Med Phys [JOURNAL]
Sun
200 papers
Monte Carlo-based optimization of CTV-to-PTV margins for image-guided VMAT prostate radiotherapy.
Chen Q, Ammar A, Yu NY … +3 more , Zhu L, Rong Y, Vargas CE
Med Phys · 2026 Apr · PMID 41912350 · Publisher ↗
BACKGROUND: The van Herk margin formula, derived for 3-D conformal radiotherapy with a uniform 3.2 mm dose penumbra and no intrinsic dose buffer, remains widely used to design CTV-to-PTV expansions in contemporary image-...
BACKGROUND: The van Herk margin formula, derived for 3-D conformal radiotherapy with a uniform 3.2 mm dose penumbra and no intrinsic dose buffer, remains widely used to design CTV-to-PTV expansions in contemporary image-guided prostate radiotherapy. However, coplanar VMAT techniques often feature broader penumbrae and explicit PTV-isodose clearances, potentially violating the assumptions underlying the original formulation and leading to overly conservative margins. PURPOSE: We investigated whether the widely adopted van Herk margin formula overestimates clinical target volume (CTV)-to-planning target volume (PTV) expansions for contemporary coplanar volumetric modulated arc therapy (VMAT) prostate treatments. METHODS: Fifty consecutive intact-prostate VMAT radiotherapy plans (two coplanar arcs; clinical margins 3 mm, except for 2 mm posterior) were exported. Direction-specific 90% isodose-to-PTV gaps and penumbra widths were measured. Candidate anisotropic margins were tested by eroding the PTV to create CTV_eval. Monte-Carlo simulations combined systematic (Σ) shifts with Gaussian random (σ') blurring kernels of 0-2 mm were performed. Acceptability criteria of (i) CTV Dmin0.03 cc ≥ 90% Rx in ≥ 90% of simulated scenarios or (ii) population tumor-control probability (TCP) loss < 1 % were used. RESULTS: VMAT plans exhibited intrinsic 90% isodose clearances of 3-5 mm laterally/anteriorly and 1-2 mm superior-inferiorly, while axial-plane penumbras were up to fivefold broader than van Herk's assumption. With a [0, 0, 2] mm (LR/AP/SI) margin, ≥ 90% of patients maintained Dmin0.03 cc ≥ 90% Rx provided Σ lay within an ellipsoid of [2.0, 1.5, 1.8] mm and σ' ≤ [1.5, 2.0, 1.5] mm. Under TCP criteria the safe Σ ellipsoid for high-risk disease was [2.5, 1.9, 2.2] mm, while low-intermediate risk was even less sensitive. CONCLUSIONS: For image-guided coplanar VMAT prostate radiotherapy, an anisotropic [0, 0, 2] mm CTV-to-PTV margin is sufficient for target coverage. A modified margin expression that subtracts the measured isodose-to-PTV gap and uses reduced random-error coefficients better reflects modern practice.
Head radiotherapy positioning guidance system based on feature recognition and automatic annotation: Clinical validation and error analysis.
Wang Y, Wang C, Hua G … +3 more , Zhu L, Li Y, Lin Q
Med Phys · 2026 Apr · PMID 41904706 · Publisher ↗
BACKGROUND: Positioning accuracy in radiotherapy is critical for treatment outcomes, especially in head tumor radiotherapy, where the target area is small and surrounded by dense organs, requiring higher precision. PURPO...
BACKGROUND: Positioning accuracy in radiotherapy is critical for treatment outcomes, especially in head tumor radiotherapy, where the target area is small and surrounded by dense organs, requiring higher precision. PURPOSE: To explore the feasibility of constructing a radiotherapy positioning guidance system using an RGB-D camera and deep learning algorithms, and analyze the positioning errors of the system in head radiotherapy localization. METHODS: This study proposes an innovative positioning method that integrates deep learning algorithms into the radiotherapy workflow. An RGB-D camera was used in both the CT simulation and radiotherapy rooms to capture patient surface and facial images, which were used to develop the DeepLab-Opt and Fast Face Marker Detector (FFMD) algorithms. DeepLab-Opt was applied for coarse positioning through surface contour extraction, whereas FFMD was used for fine positioning by detecting facial landmarks. In the CT simulation room, color and depth images were acquired to generate reference contour and 3D facial landmark data, which were stored in the patient positioning database. During treatment setup in the radiotherapy room, real-time calibrated images were acquired and compared with the archived reference data to provide deviation and calibration-rate feedback for therapist adjustment. The system performance was then compared with that of the traditional cross-laser positioning method. Using an alternating design, 22 patients with head tumors underwent positioning with the proposed system and the conventional method on different treatment days, and positioning accuracy was evaluated by MVCT verification. RESULTS: The Mann-Whitney U test was used to compare the MVCT verification positioning deviation data from 246 cases. The system's positioning errors in the lateral, longitudinal, vertical directions, and roll were 1.73 ± 1.35 mm, 1.53 ± 1.20 mm, 0.82 ± 0.94 mm, and 0.69° ± 0.51°, respectively, all significantly lower than those of the traditional cross-laser positioning method (p < 0.05). Additionally, the method reduced the positioning and registration time from 345.9 ± 93.4 to 307.8 ± 36.2 s (p < 0.001), with MVCT verification passing on the first attempt, reducing the need for multiple verifications and effectively reducing the radiation dose the patient receives during positioning. CONCLUSION: The radiotherapy positioning guidance system is feasible and can provide real-time feedback on the patient's outer contour and facial feature point deviations, achieving precise mapping between CT simulation positioning and treatment positioning. It effectively improves the accuracy and efficiency of head radiotherapy positioning, demonstrating strong clinical application potential.
Real-time energy measurement of clinical carbon ion beams using a cross-correlation time-of-flight method with parallel-plate chambers.
Kwon NH, Choi SW, Han S … +13 more , Yun Y, Han MC, Hong CS, Kim HJ, Lee H, Kim C, Kim DW, Koom WS, Kim JS, Carolino N, Lopes L, Kim DW, Fonte PJR
Med Phys · 2026 Apr · PMID 41904704 · Full text
BACKGROUND: In carbon-ion radiotherapy (CIRT), the beam energy determines both the particle range and the overall dosimetric quality. Range-verification QA devices such as Zebra and Giraffe, which are based on multilayer...
BACKGROUND: In carbon-ion radiotherapy (CIRT), the beam energy determines both the particle range and the overall dosimetric quality. Range-verification QA devices such as Zebra and Giraffe, which are based on multilayer ionization chambers (MLICs), can verify the range but only under dedicated QA conditions, leaving any energy deviations introduced by nozzle components undetected in real time. In particular, nozzle structures such as ridge filters can broaden or modulate the energy spectrum, causing the effective energy delivered to the patient to differ from the nominal accelerator setting. These limitations highlight the need for a real-time method capable of verifying the beam energy under actual clinical operating conditions. PURPOSE: We proposed a TOF-based beam-energy measurement concept that leverages a cross-correlation analysis of full detector waveforms. Compact and radiation-hard parallel-plate chambers (PPCs) were developed and evaluated, in contrast to prior TOF systems based on semiconductor detectors. METHODS: PPCs (2.5 cm diameter active area, 0.4 mm gas gap) were operated in CO. Two detectors were mounted coaxially with detector separations of 22.5 and 46.3 cm. Experiments were performed at Yonsei Heavy-ion Therapy Center (HITC) using four nominal energies (102.6, 140.4, 250.3, 430 MeV/nucleon) and three intensities, covering the clinically interesting ranges. Signals were digitized with a 1 GHz bandwidth oscilloscope. For each spill, paired waveforms were cross-correlated, and peak times were refined by parabolic interpolation to determine TOF. Precision and accuracy were evaluated across energies, intensities, and detector separations. RESULTS: The PPCs operated stably for all beam conditions. Under pencil-beam delivery and normalized to 1 s acquisitions, the timing precision of the mean TOF (standard error) remained within 1 ps for both detector separations, scaling with (N: number of TOF samples per acquisition) and not representing the single-particle TOF resolution. Residuals between measured and theoretical TOF remained within 80 ps across energies and distances. After relativistic conversion from TOF to kinetic energy and then to water-equivalent range, all deviations were within a 1 mm range shift, meeting the recommended clinical criteria for range verification. CONCLUSIONS: We demonstrated that compact CO-filled PPCs, operated as a TOF pair, can measure carbon-ion beam energy across the clinically relevant range of energies (≈100-430 MeV/u) and intensities used in routine treatment delivery. We achieved sub-picosecond timing precision on the TOF mean (standard error) per 1 s acquisition and submillimeter water-equivalent range accuracy using a robust cross-correlation analysis method. These results open the way to the integration of PPC-based TOF monitoring to tighten beam-delivery tolerances and improve the reliability and safety of carbon-ion radiotherapy.
Pushing the limits of spatial resolution in clinical PCD-CT using a dedicated high-resolution convolutional neural network (HR-CNN).
Zhou Z, Bratt AK, Koo CW … +3 more , Horst KK, McCollough CH, Yu L
Med Phys · 2026 Apr · PMID 41904700 · Full text
BACKGROUND: Photon-counting-detector (PCD) CT systems offer ultra-high spatial resolution, yet the visual spatial resolution on clinical images often constrained by large pixel size, yielding resolutions below system cap...
BACKGROUND: Photon-counting-detector (PCD) CT systems offer ultra-high spatial resolution, yet the visual spatial resolution on clinical images often constrained by large pixel size, yielding resolutions below system capabilities. While reducing pixel size and using sharp kernels enhance visual spatial resolution, it increases noise, compromising image quality. PURPOSE: To investigate the combined effects of pixel size and reconstruction kernels on visual spatial resolution using phantom and clinical images and to develop a dedicated high-resolution deep convolutional neural network (HR-CNN) to better utilize the intrinsic high spatial resolution of PCD-CT in clinical imaging. METHODS: The relationship between spatial resolution, reconstruction kernel, and pixel size was investigated to identify strategies for utilizing the full spatial resolution potential of PCD-CT. To overcome the increased noise associated with high-resolution settings, a dedicated HR-CNN was developed to push the limit of spatial resolution in routine PCD-CT exams. The HR-CNN was trained using patient exams acquired with ultra-high-resolution (UHR) mode and reconstructed with a 150-mm field of view (FOV), matrix size of 1024×1024 (0.15-mm pixel size) and sharpest quantitative kernel (Qr89). The impact of FOV, kernel, and denoising on spatial resolution was studied using bar-pattern phantoms and a pilot clinical evaluation including 5 patients with interstitial lung diseases. Two thoracic radiologists evaluated 4 different FOV/reconstruction conditions: (1) FOV-410/Qr56-Iterative reconstruction (IR), (2) FOV-410/Qr89-IR, (3) FOV-150/Qr89-IR, and (4) FOV-150/Qr89-HR-CNN in terms of overall image quality, noise, visual spatial resolution, and overall preference. RESULTS: With a FOV of 410 mm, the Qr89 sharp kernel displayed bar-patterns up to 14 lp/cm, not much higher than the routine lung kernel Qr56. When the FOV was reduced to 150 mm, Qr89-IR allowed for the visualization of line pair patterns ranging from 18 to 20 lp/cm, with 20 lp/cm being moderately discernible. The application of Qr89-HR-CNN yielded further improvement, enabling the display of line pair patterns as high as 20-22 lp/cm. In patient cases, both radiologists consistently ranked the FOV-150 images processed with HR-CNN as superior across metrics including overall image quality, noise reduction, visual spatial resolution, and overall preference. The HR-CNN reduced the noise in patients' images by 93.0 ± 0.6% and 44.9 ± 5.3% in comparison with the original FBP and IR images, respectively. CONCLUSIONS: The spatial resolution of PCD-CT is not maximized in routine practice due to the large FOV and high noise levels at sharp kernels. The proposed HR-CNN denoising method, along with small pixel size, may allow the high spatial resolution toward the system limit to be implemented in practice, which is beneficial in the diagnosis of many diseases, including interstitial lung disease.
Independent verification of vendor-issued dosimetric data for Ru brachytherapy using diode detectors with traceability to external beam standards for absorbed dose to water.
Dahlander S, Billas I, Sander T … +3 more , Bass G, Persson L, Tedgren ÅC
Med Phys · 2026 Apr · PMID 41904697 · Full text
BACKGROUND: Independent verification of the manufacturer-provided dosimetry data for Ru ophthalmic brachytherapy applicators is crucial for safe and accurate treatment, yet a standardized, traceable method for clinical a...
BACKGROUND: Independent verification of the manufacturer-provided dosimetry data for Ru ophthalmic brachytherapy applicators is crucial for safe and accurate treatment, yet a standardized, traceable method for clinical absolute dosimetry has been lacking. PURPOSE: This work establishes a complete framework for traceable absorbed dose to water measurements of Ru eye plaques in absolute units of Gray, complementing the high-precision BetaCheck-106™ setup with a robust detector calibration methodology independent of the source manufacturer. METHODS: Three microSilicon diode detectors were calibrated in traceable Co and 6 MeV electron reference beams. Depth-dose measurements for four CCB-type Ru plaques were performed in water using the BetaCheck-106™ setup. Monte Carlo (MC) simulations were employed to calculate depth-dependent beam quality correction factors which account for the detector's response in the Ru field relative to the calibration beams. The method was validated against measurements performed at the National Physical Laboratory (NPL) using alanine dosimetry. MC simulations were also used to investigate the water-equivalence of the NPL alanine/PMMA phantom setup. RESULTS: The MC-calculated correction factors for the diodes showed a significant depth-dependence, underscoring the necessity of such corrections in the steep Ru depth-dose gradient. The dose rates determined with the calibrated diodes were in agreement with the NPL alanine results. Both methods yielded dose rates systematically lower than those provided in the manufacturer's certificates, though generally within the stated uncertainties. The MC simulations revealed substantial non-water equivalence correction factors for the alanine/PMMA phantom, highlighting the advantage of direct measurements in water. CONCLUSIONS: We present a novel, comprehensive methodology for independent and traceable absolute dosimetry of Ru applicators. By combining a dedicated water phantom setup with diode detectors calibrated against external beam standards and MC-derived correction factors, this framework empowers clinical users to perform robust verification measurements, filling a critical gap in the quality assurance of ocular brachytherapy.
CT data harmonization via learned virtual monoenergetic imaging for cross-kV scan translation and radiomics reproducibility.
Chang S, Swicklik JR, Zhou Z … +7 more , Kharat S, Wellinghoff J, Gong H, Williamson EE, Foley TA, McCollough CH, Leng S
Med Phys · 2026 Apr · PMID 41881591 · Full text
BACKGROUND: Radiomics extracts quantitative imaging features from computed tomography (CT) data for clinical decision-making. However, variations in acquisition parameters-particularly x-ray tube voltage (kV)-introduce n...
BACKGROUND: Radiomics extracts quantitative imaging features from computed tomography (CT) data for clinical decision-making. However, variations in acquisition parameters-particularly x-ray tube voltage (kV)-introduce non-biological variability in attenuation values, limiting the reproducibility of radiomic features across scanners, protocols, and institutions. PURPOSE: To develop and evaluate a CT dAta harmoNiZAtion framework based on deep learNed vIrTual monoEnergetic imaging (TANZANITE), which leverages the keV flexibility of virtual monoenergetic images (VMIs) to enable cross-kV scan translation and radiomics harmonization. METHODS: TANZANITE is a model hub consisting of multiple pre-trained convolutional neural networks (CNNs), each designed to translate VMIs from 1 keV level to another. Phantom-based calibration was first used to determine energy-equivalent (Eff_E) keV levels corresponding to each tube potential (e.g., Eff_E(A) keV for source kV and Eff_E(B) keV for target kV). A CNN trained using 69 120 patches from seven patient cases to map VMIs from Eff_E(A) to Eff_E(B) was selected from the TANZANITE hub and applied directly to clinical CT images acquired at the source kV. This harmonized the images to match the attenuation characteristics of the target kV setting. Evaluation was conducted on independent dual-energy CT datasets acquired at 100/Sn150 kV. Regions of interest (ROIs) were placed in the kidney, liver, and spine to assess CT number consistency and radiomic feature reproducibility. The concordance correlation coefficient (CCC) was calculated across 93 non-shape radiomic features. RESULTS: After TANZANITE processing with 100 kV images, CT numbers in evaluated organs closely matched the Sn150 kV reference values in four testing patient cases. For example, mean kidney CT numbers changed from 320 HU (100 kV) to 156 HU (TANZANITE), approximating the Sn150 kV value of 160 HU. Similar changes were observed in the liver (157-105 HU vs. 104 HU reference) and spine (45-22 HU vs. 19 HU reference). Radiomic reproducibility improved substantially across organs: mean CCC increased from 0.590 to 0.995 in the liver, 0.300 to 0.970 in the kidney, and 0.630 to 0.968 in the spine. Post-TANZANITE, over 98% of features exceeded the stability threshold (CCC ≥ 0.900) in all three representative organs. CONCLUSION: TANZANITE provides a flexible, image-domain harmonization framework by learning the keV-to-keV translation in the VMI domain and applying pre-trained CNNs to clinical kV images. It improves CT number consistency and organ-specific radiomic reproducibility without requiring raw projection data or scanner-specific training. This approach supports consistent quantitative imaging across multiple-kV acquisition protocols, enhancing radiomics reliability in clinical settings.
Anatomical feature-guided semi-supervised recognition of the vertebrobasilar artery for microvascular decompression.
Zhang J, Xie L, Huang J … +7 more , Xing Z, Li Y, Wen C, Zhuge Q, Pan Y, Zeng Q, Feng Y
Med Phys · 2026 Apr · PMID 41881590 · Publisher ↗
BACKGROUND: Microvascular decompression (MVD) surgery is a surgical procedure commonly used in the treatment of trigeminal neuralgia, which is based on the principle of relieving compression of the responsible blood vess...
BACKGROUND: Microvascular decompression (MVD) surgery is a surgical procedure commonly used in the treatment of trigeminal neuralgia, which is based on the principle of relieving compression of the responsible blood vessels. Accurate segmentation of the vertebrobasilar artery can provide physicians with more intuitive spatial location relationships, effectively improving the success rate of MVD. In recent years, various learning-based methods have been widely used for artery segmentation. However, accurately segment the vertebrobasilar artery in complex skull base structures with little labeling data remains a challenge. PURPOSE: This study seeks to identify the vessel responsible for MVD using an anatomical feature-guided semi-supervised automated identification approach in clinical data. METHODS: We proposed a multiscale uncertainty cross-pseudo-labeling network with a residual adaptive attention module. Predictions at different scales are generated through multiple decoder levels, emphasizing the reliable part of the prediction and ignoring the regions with low confidence, so that the prediction results at different scales are consistent. Meanwhile, we designed a module called Res-AdaptiveAttention, which enables the utilization of prior knowledge related to the vertebrobasilar artery. RESULTS: Experiments indicated that our method could maintain the overall structure of the vertebrobasilar artery with minimal annotation labels, outperforming current semi-supervised methods across various metrics. In terms of the dataset, this paper divided the data into a training set, validation set, and test set, with the number of cases being 58 for the training set, eight for the validation set, and 13 for the test set. Additionally, through the evaluation of clinical data, it can accurately display the lesion location, providing significant clinical application value. CONCLUSIONS: Our method can accurately segment the vertebrobasilar artery in clinical data, and provide visual anatomical references for its positional relationship with the trigeminal nerve. This work lays a foundation for MVD preoperative planning, while further research will focus on quantifying the positional relationship between the segmented artery and trigeminal nerve at multiple anatomical locations to enhance clinical applicability.
Kinetic model of radiochemical oxygen depletion (ROD) in FLASH radiotherapy.
Seco J, Freitas H
Med Phys · 2026 Apr · PMID 41881566 · Full text
BACKGROUND: The role of oxygen in "Ultra-High Dose Rate" (UHDR) radiotherapy is currently subject to active debate, due to its importance in the FLASH effect. Radiochemical oxygen depletion (ROD) is used to characterize...
BACKGROUND: The role of oxygen in "Ultra-High Dose Rate" (UHDR) radiotherapy is currently subject to active debate, due to its importance in the FLASH effect. Radiochemical oxygen depletion (ROD) is used to characterize the removal of oxygen by its interaction with the free radicals produced by the radiation. Currently, there is a need to understand why ROD depends on the radiation dose rate and the initial oxygen pressure. PURPOSE: Development of a kinetic model of ROD that explains its dependence on (i) radiation dose rate and (ii) initial oxygen pressure, . METHODS: The current work uses a variety of published ROD studies performed in vitro and in vivo in mice to evaluate the kinetic model prediction of ROD. The in vitro studies include evaluation of ROD in water, bovine serum albumin (BSA), and CELL medium consisting of HEPES ( ), glycerol ( ), glucose ( ), and glutathione ( ). Published in vivo studies were performed in C57BL/6 mice (male and female) and NU(Ico)-Foxn1nu mice (female Swiss nude) using proton FLASH and electron, respectively. Oxygen pressure measurements were performed with a variety of different probes such as (i) TROXSP5 sensors, (ii) Oxyphor PtG4, and (iii) Oxylite (NX-BF/OT/E). Two definitions of ROD were used in the current work to represent separately the ROD dependence in "time" ( ) and "dose" ( ). RESULTS: The kinetic model prediction agreed well with published measurements, yielding reduced values near the unity for water, BSA, and CELL medium, and comparably strong agreement for the animal-study datasets, within the reported or estimated uncertainties used in this work. The solvated electron G-value, , was shown to be dose rate, LET dependent and medium specific. For a medium with radical scavenging capacity (such as BSA and CELL), a higher value of was observed compared to water, which had a much lower radical scavenging capacity. The kinetic model dose rate predictions also achieved very good agreement, with the published in vitro in water and BSA medium. The kinetic model's dose-rate predictions for in vivo mice studies also showed excellent agreement once the raw oxygen consumption data were corrected for oxygen diffusion during radiation delivery. CONCLUSIONS: A systematic review of all published ROD studies was performed and used as the basis for testing the novel kinetic model for ROD. The kinetic model prediction of ROD showed that the radiolysis products, , , , , play an important role in ROD and provide an explanation why ROD depends on (1) dose rate and (2) initial oxygen pressure.
Predicting ventilation from single breathing phase non-contrast CT using Swin Transformers.
Liu YK, Kuo HT, Melek A … +5 more , Castillo R, Vinogradskiy Y, Zhao L, Nair G, Castillo E
Med Phys · 2026 Apr · PMID 41881558 · Full text
BACKGROUND: Pulmonary ventilation imaging enables functional avoidance radiotherapy treatment plans by quantifying regional lung function. However, current clinical standards, such as 99𝑚Tc-based single-photon emission c...
BACKGROUND: Pulmonary ventilation imaging enables functional avoidance radiotherapy treatment plans by quantifying regional lung function. However, current clinical standards, such as 99𝑚Tc-based single-photon emission computed tomography (SPECT), rely on radioactive tracers, which can introduce imaging deposition artifacts. CT ventilation imaging (CTVI) methods based on both physical models and deep learning approaches currently require multiple CT images as input, such as the inhale/exhale phases of a 4DCT. While the theoretical foundation of physics-based CTVI is built on multi-phase information, the feasibility of single-phase deep learning CTV models has not been determined. PURPOSE: While deep learning methods have predicted SPECT ventilation from multi-phase 4DCT, the benefit of including more than one respiratory phase remains unclear. Predicting ventilation using only single-phase CTs reduces computational expense, potentially simplifies the image acquisition process, and avoids artifacts introduced by image registration, thereby making deep learning-based CTV approaches more feasible for clinical applications outside of radiotherapy. This study (1) develops a deep learning model to predict SPECT ventilation using only the inhale phase of non-contrast 4DCT and (2) evaluates the impact of adding the exhale phase. METHODS: We developed a SwinUNETR-based architecture using the maximum inhale 4DCT phase to predict pulmonary ventilation. A total of 44 cases with paired inhale CT and SPECT scans were used in the training. To assess multi-phase benefits, we compared: (1) InhaleCT-Swin Model-trained on inhale CT only; (2) ExhaleCT-Swin Model-trained on exhale CT only; (3) Hybrid Models IECT-Swin-FTD, IECT-Swin-FTDE, IECT-Swin-FTDES, fine-tuned on inhale/exhale CT pairs (IECT) with varying network components updated. A standard U-Net was also trained on inhale CT (InhaleCT-UNet), exhale CT (ExhaleCT-UNet), and IECT (IECT-UNet) for cross-architecture evaluation. RESULTS: The SwinUNETR-based Hybrid Model, IECT-Swin-FTD, achieved mean voxel-wise Spearman correlation of 0.762 ± 0.035, outperforming the current state-of-the-art methods. Our transformer-based model trained on inhale CT slightly outperformed exhale CT with no significant differences ( ). U-Net achieved lower overall accuracy, though its highest performance occurred with IECT. No significant difference was found between InhaleCT-Swin Model and the best-performing hybrid UNet Model, IECT-UNet ( ). CONCLUSIONS: A transformer-based model with its decoder fine-tuned on IECT (IECT-Swin-FTD) achieved state-of-the-art accuracy for SPECT ventilation prediction. Moreover, our InhaleCT-Swin Model achieved comparable results with widely used UNet-based models that require multi-phase CT, showing that single CT may be sufficient for accurate ventilation prediction and may improve clinical workflow by reducing acquisition requirements and registration-related artifacts.
Unsupervised 1D CNN -bidirectional long short-term memory model with multi-head attention for generating intravoxel incoherent motion maps.
Li ZY, Huang HM
Med Phys · 2026 Mar · PMID 41855015 · Publisher ↗
BACKGROUND: Intravoxel incoherent motion (IVIM) imaging, a diffusion magnetic resonance imaging technique, is commonly used to quantify tissue perfusion and diffusion. Traditional pixel-by-pixel fitting methods, however,...
BACKGROUND: Intravoxel incoherent motion (IVIM) imaging, a diffusion magnetic resonance imaging technique, is commonly used to quantify tissue perfusion and diffusion. Traditional pixel-by-pixel fitting methods, however, often suffer from high noise, causing unreliable parameter estimates. PURPOSE AND METHODS: To address this issue, a novel unsupervised learning-based framework combining a one-dimensional (1D) convolutional neural network (CNN) with a bidirectional long short-term memory (BiLSTM) network and a multi-head attention mechanism (MHAM) was proposed. Several techniques were proposed to reduce the effect of random weight initialization, noisy input data, and overfitting/underfitting on the estimation of IVIM parameters. The performance of the proposed method was evaluated using both simulated and experimental data, and the results were compared with those obtained using the deep neural network (DNN) method and the Bayesian-Markov random fields (MRF) method. RESULTS: Simulation results showed that the proposed method achieved lower root mean square error values than the other two methods, indicating more reliable IVIM parameter estimates. The only exception was at a signal-to-noise ratio of 100, where it performed similarly to the Bayesian-MRF method. For the abdominal datasets, the proposed method yielded IVIM parameter estimates that closely matched the literature-reported values and avoided the overestimation of pseudo-diffusion coefficients (D) observed in the other two methods. For the brain dataset, the perfusion fractions and diffusion coefficients obtained from all three methods were consistent with the literature-reported ranges; however, only the DNN method tended to overestimate D. CONCLUSIONS: These findings suggest that the proposed CNN-BiLSTM-MHAM model is a promising approach for IVIM parameter estimation.
A physics-driven neural network with parameter embedding for generating quantitative MR maps from weighted images.
Chen L, Zhang C, Yi Y … +9 more , Wang Y, Song Y, Yan X, Xu S, Zhu D, Cao M, Zhou Y, Wang C, Yang G
Med Phys · 2026 Mar · PMID 41854958 · Publisher ↗
BACKGROUND: Traditional quantitative MRI (qMRI) techniques require acquisition of multiple weighted images to create a single quantitative mapping, which prolongs the scan time and limits their clinical applications. Dee...
BACKGROUND: Traditional quantitative MRI (qMRI) techniques require acquisition of multiple weighted images to create a single quantitative mapping, which prolongs the scan time and limits their clinical applications. Deep learning (DL) has emerged as a promising solution for synthesizing quantitative maps from conventional weighted MRI images. However, existing DL methods often overlook the underlying physical principles inherent in MR signals, which inevitably compromises the performance and generalizability of the model. PURPOSE: To develop a deep learning-based approach that integrates MRI sequence parameters to improve the accuracy and generalizability of quantitative image synthesis from clinical weighted MRI. METHODS: We proposed a physics-driven neural network that embeds MRI sequence parameters-repetition time (TR), echo time (TE), and inversion time (TI) - directly into the model via parameter embedding. This design enables the network to learn the underlying physical principles of MRI signal formation. The model takes conventional T1-weighted, T2-weighted, and T2-fluid-attenuated inversion recovery (T2-FLAIR) images as input and synthesizes T1, T2, and proton density (PD) quantitative maps. The model was trained on healthy brain MR images and evaluated on both internal and external test datasets. RESULTS: The proposed method consistently achieved the best performance across all evaluation metrics compared with conventional deep learning methods (pGAN and U-Net). On the internal test set, the model achieved mean percentage errors (MPE) below 6% for T1, 10% for T2, and 5% for PD, with corresponding global voxel-wise mean absolute errors (MAE) of approximately 60 ms for T1, 10 ms for T2, and 30 ms for PD. Notably, the proposed model accurately generated quantitative maps for previously unseen pathological regions, highlighting its superior generalization capability. CONCLUSION: Incorporating MRI sequence parameters via parameter embedding allows the neural network to better learn the physical characteristics of MR signals, significantly enhancing the performance and reliability of quantitative MRI synthesis. This method shows great potential for accelerating qMRI and improving its clinical utility.
Revisiting the segmentation threshold for Lu-177 SPECT.
Liu Y, Yang J, Wang P … +3 more , Wang Y, Chen Y, Mok GSP
Med Phys · 2026 Mar · PMID 41854844 · Full text
BACKGROUND: The mostly used threshold-based segmentation method for SPECT, i.e., 42% of the maximum intensity, was derived from Tc and may not be directly applicable to Lu. PURPOSE: This study aims to revisit the optimal...
BACKGROUND: The mostly used threshold-based segmentation method for SPECT, i.e., 42% of the maximum intensity, was derived from Tc and may not be directly applicable to Lu. PURPOSE: This study aims to revisit the optimal segmentation threshold for Lu SPECT. METHODS: A cylindrical Jaszczak phantom containing six spheres (2-113 mL) was imaged via simulation and physical experiments using a clinical dual-head NaI SPECT/CT system. The spheres were filled with Tc and Lu, with different sphere-to-background ratios (SBRs). One hundred and twenty projections were acquired and reconstructed using filtered back-projection (FBP) and 3D ordered subset expectation maximization (OS-EM) algorithms with attenuation and scatter corrections, followed by Gaussian filtering (σ = 3.8 mm). Thresholds from 1% to 99% (1% interval) of peak intensity were applied to minimize the absolute volume error (AVE) of the spheres. The newly derived Lu threshold was further validated on Lu-PSMA-617 (n = 6), Lu-DOTATATE (n = 5), Lu-FAP-2286 (n = 5) and Lu-DOTA-IBA (n = 4) SPECT images, comprising 45 tumors with manual segmentations used as reference. Mean Dice, HD95, and AVE were calculated for all tumors and compared between the conventional threshold (42%) and the newly derived threshold using the Mann-Whitney U test. RESULTS: The optimal threshold increased along with the decrease in sphere volume or SBR. For SBR ≥ 3.5:1 and volume ≥ 16 mL, the mean optimal threshold of Lu converged to 56% for FBP and 50% for OS-EM. The derived 50% threshold significantly improved tumor segmentation performance compared to the 42% threshold, with a higher Dice score (0.5999 ± 0.1589 vs. 0.6694 ± 0.1361) (p < 0.05), lower HD95 (2.0070 ± 1.1508 mm vs. 1.7392 ± 1.0643 mm), and lower AVE (130.72% ± 101.87% vs. 69.21% ± 63.49%) (p < 0.05). CONCLUSIONS: An optimal Lu-specific threshold (∼50%) was derived and clinically validated, differing from the conventional 42% threshold used for Tc. The new threshold improved segmentation accuracy across different therapeutic radiopharmaceutical distributions.
Mining whole-brain information with deep learning to predict EGFR mutation and subtypes in brain-metastatic NSCLC: A multicenter study.
You S, Fan Y, Zhang J … +7 more , Yang C, Sun Y, Jiang M, Chen H, Guo W, Yang H, Jiang W
Med Phys · 2026 Mar · PMID 41854843 · Publisher ↗
BACKGROUND: Epidermal growth factor receptor (EGFR) and its mutation subtypes play a pivotal role in the treatment of non-small cell lung cancer (NSCLC) patients. Therefore, developing an accurate, noninvasive quantitati...
BACKGROUND: Epidermal growth factor receptor (EGFR) and its mutation subtypes play a pivotal role in the treatment of non-small cell lung cancer (NSCLC) patients. Therefore, developing an accurate, noninvasive quantitative method to predict EGFR genotype is crucial for personalized treatment. PURPOSE: To explore a deep learning-based method with whole-brain information for predicting EGFR mutation and subtypes utilizing MRI images in NSCLC patients presenting with brain metastasis (BM). METHODS: This study enrolled 293 patients with BM. A primary set was built with 170 patients from Center 1 (between January 2017 and December 2021). External sets were constructed with 62 patients from Center 2 (between July 2014 and October 2021) and 61 patients from Center 3 (between January 2020 and October 2022). All patients underwent contrast-enhanced T1-weighted (T1CE) and T2-weighted (T2W) brain MRI scans before genetic testing. An EGFR site recognition network (ESR-Net) was developed by mining whole-brain information to predict EGFR mutations and subtypes. The ESR-Net integrated deformable convolution and an auxiliary network to seek informative mutation features and enhance tumor features, respectively. Predictive performances of deep learning models were assessed using area under the curve (AUC) analysis. RESULTS: For the prediction of EGFR mutations, the ESR-Net demonstrated superior performance with AUCs ranging from 0.835 to 0.840 across primary and external validation sets, surpassing conventional state-of-the-art methodologies. Furthermore, the ESR-Net exhibited AUCs ranging from 0.858 to 0.904 for predicting EGFR exon 19 (Del19) mutation and 0.838 to 0.903 for predicting EGFR exon 21 (L858R) mutation across primary and external sets. CONCLUSIONS: The developed ESR-Net demonstrates promising potential for early detection of EGFR mutations and subtypes with multicenter data, which may promote optimal treatment management for patients with brain metastatic NSCLC.
Simplified dosimetry using two-time-point kinetic modeling of I-MIBG PET for I-MIBG therapy in neuroblastoma.
Wang Y, Huh Y, Matthay KK … +4 more , Vo KT, Pampaloni MH, DuBois SG, Seo Y
Med Phys · 2026 Mar · PMID 41854810 · Full text
BACKGROUND: I-metaiodobenzylguanidine (I-MIBG) therapy is an established and effective treatment for metastatic neuroblastoma. Due to the substantial variability in absorbed dose across different tumor sites and organs,...
BACKGROUND: I-metaiodobenzylguanidine (I-MIBG) therapy is an established and effective treatment for metastatic neuroblastoma. Due to the substantial variability in absorbed dose across different tumor sites and organs, I-MIBG dosimetry, such as achieved via SPECT imaging, is critical for enabling personalized therapy planning. However, conventional imaging-based dosimetry typically requires three or more imaging sessions to reliably estimate time-integrated activity (TIA) of tumors and organs, which imposes workflow burdens and increases patient inconvenience. Therefore, there is a clear need for dosimetry methods that can maintain accuracy while requiring fewer imaging sessions. PURPOSE: This study aims to develop and validate a simplified dosimetry method for I-MIBG therapy that enables robust estimation of TIA using only two imaging time points. The method leverages kinetic modeling to estimate tumor and organ time-activity curves (TACs) and TIAs from limited imaging data and was validated using I-MIBG PET imaging data. METHODS: Five subjects with neuroblastoma underwent I-MIBG PET/CT imaging at three or four time points post-administration. Two imaging time points (∼28 and ∼113 h post-administration) were selected for TIA estimation using a kinetic modeling framework. To obtain the blood input function, left ventricular activity at the two time points was extracted and fitted to a mono-exponential function. With this input function, a one-tissue compartmental model was then applied to estimate tumor and organ TACs from the two-time-point data, and the corresponding TIAs were calculated by integrating the modeled TACs. The proposed method was compared with (1) a conventional mono-exponential fitting method using the same two-time-point data, and (2) a reference standard based on bi-exponential fitting of all available three- or four-time-point data. To evaluate the performance of the proposed method, relative errors in TIA estimation for tumors and organs were calculated using the bi-exponential fitting results as the reference. RESULTS: The proposed method achieved substantially improved accuracy over mono-exponential fitting. Taking the bi-exponential method as the reference, the proposed method yielded an average TIA estimation bias of 0.3%, a standard deviation of 13.8%, and a root mean square error (RMSE) of 14.2%. In contrast, mono-exponential fitting resulted in a higher bias of 14.9%, a standard deviation of 36.3%, and an RMSE of 39.5%. Specifically, the proposed method outperformed mono-exponential fitting in tumors, adrenal glands, brain, and thyroid. CONCLUSIONS: We developed a novel dosimetry method based on two-time-point imaging and kinetic modeling that enables simplified TIA estimation in I-MIBG therapy. Validated using I-MIBG PET data, this approach demonstrated improved TIA estimation performance compared with conventional mono-exponential fitting. It may offer a physiologically motivated and more clinically applicable solution that supports personalized dosimetry and facilitates individualized treatment planning in radiopharmaceutical therapy.
Tensor-decomposition regularized learning for fast and high-fidelity multi-parametric microstructural MR imaging.
Fan W, Cheng J, Tian Q … +5 more , Wu R, Zou J, Si W, Chen Z, Wang S
Med Phys · 2026 Mar · PMID 41846469 · Publisher ↗
BACKGROUND: Deep learning (DL) has emerged as a promising approach for learning the nonlinear mapping between diffusion-weighted MR images and tissue parameters, enabling automatic and in-depth understanding of brain mic...
BACKGROUND: Deep learning (DL) has emerged as a promising approach for learning the nonlinear mapping between diffusion-weighted MR images and tissue parameters, enabling automatic and in-depth understanding of brain microstructures. However, the efficiency and accuracy of jointly estimating multiple microstructural parameters derived from various diffusion models remain limited due to isolated signal modeling and dense sampling requirements. PURPOSE: This study aims to develop a unified DL framework for fast and high-fidelity estimation of multiple microstructural parameters derived from different diffusion models using sparsely sampled q-space data. METHODS: We propose DeepMpMRI, an efficient and extendable framework equipped with a novel tensor-decomposition-based regularizer that captures fine structural details by exploiting high-dimensional correlations across parameters. In addition, a Nesterov-based adaptive learning algorithm is introduced to dynamically optimize the regularization parameter, improving both efficiency and reconstruction accuracy. RESULTS: Experimental results on the Human Connectome Project (HCP) dataset and an Alzheimer's disease dataset demonstrate that DeepMpMRI outperforms five state-of-the-art methods in simultaneously estimating DKI- and NODDI-derived parameter maps, achieving 4.5-15 acceleration compared to dense sampling with 270 diffusion gradients. CONCLUSIONS: DeepMpMRI enables accurate and robust multi-parametric microstructural imaging under sparse sampling conditions, showing strong potential for clinical translation in efficient diffusion MRI-based tissue characterization.
Impact of CT dose on AI performance: A comparison of radiomics, deep, and foundation models in a multicentric anthropomorphic phantom study.
Martín Asiain M, Amirian M, Jimenez Del Toro O … +10 more , Aberle C, Schaer R, Bach M, Obmann M, Flouris K, Müller H, Stieltjes B, Konukoglu E, Andrearczyk V, Depeursinge A
Med Phys · 2026 Mar · PMID 41846467 · Full text
BACKGROUND: Computed tomography (CT) is widely used in clinical practice due to its ability to provide detailed anatomical information. However, variations in radiation dose can affect image quality, potentially compromi...
BACKGROUND: Computed tomography (CT) is widely used in clinical practice due to its ability to provide detailed anatomical information. However, variations in radiation dose can affect image quality, potentially compromising the performance and reliability of artificial intelligence (AI) models applied to these images. PURPOSE: To evaluate the robustness of radiomics-based and deep learning-based models to variations in CT dose levels using a standardized dataset obtained from a 3D-printed anthropomorphic phantom simulating liver tissue with anomalies, as well as in the publicly available dataset CT-ORG with real patient data for organ classification. This study is in an early experimental stage, tested only on retrospective data. METHODS: A total of 1378 image series from 649 scans were acquired across 13 scanners from four manufacturers at five dose levels. Features were extracted from six regions of interest (ROIs), representing four liver tissue types (normal, cyst, hemangioma, metastasis), using four methods: PyRadiomics, a shallow convolutional neural network (CNN), SwinUNETR, and a CT foundation model (CT-FM). Feature stability was assessed using the Intraclass Correlation Coefficient (ICC), while Uniform Manifold Approximation and Projection (UMAP) was employed to evaluate tissue types separability and the influence of scanner variations. Generalizability was tested by training liver tissue classifiers on one dose level and testing on others, alongside a dose classification task (10-fold cross-validation) to determine the sensitivity of each method to dose variations. In addition, we compared the four methods in addressing the task of organ classification (10-fold cross-validation) with the CT-ORG dataset containing 140 CT scans acquired with varying dose levels. RESULTS: Radiomic features showed limited robustness to dose variations, leading to reduced performance in liver tissue classification and the lowest ICC among methods (ICC: 0.8355 0.1705). SwinUNETR and CT-FM exhibited the highest stability (SwinUNETR ICC: 0.9528 0.0272; CT-FM ICC: 0.9347 0.0420), clearly above the Shallow CNN (ICC: 0.8416 0.2018). CT-FM also showed strong generalization across dose levels: its features effectively distinguished between liver tissue types and dose levels simultaneously, without compromising performance in either task. Consistent with these trends in dose sensitivity, CT-FM obtained the highest dose-classification accuracy (0.6517 0.0179), whereas SwinUNETR showed the lowest (0.3796 0.0250). These trends were confirmed in the context of organ classification with real patient data on the CT-ORG dataset, where CT-FM achieved the highest accuracy (0.965). CONCLUSIONS: The study highlights the limited robustness of traditional radiomics and deep models to CT dose variation and underscores the potential of foundation models like CT-FM to enable robust clinical applications by mitigating dose-related variability. This enhanced performance is likely due to the model's pretraining on large and diverse datasets, allowing it to learn robust and generalizable representations across varying acquisition conditions.
Deep learning-based upsampling of 2D detector array measurements for patient plan verification in radiotherapy.
Pflaum A, Brand N, Kempf E … +5 more , Weidner J, Eulenstein D, Delfs V, Poppe B, Looe HK
Med Phys · 2026 Mar · PMID 41840888 · Full text
BACKGROUND: Detector arrays are commonly used for treatment plan verifications in intensity modulated radiation therapy. However, the intrinsic resolution of detector arrays is limited by the physical dimensions of each...
BACKGROUND: Detector arrays are commonly used for treatment plan verifications in intensity modulated radiation therapy. However, the intrinsic resolution of detector arrays is limited by the physical dimensions of each single detector and the detector-to-detector distance. This may lead to inaccurate representations of steep gradients and narrow dose peaks. PURPOSE: This work presents a deep learning approach for increasing the effective spatial resolution of detector arrays used for patient plan verification. The presented approach aims to augment missing values in the insensitive areas of the detector matrix and to increase the sampling frequency of the measured 2D dose profile. Furthermore, perturbations caused by finite detector's dimensions via the volume-averaging effect are corrected during the upsampling process. METHODS: In this work, Monte Carlo simulation methods were employed to synthetically generate training data, enabling a wide coverage of different linear accelerator setups and field shapes. The approach was implemented for the OCTAVIUS Detector 1500 (PTW Freiburg, Germany), which consists of 1405 air-filled ionization chambers arranged in a checkerboard pattern. This arrangement enables a threefold increase in resolution from 5 mm, achieved with the standard bilinear interpolation, to 1.7 mm using neural networks. The implemented neural networks are based on a deep convolutional architecture and were trained using PyTorch. Initially, the models were tested by comparing the upsampled measurements of individual step-and-shoot IMRT segments with measurements obtained using a high-resolution OCTAVIUS Detector 1600 SRS liquid-filled ionization chamber array. In addition, radiochromic film measurements of fields with leaf gaps of 1 and 2 cm were used to demonstrate the differences between measurements and interpolation results in the presence of steep gradients and narrow dose peaks. Finally, reconstructed 3D dose distributions of VMAT plans, using both the original and upsampled measurements, were compared to the treatment planning system calculations. RESULTS: The comparison of the individual IMRT segments with a standard bilinear interpolation showed an average increase in the gamma index passing rate of up to 20%. In the case of 3D dose reconstructions from the field-by-field IMRT measurements in the OCTAVIUS 4D phantom, the neural network upsampling yielded an average increase in passing rate of 22% as compared to bilinear interpolation, when using the OD 1600SRS array measurements as reference and 19% with the TPS calculated dose distribution as reference. Similarly, for VMAT plans, the passing rate showed an average increase of 8% for the measurement at an Elekta accelerator and 7% at a Varian Ethos accelerator, both using the TPS dose distribution as reference. CONCLUSION: It has been shown that a neural network can be applied to upsample the detector array resolution. This results in a better interpolation of measurement points, especially in regions of steep gradients, than compared to a standard bilinear interpolation. The passing rates of all investigated VMAT plans are increased by applying the proposed neural network upsampling approach.
Assessment of discoid meniscus injury using solid-state nuclear magnetic resonance (SSNMR).
Li M, Fatima S, Wabasa N
Med Phys · 2026 Mar · PMID 41840873 · Publisher ↗
BACKGROUND: Lateral discoid meniscus (LDM) injuries have traditionally been diagnosed using magnetic resonance imaging (MRI), which provides detailed macroscopic visualization of the meniscus. MRI remains the gold standa...
BACKGROUND: Lateral discoid meniscus (LDM) injuries have traditionally been diagnosed using magnetic resonance imaging (MRI), which provides detailed macroscopic visualization of the meniscus. MRI remains the gold standard for evaluating meniscal morphology and tears, allowing for differentiation between complete and incomplete discoid meniscus types. However, while MRI effectively identifies macroscopic alterations such as meniscal thickening, hypermobility, and tears, it may not always detect the early degenerative changes at the same molecular resolution. PURPOSE: This study explores the potential of solid-state nuclear magnetic resonance (SSNMR) in diagnosing adult LDM injuries. METHODS: The study involved 80 adult participants, divided into two groups: 40 individuals with LDM injuries (experimental group) and 40 healthy controls. SSNMR assessed meniscal integrity by Chemical Shift Anisotropy, dipolar coupling strength, and T1 and T2 relaxation times. MRI assessed morphological changes such as free edge height and body width. Independent sample t-tests were applied for statistical comparison, and effect sizes (Cohen's d) were calculated to determine the practical significance. RESULTS: The injured menisci showed higher Chemical Shift Anisotropy (75.2 ± 4.1 ppm vs. 60.3 ± 3.8 ppm, p < 0.001, Cohen's d = 1.50) and dipolar coupling strength (8.4 ± 1.1 kHz vs. 5.2 ± 0.9 kHz, p < 0.001, Cohen's d = 1.25), indicating collagen degradation. T1 (710 ± 50 ms vs. 530 ± 40 ms, p < 0.001, Cohen's d = 1.65) and T2 (48 ± 3.4 ms vs. 32 ± 2.1 ms, p < 0.001, Cohen's d = 1.58) relaxation times were also significantly prolonged in the injured group, reflecting altered hydration. Morphologically, the injured group had higher free edge height (5.39 ± 0.71 mm vs. 1.69 ± 0.30 mm, p < 0.001, Cohen's d = 2.0) and body width (31.1 ± 3.16 mm vs. 2.39 ± 0.31 mm, p < 0.001, Cohen's d = 1.8). Capsular edge height was significantly lower (4.09 ± 0.33 mm vs. 5.11 ± 0.59 mm, p < 0.001, Cohen's d = 1.3). The fat angle sign was higher (65% vs. 0%, p < 0.001), and the wedge sign was lower (10% vs. 72.5%, p < 0.001). CONCLUSION: SSNMR offers a novel approach and underscores the potential of SSNMR in orthopedic imaging, facilitating improved precision in early diagnosis and treatment planning.
Single-field-uniform-dose-per-fraction simultaneous dose and dose rate optimization (SFUDPF-SDDRO) method for proton FLASH therapy.
Luo Y, Zhu YN, Setianegara J … +6 more , Hong X, Zhang W, Wang C, Lin Y, Li Q, Gao H
Med Phys · 2026 Mar · PMID 41833534 · Publisher ↗
BACKGROUND: The FLASH effect can significantly reduce radiation-induced normal tissue damage while maintaining tumour control, but requires ultra-high dose rates and high doses. PURPOSE: This work proposes a single-field...
BACKGROUND: The FLASH effect can significantly reduce radiation-induced normal tissue damage while maintaining tumour control, but requires ultra-high dose rates and high doses. PURPOSE: This work proposes a single-field-uniform-dose-per-fraction simultaneous dose and dose rate optimization (SFUDPF-SDDRO) method for proton FLASH radiotherapy to ensure both dose rate and dose meet FLASH effect thresholds. METHODS: The SFUDPF method focuses on delivering the prescription dose for each fraction from only a single field instead of multiple fields, which inherently supports the ultra-high dose rate and high dose necessary for the FLASH effect. We performed retrospective FLASH treatment planning utilizing SFUDPF-SDDRO on four clinical head-and-neck (HN) cases for this study. SFUDPF planning involves delivering each prescription fraction (8 Gy x 5 fx) in 1 beam angle as opposed to multiple beam angles per fraction for IMPT. For each beam delivery, we maximized the FLASH effect in a 1 cm expansion of the HN CTV (CTV+1 cm) by enforcing FLASH dose-rate and dose thresholds of 40 Gy/s and 5 Gy, respectively, in this region. The pencil-beam-scanning dose rate (PBSDR) was calculated voxel-wise by modeling the raster-scanning spot trajectory, while neglecting energy switching times under the assumption of a range modulator capable of expanding a single-energy beam into a spread-out Bragg peak (SOBP). Robust optimization at 3 mm/3.5% was performed to address setup and range uncertainties. We employed iterative convex relaxation and alternating direction method of multipliers algorithms to solve the non-convex optimization problem posed by the SFUDPF-SDDRO model. The FLASH effect was modelled within this work by multiplying the proton dose with a constant 0.7 dose modification factor for voxels fulfilling the dose-rate and dose thresholds to obtain the FLASH effective dose (FED). Effects of FLASH sparing maximization via SFUDPF-SDDRO are verified by comparing with IMPT and VMAT on plan qualities such as (i) high-dose area sparing, (ii) conformity index (CI), and (iii) OAR doses. RESULTS: FLASH RT via SFUDPF-SDDRO compared with IMPT and VMAT was evaluated for four clinical HN cases with different tumor geometries. When compared with their VMAT counterparts, SFUD-SDDRO achieved a considerable reduction of FED for OAR directly adjacent to the CTV. Specifically in case 1, the brainstem D decreased from 87.57% to 62.26%, and the spinal cord D decreased from 87.36% to 60.74%; in case 2, the D of the carotid decreased from 102.46% to 63.30%; in case 3, the Dof the oral cavity decreased from 94.72% to 62.66%, and the D of the oropharynx decreased from 102.5% to 69.09%; in case 4, the D of the oral cavity decreased from 88.56% to 59.81%. The SFUDPF-SDDRO achieved a satisfactory CI in terms of FED, indicating that conformity was not sacrificed to achieve the FLASH effect. CONCLUSION: The proposed SFUDPF-SDDRO method is feasible and shows potential clinical benefits for FLASH treatment planning. Maximizing the FLASH effect within a 1 cm ring around the target substantially limits high-dose spillage and enhances OAR sparing compared with conventional approaches.
Semantic edge-guided single-view 2D/3D registration for vertebrae in X-rays.
Shen A, Jiang J, Tang Y … +5 more , Chen Z, Nong L, Chen Q, Wang F, Wu X
Med Phys · 2026 Mar · PMID 41833531 · Publisher ↗
BACKGROUND: The integration of artificial intelligence into image-guided intraoperative interventions holds considerable promise for deriving 3D geometric information from 2D imaging. 2D/3D registration establishes the s...
BACKGROUND: The integration of artificial intelligence into image-guided intraoperative interventions holds considerable promise for deriving 3D geometric information from 2D imaging. 2D/3D registration establishes the spatial relationship between preoperative computed tomography (CT) and intraoperative X-rays. However, existing methods are often limited by the image domain gap and imprecise feature extraction, causing coarse registration to provide inadequate initial poses and subsequent fine registration to fall into local optima, thereby reducing accuracy. PURPOSE: We aim to develop a robust single-view lumbar spine 2D/3D registration framework that balances high clinical accuracy with intraoperative efficiency requirements by aligning preoperative CT with intraoperative X-rays. METHODS: We propose utilizing vertebral body edges in X-rays as novel semantic features to guide 2D/3D registration. For robust edge extraction, we develop ESegMamba, an efficient U-shaped Mamba network incorporating Group multi-axis Hadamard Product Attention (GHPA) and Group Aggregation Concatenation (GAC) modules. Experiments for semantic edge extraction were performed on a dataset of 710 images (comprising X-rays and Digitally Reconstructed Radiographs) derived from 10 patients. The dataset was partitioned using a 4:1 patient-specific split, resulting in 568 training and 142 test images. The training set was further utilized via 5-fold cross-validation for network fine-tuning. ESegMamba was benchmarked against SegMamba, SwinUNETR, and UNETR using Dice and mIoU metrics. For 2D/3D registration, experiments were conducted separately on 300 simulated samples and 90 real clinical samples, following the same patient-specific split. The proposed framework was compared with landmark-based, intensity-based, and learning-based methods using mean Target Registration Error (mTRE). Statistical significance was assessed using the Wilcoxon signed-rank test with a significance level of 0.05, applying Bonferroni correction for multiple comparisons. RESULTS: ESegMamba outperforms representative networks with fewer parameters (99.18 M), achieving 90.36% Dice and 85.49% mIoU on the test set. Compared to the strong baseline SegMamba, ESegMamba demonstrated a large effect size in Dice improvement (Cohen's , ). For 2D/3D registration, the proposed method demonstrated superior performance over representative benchmarks. Specifically, compared to Xreg and PSSS, our method achieved large practical improvements in mTRE ( and , respectively; ). On real clinical data, the method achieved a mean in-plane translation error of approximately 1.5 mm and an average registration time of approximately 10 s. CONCLUSIONS: The proposed method, empowered by ESegMamba, yields statistically significant improvements over intensity-based benchmarks ( ). The achieved sub-2mm accuracy and 10 s processing time on clinical data confirm its efficacy for intraoperative spinal navigation. The code for the proposed method is available at github.com/shenao1995/lineReg.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe