their response, time structure and position response. RESULTS: A linearity better than 3.5% is observed up to 10 Gy per pulse of the single diode device only limited by the reference dosimetry. The pulse structure measured is consistent with the reference beam current transformer installed in the LINAC, allowing for instantaneous pulse discrimination at UHDR and its verification in the measurement point. Moreover, results demonstrate the viability of using SiC arrays to quantify the dose per pulse in a 70 50 area with a granularity of 1 2.2 , paving the way to larger arrays and thus toward potential 2D dose monitoring. CONCLUSIONS: The possibility of a position sensitive SiC dose monitor for UHDR is demonstrated, as the technology demonstrator has been proven to maintain good linearity up to at least 10 Gy per pulse, with a time resolution enough to observe microsecond pulses and position sensitive readout.
Real-time Cherenkov imaging will make radiation therapy safer.
Pogue BW, Harms J, Das IJ
Med Phys · 2026 Mar · PMID 41757393 · Publisher ↗
Abstract loading — click title to view on PubMed.
First evaluation of a heteroepitaxial diamond ionization chamber operating at low voltage for diagnostic X-ray dosimetry.
Shinsho K, Koyama K, Hitomi K … +10 more , Nogami M, Maida O, Onodera T, Kikkawa K, Hashimoto S, Hirai Y, Koba Y, Haga A, Maruyama D, Kim SW
Med Phys · 2026 Mar · PMID 41755805 · Full text
BACKGROUND: Heteroepitaxial diamond has recently gained attention as a radiation detector material due to its wide bandgap, radiation hardness, and near-tissue equivalence. Despite these advantages, its use as a solid-st...
BACKGROUND: Heteroepitaxial diamond has recently gained attention as a radiation detector material due to its wide bandgap, radiation hardness, and near-tissue equivalence. Despite these advantages, its use as a solid-state ionization chamber for diagnostic X-ray dosimetry has not yet been established. Demonstrating stable, high-efficiency operation at low voltage would enable compact dosimeters with a very small sensitive volume, which is difficult to achieve with conventional air ionization chambers. PURPOSE: To perform the first characterization of a heteroepitaxial diamond ionization chamber (HED-IC) operated at low bias voltage under diagnostic X-ray conditions and to evaluate its feasibility as a compact, high-efficiency dosimeter. METHODS: A heteroepitaxial diamond detector (4 × 4 × 0.5 mm) with Ti/Au electrodes was fabricated and evaluated using diagnostic X-ray beams at tube voltages from 50 to 120 kV. Charge-collection characteristics, dose linearity, energy dependence, and temporal response were assessed at negative bias voltages with magnitudes between -1 and -100 V. Monte Carlo simulations were performed using PHITS to compute the expected diamond-to-air sensitivity ratio under the same beam qualities for comparison with the experimental measurements. RESULTS: The HED-IC exhibited excellent dose linearity (R > 0.997) and weak energy dependence (< 10%) across effective energies from 28.4 to 40.1 keV. The detector enables dose measurements within a very small sensitive volume, only 1/1250 of that of a typical air ionization chamber. The volume-normalized sensitivity exceeded theoretical expectations, suggesting enhanced effective ionization efficiency. An increased response with higher bias voltage further indicated potential for high-sensitivity operation. CONCLUSIONS: The results demonstrate that the HED-IC can operate as a low-voltage, high-efficiency solid-state ionization chamber under diagnostic X-ray conditions. Owing to the scalability of heteroepitaxial diamond growth, this detector concept provides a promising basis for compact, tissue-equivalent dosimeters capable of real-time dose monitoring across a wide range of radiological applications.
Quantifying uncertainty in calibration and clinical application of EBT-4 Gafchromic film dosimetry.
Lombardi R, Zancopè N, Cavinato S … +4 more , De Zordi N, Giannone A, Scaggion A, Paiusco M
Med Phys · 2026 Mar · PMID 41755802 · Full text
BACKGROUND: GAFchromic films (GAFs) are essential tools in radiotherapy dosimetry, but uncertainties in calibration methods limit their accuracy and hinder consistency across clinical practice. PURPOSE: This study aimed...
BACKGROUND: GAFchromic films (GAFs) are essential tools in radiotherapy dosimetry, but uncertainties in calibration methods limit their accuracy and hinder consistency across clinical practice. PURPOSE: This study aimed to perform a comparative analysis of different calibration functions for GAFchromic EBT4 films in single- and double-channel dosimetry to quantify the uncertainty associated with their calibration and clinical use. METHODS: GAFchromic EBT4 films were irradiated to doses between 0.2 and 10 Gy using a 6 MV photon beam. A custom irradiation setup in a water-equivalent phantom ensured reference conditions, with dose verified by an ionization chamber. Films were scanned on an Epson Expression 13000XL flatbed scanner at 72 dpi, with pre- and postirradiation scans taken at optimized time points (48 h postirradiation to ensure polymer growth stabilization). A custom-developed Python script was used for comprehensive film data processing, including lateral response artifact (LRA) correction. Eight calibration functions, including invertible and non-invertible forms from existing literature, were evaluated and compared. A comparison of the performance of the proposed approach with an established commercial software was performed. RESULTS: Among the tested functions, two of them consistently emerged as the most accurate for EBT4 film calibration, yielding also the lowest relative errors. Specifically, the 2.5th order polynomial function showed an average percentage residual error of 2.7% (ranging from 1.6-4.4%), while the same function on corrected net optical density (netOD) demonstrated a comparable 2.9% (ranging from 1.9-4.4%). In general, calibration functions relying on netOD showed larger uncertainty compared to their analog using the red channel only. The best performing function showed residual values comparable to those obtained with the commercial benchmark. The LRA was observed to cause deviations of up to 5% at scanner extreme lateral positions, reinforcing the necessity of correction for large fields. CONCLUSIONS: We identified an optimal calibration function that incorporates a linear term and a 2.5th-order component to accurately model the relationship between radiation dose and the netOD in the red channel. The significant variability in performance among calibration functions underscores the critical need for independent verification and transparent dosimetry tools. To address this, our study makes the full analysis code and data publicly available, facilitating independent validation and enhancing the accuracy, consistency, and reproducibility of film dosimetry in clinical practice.
Filling of incomplete sinograms from sparse PET detector configurations using a residual U-Net.
Leffler K, Luppino LT, Kuttner S … +2 more , Söderkvist K, Axelsson J
Med Phys · 2026 Mar · PMID 41755757 · Full text
BACKGROUND: Long axial field-of-view PET scanners are becoming increasingly available worldwide for clinical and research nuclear medicine examinations, providing an increased field-of-view and sensitivity compared to tr...
BACKGROUND: Long axial field-of-view PET scanners are becoming increasingly available worldwide for clinical and research nuclear medicine examinations, providing an increased field-of-view and sensitivity compared to traditional PET scanners. However, a significant cost is associated with manufacturing the densely packed photodetectors required for the extended-coverage systems. Despite improved performance allowing ultralow dose or ultrafast scans, the financial barrier remains, limiting clinical utilisation. PURPOSE: To mitigate the cost limitations, alternative sparse system configurations with strategically placed inter-detector gaps have been proposed, allowing an extended field-of-view PET design with detector costs similar to a standard PET system, albeit at the expense of image quality. METHODS: To address the challenges posed by sparse detector configurations, particularly the heavy undersampling of PET measurements, we propose a deep sinogram restoration network to fill in the missing sinogram data. The network, a modified Residual U-Net, is trained end-to-end using standard clinical PET scans from a GE Signa PET/MR. The training involves simulating the removal of 50% of the detectors in chessboard patterns of varying sizes, leading to incomplete sinograms with significant count losses (thus retaining only 25% of all lines of response). RESULTS: The model successfully recovers missing counts in incomplete sinograms, with a mean absolute error consistently below two events per pixel for typical injected radioactivity, outperforming 2D interpolation of incomplete sinograms based on mean absolute error and structural similarity in both sinogram and reconstructed image domain. Notably, the predicted sinograms exhibit a smoothing effect, leading to reconstructed images lacking sharpness in finer details. Despite these limitations, the model demonstrates a substantial capacity for compensating for the undersampling caused by the sparse detector configuration. CONCLUSIONS: This proof-of-concept study suggests that sparse detector configurations, combined with deep learning techniques, offer a viable alternative to conventional PET scanner designs. This approach supports the development of cost-effective, total body PET scanners, allowing a significant step forward in medical imaging technology.
A prescription-free, radiobiology-based framework for automated VMAT planning: A feasibility study in primary prostate cancer radiotherapy.
Kuhn D, Spohn SKB, Zamboglou C … +3 more , Grosu AL, Baltas D, Sachpazidis I
Med Phys · 2026 Mar · PMID 41746195 · Full text
BACKGROUND: Current VMAT planning workflows for prostate cancer primarily depend on conventional dose-volume criteria specified at discrete dose or volume points. These point-based objectives, however, do not necessarily...
BACKGROUND: Current VMAT planning workflows for prostate cancer primarily depend on conventional dose-volume criteria specified at discrete dose or volume points. These point-based objectives, however, do not necessarily lead to globally optimal, patient-specific treatment plans. While radiobiological models such as Tumor Control Probability (TCP) and Normal Tissue Complication Probability (NTCP) can provide more meaningful, individualized targets, previous implementations have either employed these for plan evaluation or integrated biological objectives without providing a comprehensive set of deliverable trade-off plans. To date, no prescription-free, automated VMAT planning method has been introduced that generates clinically deliverable, patient-specific Pareto fronts that are biologically interpretable and useful for radiobiological trade-off analysis. PURPOSE: The purpose of this study was to develop and clinically evaluate a fully automated, prescription-free VMAT planning framework for primary prostate cancer that generates Pareto-optimal, clinically deliverable treatment plans in radiobiological objective space, constrained by predefined TCP and NTCP levels. METHODS: The proposed framework was implemented within a commercial treatment planning system (TPS). 17 patients with unfavorable intermediate-risk prostate cancer were retrospectively selected for evaluation. For each patient, TCP and NTCP levels were predefined for three target volumes and seven organs at risk (OARs), restricting the optimization to clinically meaningful regions of the solution space. Plan optimization was performed using Particle Swarm Optimization (PSO) to iteratively adjust VMAT parameters, with the complication-free tumor control probability (P) serving as the sole objective function. All resulting, clinically deliverable plans were generated in the TPS and subsequently analyzed in the bi-objective radiobiological space defined by injury probability (P) versus one minus the benefit probability (1 - P). The plan yielding the highest P and the corresponding individualized pseudo-Pareto front were identified for each patient. The proposed method was benchmarked against clinical moderately hypofractionated simultaneous integrated boost (SIB) plans. RESULTS: The proposed prescription-independent planning approach successfully generated individualized pseudo-Pareto fronts for all 17 patients in the radiobiological space of P versus (1 - P). This enabled clinicians to visualize and interpret trade-offs between tumor control and normal tissue complication risk within the predefined TCP and NTCP levels. For each patient, the plan with highest P achieved superior predicted tumor control and reduced normal tissue toxicity compared to manually optimized clinical plans. The method effectively individualized dose distributions according to patient-specific anatomy and tumor biology, without reliance on fixed dose prescriptions or conventional constraints. All highest P treatment plans fulfilled the clinical dose requirements. Sensitivity analyses demonstrated robustness of the framework with respect to variations in TCP model parameters. CONCLUSION: This study demonstrated the feasibility of a fully automated, prescription-free VMAT planning framework for primary prostate cancer, indicating its potential for future clinical implementation. The proposed framework directly optimized treatment plans in radiobiological objective space, producing Pareto-optimal, clinically deliverable solutions using predefined TCP and NTCP levels. It enables patient-specific trade-off analysis taking into account tumor control and normal tissue complication risk. The work provides a foundation for further development, including the incorporation of geometric uncertainties, acceleration through parallel or GPU-based computation, and application to additional tumor sites.
Simulation-free spine palliative radiotherapy enabled by AI-adapted diagnostic CT.
Han Y, Hanania AN, Siddiqui ZA … +6 more , Ugarte V, Zhou B, Mohamed ASR, Pathak P, Hamstra DA, Sun B
Med Phys · 2026 Mar · PMID 41746191 · Full text
BACKGROUND: Radiotherapy planning traditionally requires a dedicated simulation CT (sCT), which can introduce delays in initiating treatment. This is particularly impactful in spinal palliative care, where timely treatme...
BACKGROUND: Radiotherapy planning traditionally requires a dedicated simulation CT (sCT), which can introduce delays in initiating treatment. This is particularly impactful in spinal palliative care, where timely treatment is often important for symptom control and prevention of neurological deterioration. Although diagnostic CT (dCT) is frequently available earlier in the workflow, it can lead to geometric and dosimetric inaccuracies when used directly for treatment planning due to discrepancies in patient positioning, vertebral alignment, and table curvature. PURPOSE: To develop and evaluate an AI-based method that transforms dCT into a simulation-equivalent planning CT (AI-pCT), enabling a clinically feasible simulation-free workflow for spinal palliative radiotherapy. METHODS: Two neural networks were trained to correct spine position and body contour using paired dCT-sCT images from 50 patients (42 train/validation, 8 internal tests) in a safety net hospital and externally evaluated on 7 additional academic medical center (AMC) patients. After rigid bone-based alignment to sCT, dosimetric accuracy was assessed by comparing DVH endpoints (Dmean, Dmax, D95, D99, V100, V107) and DVH Root-Mean-Square (RMS) error for plans recalculated on dCT versus AI-pCT versus sCT. Four radiation oncologists scored image suitability. Significance was evaluated using the Wilcoxon signed-rank test. RESULTS: In the safety net cohort, AI-pCT substantially reduced geometric and dosimetric error relative to dCT (e.g., Dmean error 2.0%→0.57%; RMS DVH error 6.4%→2.2%, all p < 0.05), improved physician plan-quality ratings from "Acceptable" to "Good-Perfect," and increased plan-level clinical goal achievement from 37.5% to 100%. In the AMC cohort, where baseline dCT was already closely aligned to sCT, AI-pCT produced smaller but still statistically significant gains. CONCLUSION: AI-pCT achieves sCT-level geometric and dosimetric fidelity without requiring a separate simulation scan, enabling a simulation-free planning workflow for spinal palliative RT. This approach has the potential to reduce treatment delays and improve access, particularly in resource-constrained environments.
Visualization of small vibrations inside an MRI scanner using video motion amplification.
Seo Y, Wang ZJ
Med Phys · 2026 Mar · PMID 41746189 · Full text
BACKGROUND: Diffusion-weighted magnetic resonance imaging (DW-MRI) acquisition requires the application of strong magnetic field gradients, which can induce mechanical vibrations in tissues or phantoms, potentially leadi...
BACKGROUND: Diffusion-weighted magnetic resonance imaging (DW-MRI) acquisition requires the application of strong magnetic field gradients, which can induce mechanical vibrations in tissues or phantoms, potentially leading to signal loss or degradation. A qualitative assessment of these vibrations would be valuable for quality assurance (QA). Conventional methods, such as piezoelectric accelerometers and laser interferometry, have limitations in their applicability and availability. There remains a need for a readily accessible method to detect and characterize these vibrations as part of a robust QA protocol. PURPOSE: The objective of this study is to investigate the feasibility of using motion amplification of high-speed video to assess vibrations during DW-MRI scanning. METHODS: A gel phantom simulating a human head was positioned supine within a head receive coil inside an MRI scanner. A 45-degree angled mirror was placed to visualize the phantom's face, while a high-speed camera, positioned outside the scanner, was used to record videos under two conditions: (1) the scanner in an idle state (still condition), and (2) during a DW-MRI scan (vibration condition). The recorded videos were processed using a motion amplification software tool to enhance subtle movements. The motion of multiple position markers affixed to the phantom was quantitatively analyzed. RESULTS: No motion was visible to the naked eye under either still or scanning conditions. However, motion amplification revealed clear marker displacement during DW-MRI, with substantially smaller movement during scanner idling. Across all facial markers and directions (X (L-R), Y (A-P) and Z (I-S)), median root-mean-square displacement increased from 0.85 µm (range: 0.61-1.50 µm) at idle to 2.59 µm (2.04-4.31 µm) during DW-MRI scanning (b = 2500 s/mm). Similarly, median peak-to-peak displacement rose from 7.71 µm (3.92-10.37 µm) to 18.47 µm (15.22-31.77 µm). CONCLUSIONS: Motion amplification of high-speed video provides a viable method for detecting and analyzing vibrations during MRI scans. This approach could serve as a valuable tool for QA, offering an alternative to conventional vibration assessment techniques.
Robustness testing of MLC optical sensors for use in leaf open time reconstruction and online delivery verification.
Corradini NA, Vite C, Urso P
Med Phys · 2026 Mar · PMID 41746182 · Publisher ↗
BACKGROUND: Online adaptive radiotherapy (OART) requires the patient to remain on the treatment couch, necessitating alternative patient-specific quality assurance (PSQA) solutions to guarantee treatment delivery integri...
BACKGROUND: Online adaptive radiotherapy (OART) requires the patient to remain on the treatment couch, necessitating alternative patient-specific quality assurance (PSQA) solutions to guarantee treatment delivery integrity. Multileaf collimator (MLC) optical sensors (OS) record leaf states on the Radixact linear accelerator and allow for leaf open time (LOT) reconstruction for online delivery verification. Assessment of the quality of the OS data and the OS reconstruction algorithm is important for understanding the reliability of this QA approach. PURPOSE: To assess the robustness of the OS LOT reconstruction method currently available to users on the Radixact platform. METHODS: Raw OS and MVCT detector data were acquired daily for LOT latency curve testing over an 8-month period. More than 8600 datapoints were analyzed for each leaf and the datasets were used to assess OS response stability and OS response uncertainty. An in-house algorithm was developed and used to reconstruct delivered treatments for 55 patients from the raw OS data, which were compared against the vendor's Delivery Analysis (DA) software reconstructions to assess algorithm-based uncertainties. RESULTS: OS LOT reconstructions remained stable when compared to those of the detector; however, statistically significant drift from baseline, which averaged 0.1 ms, was found in approximately 20% of evaluated signals. The standard deviation in LOT reconstruction differences between methods was between 0.5 and 0.6 ms when measuring leaf open times ≥25 ms. LOT reconstruction differences between methods was found to increase with the number of leaves simultaneously opening. A systematic shift of 0.4 ms in LOT method differences was found for the MLC's two leaf banks. In-house OS-reconstructed patient treatments were on average in agreement with the vendor's DA software reconstructions to within ±1 ms for 98.1% of all sinogram LOTs. CONCLUSIONS: Leaf OS signals provide a reliable measurement method for LOT reconstruction on the Radixact platform. The OS LOT reconstructions are a practical solution for online treatment delivery verification as part of the PSQA process for OART workflows. This work provides the basis for future needs in MLC OS QA on the Radixact system as well as evidencing future studies to better understand the relationship between individual LOT reconstruction and delivered dose.
Demographic distribution matching between real-world and virtual phantom population.
Ghosh D, Tushar F, Dahal L … +5 more , Vancoillie L, Lafata KJ, Samei E, Lo JY, Luo S
Med Phys · 2026 Mar · PMID 41746164 · Full text
BACKGROUND: The adoption of virtual imaging trials (VITs) is rapidly expanding, offering a cost-effective and ethically viable alternative to large-scale clinical trials for imaging system evaluation. However, difference...
BACKGROUND: The adoption of virtual imaging trials (VITs) is rapidly expanding, offering a cost-effective and ethically viable alternative to large-scale clinical trials for imaging system evaluation. However, differences in demographic composition between virtual phantom populations and real-world clinical cohorts can introduce bias in imaging performance assessments, particularly for underrepresented populations. Such discrepancies, if unaddressed, can limit the translational relevance of VIT findings by misrepresenting diagnostic performance across diverse patient groups. PURPOSE: To address this limitation, we introduce DISTINCT (Distributional Subsampling for Covariate-Targeted Alignment), a statistical framework for selecting demographically aligned subsamples from large clinical datasets to support robust comparisons with virtual cohorts. METHODS: We applied DISTINCT to the National Lung Screening Trial (NLST) and a companion virtual trial dataset (VLST). The algorithm jointly aligned typical continuous (age, BMI) and categorical (sex, race, ethnicity) variables by constructing multidimensional bins based on discretized covariates. For a given target size, DISTINCT samples individuals to match the joint demographic distribution of the reference population. We evaluated the demographic similarity between VLST and progressively larger NLST subsamples using Wasserstein and Kolmogorov-Smirnov (K-S) distances to identify the maximal subsample size with acceptable alignment. After demographic alignment, we evaluated lung cancer risk prediction performance by applying two established NLST risk scores to the aligned subsamples and assessing their stability with receiver operating characteristic (ROC) analysis. RESULTS: The DISTINCT algorithm identified a maximal demographically aligned NLST subsample of 9974 participants that preserved similarity to the VLST population. To assess whether such aligned subsets were sufficient for downstream applications, we applied two established NLST lung cancer risk scores and evaluated their performance using ROC analysis. Area under the curve (AUC) estimates stabilized once subsample sizes exceeded approximately 6000 participants, demonstrating that moderately sized aligned subsets provide reliable predictive model evaluation. Stratified analyses revealed demographic-specific variations in AUC, underscoring the importance of covariate alignment for fair and representative comparisons. CONCLUSION: DISTINCT provides a statistically rigorous and scalable approach for covariate alignment between real and virtual imaging cohorts based on demographic factors of variability. Although demonstrated for lung cancer screening with low-dose CT, the framework is broadly applicable to other imaging modalities and diseases, and across wide ranges of factors of variability. By enabling fair and representative performance assessments, DISTINCT advances the integration of VITs into imaging research and protocol optimization workflows.
Feasibility and therapeutic effect of neutron spectra with different characteristics based on BNCT for head and lung cancers.
Dai Y, Yang Y, Zhang T … +5 more , Lin X, Jiang Q, Chen T, Ma B, Wang S
Med Phys · 2026 Mar · PMID 41741016 · Publisher ↗
BACKGROUND: Boron neutron capture therapy (BNCT) is widely recognized as an important treatment for malignant brain tumors and melanomas due to its targeting ability and the production of secondary particles with high li...
BACKGROUND: Boron neutron capture therapy (BNCT) is widely recognized as an important treatment for malignant brain tumors and melanomas due to its targeting ability and the production of secondary particles with high linear energy transfer (LET). However, clinical trials for deep-seated tumors are less common than those for superficial tumors. Considering the substantial differences in therapeutic effects that neutron energy spectra produced by different devices may have on tumors, there is a need to explore how these variations impact treatment efficacy across tumor depths. PURPOSE: This study investigates the advantages and disadvantages of neutron spectra with different characteristics for treating tumors at varying depths. The objective is to evaluate how different neutron sources influence therapeutic outcomes, providing insights into their suitability for clinical applications in BNCT. METHODS: Using the Monte Carlo Particle and Heavy Ion Transport code System (PHITS), we studied the neutron spectra generated by three neutron sources: from a nuclear reactor, from a low-energy proton accelerator with a lithium target, and from a high-energy proton accelerator with a beryllium target. Specifically, we compared their therapeutic effects on head tumors at depths of 3 and 6 cm, and a lung tumor at a depth of 9 cm. The simulations assessed dose delivery, critical organ exposure, and treatment parameters to determine the effectiveness of each neutron source. RESULTS: The results reveal that for the treatment of a head tumor at a depth of 3 cm, all three neutron sources delivered the prescribed dose to the tumor while maintaining doses to critical organs within acceptable limits. Notably, the accelerator-based neutron sources offered shorter treatment times compared to the reactor-based source. For a head tumor at a depth of 6 cm, only the neutron source from the low-energy proton accelerator with a lithium target met the treatment requirements without exceeding critical organ doses, although the homogeneity index (HI) value decreased. When treating a lung tumor at a depth of 9 cm, all three neutron sources achieved therapeutic doses to normal lung tissue, but resulted in excessive skin doses. Additionally, analysis of dose-volume histogram (DVH) curves, HI values, and two-dimensional dose distribution maps was conducted, alongside discussions of the required tumor-to-normal tissue (T/N) ratio necessary for effective BNCT of deep-seated tumors. CONCLUSIONS: This study provides a theoretical basis for the optimization and selection of neutron sources for BNCT treatment of tumors at various depths from a dosimetric perspective. The findings offer a reference for the clinical application of BNCT, particularly in addressing the challenges of treating deep-seated tumors, and highlight the need for tailored neutron spectra to enhance therapeutic outcomes.
Self-supervised out-of-distribution detection-Metal implants and other anomaly.
Ramasamy G, Tariq A, Fahrenholtz SJ … +3 more , Sensakovic WF, Patel BN, Banerjee I
Med Phys · 2026 Feb · PMID 41719005 · Publisher ↗
BACKGROUND: Despite the high precision of deep learning models on internal tests on CT, their effectiveness often drops on external validation due to artifacts caused by patient motion and implants like metal or silicone...
BACKGROUND: Despite the high precision of deep learning models on internal tests on CT, their effectiveness often drops on external validation due to artifacts caused by patient motion and implants like metal or silicone that were not accounted for in the carefully curated training data. Potential wide categories of "unknown" anomalies within a CT exam makes training a supervised model for out-of-distribution (OOD) identification impractical, especially when considering unseen external data. PURPOSE: To develop an artificial intelligence (AI) model to detect and identify anomalies/OOD data in abdominal-pelvis CT exams for the purpose of improving the performance of downstream applications. METHODS: Our proposed 2D and 3D generative architecture receives the third lumbar vertebra (L3) slice (slice-level model) or all the slices from the series (series-level model), generates a reconstruction and the secondary part of our architecture-anomaly score computation block, computes the anomalies pixels/voxels (slice-level/series-level) to identify anomalous L3-slices/volumes (slice-level/series-level). We trained on data from over 2850 abdominal-pelvis CT volumes from adults over age 50 years collected throughout multiple Mayo Clinic campuses (60% female; mean age: 66.9, 92.4% non-Hispanic White) and tested on a prospective test set of 544 CTs from July 2024 (47.3% female; mean age: 70.9, 94% non-Hispanic White) as well as an external test set. RESULTS: We found that while traditional methods show moderate success, our generative models-Vector Quantized Variational Autoencoder (VQVAE) and Vision Transformer-Masked Autoencoder (VIT-MAE)-deliver excellent results with negligible false positives (FPs) and are also superior in identifying varied types of OOD samples. Prospective analysis showed the model was able to handle the under-documentation of anomaly in radiology reports with 86.11% true positive (TP) rate. We also performed external validation using the publicly available AbdominalCT-1k dataset, which contains 1062 CT scans compiled from several existing benchmark datasets. The model achieved a 75.26% TP rate, while the 24.7% FP rate was primarily triggered by anomalies located outside the body. CONCLUSIONS: The proposed method can be leveraged to detect both intra- and interclass OOD data from abdominal CT images and can assess the quality of CT datasets to provide actionable insights. This workflow is particularly valuable for nonshareable healthcare collaborations, where it can be deployed as a service within local firewalls for automated dataset curation without prior knowledge about the OOD types. The implementation of the algorithm is available in the GitHub: https://github.com/gokul-ramasamy/implant_detection.git.
Enhancing dose conformity in head and neck intensity-modulated proton therapy using a novel dynamic multi-leaf collimator strategy.
Wakisaka Y, Tominaga Y, Miyasaka Y … +5 more , Rahimi R, Furutani KM, Nakata M, Iwai T, Nishio T
Med Phys · 2026 Feb · PMID 41712258 · Full text
BACKGROUND: Proton pencil beam scanning (PBS) enables highly conformal dose distributions; however, its lateral dose fall-off (penumbra) can be compromised by the use of range shifters (RSs) and increased air gaps. In PB...
BACKGROUND: Proton pencil beam scanning (PBS) enables highly conformal dose distributions; however, its lateral dose fall-off (penumbra) can be compromised by the use of range shifters (RSs) and increased air gaps. In PBS for head and neck regions, where critical organs at risk (OARs) are frequently adjacent to the target, penumbra degradation may lead to increased OAR doses or suboptimal target coverage. The integration of a dynamic multi-leaf collimator (dMLC), which adjusts leaf positions at each energy layer, has been shown to improve dose conformity in single-field uniform dose (SFUD) delivery. In parallel, intensity-modulated proton therapy (IMPT) offers enhanced dose shaping over SFUD by modulating beam intensity across multiple fields and does not require a single beam to encompass the entire target volume, providing greater flexibility in utilizing dMLC capabilities. PURPOSE: This study integrates dMLC into IMPT for head and neck cancer and proposes a novel leaf positioning strategy. We evaluate the dosimetric impact of this approach and assess its potential clinical benefits in terms of target coverage and OAR sparing. METHODS: Treatment plans were retrospectively created for five patients with head and neck cancer. For each patient, IMPT plans with three beam angles were generated using three techniques: (1) uncollimated PBS, (2) dMLC, in which the MLC encloses the target cross-section at each energy layer, and (3) dMLC, in which the MLC actively blocks OARs and their distal regions. Dose-volume histogram (DVH) metrics for the clinical target volume (CTV) and OARs were evaluated, including a total of 21 perturbed scenarios that combined ± 2 mm setup uncertainties (7 scenarios) and ± 3.5% range uncertainties (3 scenarios). The accuracy of dose calculations was validated by comparing calculated and measured lateral dose distributions at the isocenter plane in water using two-dimensional gamma analysis with a 2%/2 mm criterion. RESULTS: The integration of dMLC with IMPT significantly reduced the dose to surrounding OARs while maintaining comparable target coverage and robustness relative to uncollimated PBS. Notably, dMLC demonstrated an enhanced dose-sparing effect than dMLC, particularly for OARs surrounded by the target. While maintaining comparable CTV D98% across three techniques, dMLC achieved the greatest reduction in the mean doses to the eyeballs and optic nerves, as well as in the D2% to the optic chiasm, brain, and brainstem in most cases. The gamma passing rate between calculated and measured doses for dMLC exceeded 95% for all beams, confirming the accuracy of dose calculations involving complex leaf positions. CONCLUSIONS: The combination of IMPT and dMLC provides notable dosimetric advantages, supporting its potential for clinical applications. Further validation across a broader range of cases is necessary to comprehensively assess its efficacy and safety, particularly with respect to leaf positioning accuracy and potential variations in biological effectiveness.
On the degrees of freedom of gridded control points in learning-based medical image registration.
Yan W, Yang Q, Wang Y … +5 more , Punwani S, Emberton M, Stavrinides V, Hu Y, Barratt D
Med Phys · 2026 Feb · PMID 41712256 · Full text
BACKGROUND: Many registration problems are ill-posed in homogeneous/noisy regions, and dense voxel-wise decoders can be unnecessarily high-dimensional. A sparse control-point parameterization provides a compact, smooth d...
BACKGROUND: Many registration problems are ill-posed in homogeneous/noisy regions, and dense voxel-wise decoders can be unnecessarily high-dimensional. A sparse control-point parameterization provides a compact, smooth deformation representation while reducing memory and improving stability. PURPOSE: This work investigates the required control points for learning-based registration network development. In particular, as sparse as control points are configured and compared with alternative approaches, including those using scattered control points and displacements sampled at every voxel, that is, dense displacement fields. METHOD: We present GridReg, a learning-based registration frametwork that replaces dense voxel-wise decoding with displacement predictions at a sparse grid of control points. This design substantially cuts the parameter count and memory while retaining registration accuracy. Multiscale 3D encoder feature maps are flattened into a 1D token sequence with positional encoding to retain spatial context. The model then predicts a sparse gridded deformation field using a cross-attention module: Each control point attends to encoder tokens within its local grid neighborhood to estimate its displacement, which is subsequently interpolated to a dense field. We further introduce grid-adaptive training, enabling an adaptive model to operate at multiple grid sizes at inference without retraining. RESULTS: This work quantitatively demonstrates the benefits of using sparse grids. Using three data sets for registering prostate gland, pelvic organs and neurological structures, the experimental results suggest a much improved computational efficiency, due to the prediction of sparse-grid-sampled displacements. Alternatively, the superior registration performance was obtained using the proposed approach, with the similiar or less compute cost, compared with existing algorithms that predict DDFs (e.g., VoxelMorph/TransMorph) or displacements sampled on scattered key points (KeyMorph). CONCLUSION: We conclude that predicting sparsely gridded displacements provides reduced computational cost and/or improved performance, independent of the encoder architecture, and can be readily implemented. Therefore, GridReg should potentially be considered for many registration tasks with adaptive grid sizes. The code is available via git@github.com:yanwenCi/GridReg.git.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe