OBJECTIVE:Positron emission tomography (PET) reconstruction is an ill-posed inverse problem, particularly under low-count conditions where noise severely degrades image quality and quantitative accuracy. Although supervised learning approaches have demonstrated strong denoising capability, their performance often depends on large paired datasets and may suffer from limited generalization. This work aims to develop an unsupervised reconstruction framework for time-of-flight PET (TOF-PET) that improves image quality while maintaining quantitative reliability. APPROACH:We propose a TOF-PET reconstruction method based on implicit neural representations (INR). A differentiable forward projection model is incorporated to explicitly model TOF-PET imaging physics and enable reconstruction directly in the INR domain. To suppress noise and promote spatial smoothness, a ray-based total variation (TV) regularization is introduced. The reconstruction network combines a multi-resolution hash encoder with a prior-image encoder that injects structural image priors into the INR representation. MAIN RESULTS:The proposed framework was evaluated using simulated brain and whole-body datasets as well as clinical TOF-PET scans. Results show improved noise suppression and contrast recovery compared with conventional iterative reconstruction algorithms and representative unsupervised approaches. SIGNIFICANCE:The proposed approach integrates implicit neural representations with physics-consistent modeling and prior-guided regularization, providing an effective unsupervised framework for TOF-PET reconstruction and highlighting the potential of neural field representations for tomographic imaging.
Objective:Voxel-wise PET kinetic modeling of dynamic PET data enables the generation of quantitative parametric maps reflecting tracer kinetics, but it is computationally demanding and sensitive to noise. This study investigated whether embedding the underlying kinetic model of PET tracer behavior into a neural network framework (PET Kinetics-informed Neural Network, PKiNN) improves voxel-wise parameter estimation compared with conventional linearized approaches. 
Approach:PKiNN, a recurrent neural network with long short-term memory (LSTM), was trained on simulated time-activity curves generated using a one-tissue compartment model (1TCM) with randomized kinetic parameters spanning physiologically plausible ranges. A kinetics-informed loss term was incorporated into the network loss function to constrain the predicted parameters. A baseline neural network (NN-LSTM) without the kinetics-informed loss was also trained for comparison. A supplementary simplified reference tissue model (SRTM) implementation was evaluated as a proof of concept.
Main Results:Across all simulated noise levels, PKiNN consistently outperformed NN-LSTM, yielding lower absolute relative error (ARE, %) for K1, k2, and VT; for example, at an SNR of 20 dB,the REs and AREs for VTestimated by NN-LSTM were -5.0 ± 10.1 % and 8.7 ± 7.2 %, respectively, while for the VTderived from PKiNN, RE and ARE were -0.5 ± 4.6 % and 3.5 ± 3.1 %, respectively. For clinical11C-UCB-J PET data, PKiNN-derived voxel-wise maps showed close agreement with VOI-based 1TCM fitting, minimal bias in K1and k2, modest overestimation of VT, and reduced underestimation compared with Logan graphical analysis. 
Significance:Overall, PKiNN provides a fast, accurate, and noise-robust alternative to conventional 1TCM kinetic modeling, demonstrating feasibility for clinical PET, and the supplementary SRTM proof-of-concept provides preliminary evidence of its potential for extension to other kinetic models.
OBJECTIVE:Magnetic Particle Imaging (MPI) is an emerging medical imaging technique for high-sensitivity and quantitative visualization of tracer distributions. Narrowband MPI achieves high signal-to-noise ratio (SNR) by selectively detecting a few harmonics. However, in its conventional implementations, the point spread function (PSF) inevitably exhibits negative lobes, which introduce anisotropic resolution and destructive signal superposition.
Approach. To address this, we propose a Narrowband Isotropic MPI system, utilizing the Field-Free-Line (FFL)-parallel excitation (FPE) scanning strategy to eliminate negative lobes. FPE-MPI employs single-channel acquisition with coaxial excitation and receive, applies an excitation field parallel to the FFL while maintaining a constant collinear offset field, and encodes harmonic responses on the projection plane orthogonal to the FFL.
Main results. Through simulations and phantom experiments on our in-house scanner, we demonstrate that FPE-MPI eliminates the adverse effects of negative lobes, achieves isotropic imaging, and improves image quality. Compared with conventional narrowband MPI with FFL-orthogonal excitation, FPE-MPI preserves higher SNR in the acquired signals and yields robust reconstructions at low tracer concentrations, with only a modest trade-off in spatial resolution. 
Significance. FPE-MPI facilitates robust imaging at low tracer doses and provides a feasible scanning strategy for the development of future large field-of-view (FOV) narrowband MPI systems.
OBJECTIVE:Quantitative imaging-based dosimetry is essential for optimizing [177Lu]Lu-DOTATATE peptide receptor radionuclide therapy (PRRT), yet full multi-time-point (MTP) SPECT/CT is difficult to implement in routine clinical practice. This study proposes a hybrid planar-SPECT framework for organ-level TIA estimation and STP-constrained voxel-level Monte Carlo dose analysis from a single post-therapy SPECT/CT acquisition. APPROACH:The proposed framework integrates single-time-point (STP) SPECT/CT with serial planar imaging to derive organ-level time-integrated activities (TIAs). A physics-informed support vector regression (SVR) model is incorporated to refine planar-to-SPECT scaling factors and improve the robustness of planar-derived activity quantification. The resulting organ TIAs are used to scale the fixed within-organ STP SPECT pattern and generate voxelized TIA maps as source distributions for Monte Carlo (MC)-based dose calculation. The method is evaluated against reference MTP SPECT/CT data in 11 patients and further applied to a separate cohort of 23 patients to explore vertebral dose heterogeneity and its association with hematologic toxicity. MAIN RESULTS:Hybrid dosimetry showed good agreement with MTP SPECT/CT, with organ-level errors of approximately 10-15%. The SVR-based refinement modestly reduced scaling bias, particularly in anatomically challenging structures. Voxel-level vertebral dose heterogeneity metrics showed moderate but exploratory correlations with hematologic toxicity (ρ ≈ 0.4-0.5) after false-discovery-rate correction, whereas conventional organ-averaged dose metrics showed no significant association. SIGNIFICANCE:The proposed hybrid planar-SPECT framework supports clinically feasible organ-level dosimetry from a reduced acquisition protocol. The resulting voxelized maps provide STP-constrained approximate source distributions for MC dose calculation and exploratory characterization of spatial dose heterogeneity, rather than reconstructions of time-varying MTP voxel kinetics. The SVR improvement observed in the 11-patient internal validation remains preliminary, requiring larger cohorts for robustness and external validation for generalizability.
We present a novel high-resolution, high-sensitivity time-space coincidence imaging system for cascade gamma photons. The system employs a LaBr 3 ring detector with hybrid pinhole-slit collimators, combining the high spatial resolution of pinhole collimation with the high detection efficiency of slit collimation. Simulations were conducted with a point source and a Derenzo phantom using 177 Lu, a theranostic radionuclide that emits cascade gamma-ray pairs at 113 keV and 208 keV. Images were reconstructed using direct back-projection (DBP) and maximum likelihood expectation maximization (MLEM), as well as two newly proposed algorithms: refined DBP (R-DBP) and multi-information joint reconstruction (MIJR).The system achieved a central coincidence efficiency of 2.98×10⁻⁵. Point source imaging with MLEM reconstruction yielded a spatial resolution of 1.7 mm full width at half maximum (FWHM) in the transaxial plane. Derenzo phantom imaging demonstrated clear resolution of hot rods as small as 1.2 mm in diameter, with a contrast-to-noise ratio (CNR) of 9.29 for the largest rods.These results demonstrate that the proposed ring detector enables high-quality cascade gamma photon coincidence imaging, with spatial resolution and sensitivity that significantly exceed those of previously reported systems. The combination of high resolution, reasonable sensitivity, and theranostic capability positions this technology as a promising platform for integrated diagnosis and therapy.
OBJECTIVE:Childbirth is a major cause of levator ani muscle (LAM) injury, affecting over 10% of women after vaginal delivery. Progress in prevention, diagnosis, and treatment is limited by poor understanding of LAM mechanical properties particularly muscle elasticity, which is closely linked to injury mechanisms and can be measured noninvasively using shear wave elastography (SWE). Conventional SWE assumes large, isotropic tissues, making it unsuitable for the small, anisotropic, and complex LAM. This study aimed to develop fiber-network LAM phantoms and to investigate rotational SWE imaging to assess local, direction-dependent shear anisotropy properties of the LAM. 
Approach: Six LAM phantoms with varying fiber type, fiber density, freeze-thaw cycles, and fiber networks (puborectalis alone or combined puborectalis-pubococcygeus) were constructed by embedding synthetic fibers within a polyvinyl alcohol matrix. A custom rotational SWE imaging setup with a Verasonics V1 system was used. Shear wave velocities were estimated from axial velocity maps using a semi-automatic Radon sum algorithm and fitted to an elliptical model. Based on the theory of shear wave propagation in transversely isotropic (TI) materials, this model enabled the estimation of shear anisotropy metrics, including fiber direction, shear anisotropy, longitudinal and transverse shear moduli, and TI profile fit quality. 
Main results: Rotational SWE imaging differentiated puborectalis LAM phantoms with shear anisotropy (1.09-4.95), longitudinal shear moduli (8.67-60.41 kPa), and transverse shear moduli (3.09-12.26 kPa), reflecting distinct biomechanical and age-related LAM properties. Different muscle network configurations of the LAM were also distinguished across three probe positions, and local fiber curvature was detected. 
Significance: This study demonstrates that rotational SWE imaging can characterize local shear anisotropy and fiber-network architecture in anatomically informed LAM phantoms, beyond what can be obtained from conventional SWE. These findings provide a foundation for future studies investigating whether rotational SWE imaging can improve the in vivo assessment of the LAM.
OBJECTIVE:Chemotherapy-induced cardiotoxicity can lead to irreversible heart failure. Left ventricular ejection fraction (LVEF) is routinely monitored during treatment, but conventional assessment requires dedicated cardiac imaging and clinical resources. This study evaluated a deviceless data-driven gating (DDG) framework for extracting cardiac motion from routine [18F]FDG PET emission data and, secondary, assessed its feasibility for LVEF estimation. 
Approach. The DDG framework is based on histo images and uses anatomical masking for frequency-domain filtering to obtain the cardiac signals of 169 [18F]FDG PET/CT examinations. The cardiac gating performance was evaluated, and LVEF estimates were compared with the patient's respective clinical reference method; echocardiography or multigated acquisition (MUGA).
Main results. The DDG framework successfully extracted a cardiac gating signal in 138 of 169 examinations (81.7%). Sufficient myocardial [18F]FDG uptake for software-based LVEF estimation was present in 101 patients (60%), and LVEF was successfully estimated in all cases. Compared with the reference methods, the DDG-based LVEF estimates demonstrated a mean bias of 3.4% relative to echocardiography and -2.6% relative to MUGA. Agreement was strongest with echocardiography, although the limits of agreement exceeded the threshold required for interchangeable clinical use. The extracted pulse frequencies were physiologically plausible and showed good agreement with the corresponding reference examinations, supporting the validity of the DDG-derived cardiac signal.
Significance. The deviceless DDG framework can successfully extract clinically relevant cardiac motion directly from routine [18F]FDG PET acquisitions without external hardware. The method achieved a high success rate with respect to gating the images and provided LVEF estimates that showed good agreement with established clinical reference methods, supporting the feasibility of deriving functional cardiac information from standard PET examinations. The proposed approach provides a promising foundation for retrospective functional cardiac assessment and has the potential to complement conventional LVEF evaluation while reducing additional patient burden and simplifying clinical workflows.
OBJECTIVE:Deep unrolling network, as a promising deep learning approach for low-dose computed tomography (LDCT) reconstruction, can efficiently address the issues of severe noise and artifacts in LDCT imaging. However, most existing methods predominantly unrolled the fidelity term to convolutional neural networks (CNNs) while failing to unroll regularization terms, which inevitably limits the feature capture capability of the network. To further improve the performance of deep unrolling networks, this paper proposed the CTDNet, a cartoon texture decomposition-based deep unrolling network that unrolls not only the data fidelity term but also regularization terms into CNNs. APPROACH:The cartoon texture decomposition model of Meyer was incorporated as regularization terms, together with the data fidelity term, to formulate the reconstruction optimization problem. This problem was solved by the Chambolle-Pock (CP) algorithm, yielding a single-loop iterative algorithm instance. This instance was then unrolled to the deep reconstruction network for LDCT, by replacing each update step of iterative process with a simple sub-network. To further enhance the quality of reconstruction results, a lightweight image
refinement and fusion module was employed to perform detail enhancement and remove residual noise of the reconstructed images. MAIN RESULTS:Extensive experiments were conducted on the "Low-Dose CT Image and Projection Data" dataset and the "Piglet Dataset". The experiment results demonstrated that the CTDNet effectively removes artifacts and noise from LDCT images while maximally preserving textural structures, which enables it to outperform in terms of visual effects and objective metrics. SIGNIFICANCE:This work further unrolls the regularization terms to CNNs on the basis of unrolling the data fidelity term, providing a novel unrolling strategy for the future design of deep unrolling networks.
.
The development of Long Axial Field-of-View (LAFOV) positron emission tomography (PET) combined with computed tomography (CT) warrants reconsideration of the administered activity required for clinical imaging. This study investigated the relationship between body size, administered activity (18F), and image quality using two patient-like phantoms representing normal (78.5 kg) and obese (194.3 kg) body profiles. The phantoms were scanned in selected five-minute periods within a total duration of 27 and 21 hours, respectively, following the decay of the activity. 
Analysis of the trues, scatter, and randoms demonstrated a close to linear propagation of the trues with activity up to a point where saturation was reached. The normal body profile phantom achieved higher signal-to-noise ratios (SNRs), reaching system saturation at a lower activity (680 MBq) than the large phantom, representing the obese body profile (970 MBq). The SNR2 showed a nonlinear correlation with the noise equivalent count rate (NECR), whereas SNR2 had a linear correlation with the trues. 
The study demonstrated the count rate capabilities up to and beyond the saturation limit. Furthermore, we illustrated the possibilities of scanning patients with ultralow dose administration in the range of 0.03-0.07 MBq/kg body weight. However, achieving robust clinical diagnostic quality - particularly for low-contrast lesions - at these activity levels requires further clinical investigation. Utilizing these ultralow activity concentrations requires extending the scan duration to compensate for image quality, especially when examining obese patients.
OBJECTIVE:Sparse-view computed tomography (CT) reduces radiation dose and acquisition time by decreasing the number of projection views, but it also makes image reconstruction severely ill-posed, leading to structural distortion and severe artifacts. This study aims to develop an effective reconstruction framework for improving both projection-data fidelity and reconstructed image quality in sparse-view CT. APPROACH:We propose a group Convolution- and self-Attention Fusion-based Dual-domain Iterative Method (CAFDIM) for sparse-view CT reconstruction. CAFDIM follows a model-informed dual-domain iterative design. The framework consists of the Initialization Enhancement Network (IE-Net), Gradient Update Block (GUB), Projection-domain Repair Network (PR-Net), Image-domain Repair Network(IR-Net), and Momentum Update Block (MUB). The projection-domain branch employs a Deep Sparse Block (DSB) to enhance sparse projection features before full-view projection restoration, while the image-domain branch uses edge-guided residual refinement to improve anatomical structure preservation. To enhance local-global feature representation, a Convolution-Attention Fusion Block (CAFB) is embedded into both repair branches by combining group convolution with Pixel Shift Self-Attention (PSSA). RESULTS:Experiments on simulated and real clinical projection datasets demonstrate that CAFDIM effectively suppresses sparse-view artifacts, preserves anatomical structures, and achieves superior reconstruction accuracy, visual quality, and generalization ability compared with state-of-the-art methods. SIGNIFICANCE:CAFDIM provides an effective and efficient dual-domain reconstruction framework for sparse-view CT, showing strong potential for clinical applications in sparse-view and low-dose CT imaging.
Objective.To develop a practical stochastic reconstruction framework for emission
tomography that generates ensembles of data-compatible images and enables
uncertainty quantification and assessment of forward-model adequacy.Approach.
The framework combines stochastic-gradient descent initialization with
Hamiltonian Monte Carlo (HMC) sampling directly in high-dimensional voxel
space. Beyond point reconstruction, we introduce a spatially resolved
operator-weighted diagnostic, the sampled data-visible variance, which
quantifies how image fluctuations propagate through the imaging operator
and thereby probes the local conditioning of the inverse problem under
realistic acquisition physics. The methodology is evaluated using
controlled software phantoms, experimental anthropomorphic phantom
measurements, and a clinical DATSCAN SPECT acquisition.Main results.
Under ideal conditions, the HMC ensemble mean provides point-estimate
accuracy comparable to deterministic reconstruction methods, while the
sampled ensemble provides additional physically interpretable information.
The ensemble analysis helps distinguish uncertainty associated with the
intrinsic ill-posedness of the inverse problem from variability linked to
forward-model inadequacy. The clinical example demonstrates applicability
under realistic acquisition statistics rather than diagnostic performance.
Significance.The proposed stochastic reconstruction framework provides a practical
ensemble-based approach for emission tomography that extends conventional
point reconstruction with model-conditioned uncertainty estimates and
spatially resolved diagnostics of forward-model adequacy.
OBJECTIVE:Alanine has been used frequently as a reference dosimeter in clinical dosimetry. However, its high LET dependence poses a challenge for practical applications. Determination of the required beam quality correction factors requires Monte Carlo simulations. This makes the application of alanine infeasible if dedicated Monte Carlo tools or computing resources are not available. This work sets out to determine generalized alanine correction factors in clinical 12 C beams. APPROACH:Mathematical expressions were established to parametrize dosimetric correction factors. Spectral fluences in 429 MeV/u mono energetic and a range modulated 278 MeV/u 12 C beams were simulated at multiple depths along the central beam axis using GATE/Geant4. Beam quality correction factors were determined under Spencer-Attix conditions in water and in alanine-paraffin pellets. MAIN RESULTS:Analytical expressions for the water-to-medium stopping-power ratios and the beam quality correction factor were obtained. The analytical expressions represent correction factors that are well within the confidence interval, except at the Bragg peak and the distal edge of the SOBP.These points were found to be unsuitable for measurement using alanine standard dosimeters due to the high perturbation factors. SIGNIFICANCE:The results allow alanine dosimetry to be performed for mono-energetic and range-modulated beams without prior Monte Carlo simulations. This greatly reduces the effort associated with practical alanine dosimetry.
OBJECTIVE:Auger electron (AE)-emitting radionuclides are highly sensitive to subcellular localization, and even small variations in intracellular distribution can substantially alter DNA damage. Accordingly, this study quantitatively evaluates the influence of radionuclide type, target-source configuration, and, as a novel feature, the effect of intracellular spatial distribution on the DNA strand breaks using Geant4-DNA.
Approach: A detailed V79 cell model containing atomistic chromatin fibers was implemented in Geant4-DNA. Five AE-emitting radionuclides (99mTc, 111In, 123I, 125I, and 201Tl) were simulated under two target←source configurations (N←N and N←C) and both uniform and Gaussian (standard deviation (σ) = 0.5-3.5 μm) spatial distributions. Direct and indirect DNA strand breaks were quantified. 
Main results: Spatial confinement of activity exerted a quantitatively substantial, damage-type-dependent influence on DNA strand breaks. In the N←N, reducing from a uniform distribution to a Gaussian one (σ= 0.5 μm) increased DSB (HDSB) yields by up to 33% (~38%). This differential response, where complex lesions are disproportionately amplified by spatial heterogeneity relative to simple breaks, reflects the nanometer-scale co-localization required for clustered damage formation and was consistently observed across all radionuclides. N←N configuration resulted in ~44% higher deposited energy compared to N←C. In terms of DNA damage ranking, 201Tl produced the highest yields across all configurations, with deposited energy up to 36% (over 3000%) greater than 125I (99mTc), consistent with its high AE multiplicity and mean energy per decay.
Significance: The findings demonstrate that, for the idealized nucleus-centered truncated Gaussian distributions, increased spatial confinement enhances the formation of complex DNA lesions relative to uniform distributions. These results indicate that assuming uniform distributions may underestimate complex DNA damage for such confined source distributions. However, the conclusions are limited to the specific Gaussian models investigated and should not be generalized to other biologically realistic heterogeneous uptake patterns.
OBJECTIVE:Time calibration is an important procedure for correcting time shifts inherently included in TOF-PET. In this work, we developed a time skew (TS) correction method that uses the intrinsic radioactivity originating from the natural decay of176Lu, a constituent element of the scintillator. APPROACH:The TOF-PET detector has a one-to-one optical coupling between the scintillators and photosensors, enabling independent readout of each channel. For electrical TS correction caused by the electronic circuitry, the electrical TS is estimated from the time-stamp difference between beta particles from176Lu, used as a trigger, and the simultaneously emitted prompt gamma rays. The correction is performed in two stages, first within each detector and then between detectors. Although crystal TS originating from the scintillator can also be corrected using176Lu, obtaining sufficient statistics requires a long acquisition time. Therefore, we focused on the linear correlation between the light output of each crystal and the crystal TS, and implemented a simple crystal TS correction based on a formulation derived from the existing timing correction method. MAIN RESULTS:When the proposed method was applied to coincidence measurements using a pair of TOF-PET detectors, a coincidence resolving time of 232 ps was achieved using only176Lu background data, which was comparable to the 230 ps obtained with the existing method using an external source. SIGNIFICANCE:The developed method is expected not only to simplify timing correction but also to reduce downtime during detector replacement.
OBJECTIVE:The crosshair light-sharing (CLS) PET detector acquires time-of-flight (TOF) and depth-of-interaction (DOI) information by looping scintillation light through a window at the top of the crystals and distributing it to two photosensors, but light is typically detected by an average of six photosensors, making center-of-gravity (CoG) calculation essential for crystal identification. APPROACH:In this work, an improved CLS detector achieved a shorter coincidence resolving time (CRT) by employing a reflector with a thickness comparable to the 0.2 mm photosensor gap, together with a new CoG calculation method specialized for the CLS detector to address the problem that weak signals at deep interaction positions fall below the noise level caused by light leakage. In the improved CLS detector, the reflector thickness was set to 188 μm to approximate the photosensor gap, and the crystal size was optimized accordingly. The developed three-point CoG (3P-CoG) method removes light that does not contribute to position detection by using only the three outputs with the highest detected energies. MAIN RESULTS:For the improved CLS detector, the new reflective material increased light output and contributed to an improvement in the CRT. Applying 3P-CoG enabled clearer separation of all crystals regardless of interaction depth. Furthermore, 3P-CoG improves the DOI resolution by approximately 10%. Finally, 3P-CoG contributes to data-volume reduction because the crystal response remains stable even when the energy threshold is increased. SIGNIFICANCE:These results show that a simple 3P-CoG algorithm can improve interaction-position estimation in CLS detectors.
OBJECTIVE:Accurate dose verification remains a major challenge in particle therapy. Dose monitoring methods such as positron emission tomography (PET) and prompt-gamma (PG) imaging are highly sensitive to uncertainties in tissue composition derived from CT, which can be improved by dual-energy CT (DECT)-based material decomposition. This study systematically assessed the accuracy of four DECT elemental decomposition algorithms in predicting physical dose, annihilation-photon, and PG reference distributions in particle therapy, focusing on CT-based model uncertainty without detector considerations.
Approach: Three parameterization-based DECT methods, a machine-learning (ML) DECT method, and a conventional single-energy CT (SECT) method were used to predict the elemental composition of the ICRP110 human phantom and 10 head-and-neck patients. The physical dose, annihilation, and PG distributions for incident proton, helium, carbon, and oxygen pencil beams were compared using Monte Carlo simulations. For the phantom, mean relative errors (MREs) were calculated relative to the reference distributions. For the patients, the distal fall-off positions of annihilation and PG profiles were compared across the methods.
Main results: In the phantom study, the ML method yielded the lowest MREs across the four incident particle types. For oxygen ions, the ML method yielded an annihilation-photon MRE of 2.64%, compared with 6.50%-6.74% for the parameterization-based DECT methods. The corresponding PG MRE were 1.53%, 2.44%-2.60% for the ML and parameterization-based methods. The matched SECT analysis provided a conventional baseline under the same simulation and evaluation settings. In the patient analyses, the distal fall-off positions of annihilation and PG differed by up to 1.93 ± 0.42 mm and 1.76 ± 0.44 mm, respectively, between ML and the parameterization-based methods. Without voxel-wise patient ground truth, these results describe inter-method differences rather than an accuracy ranking.
Significance: Compared with the parameterization-based methods, the ML method reduced errors in predicting source-level PET and PG reference distributions under the idealized Monte Carlo conditions used here, especially for heavier incident particles. These results quantify model-related uncertainty in complementary with detector resolution in clinical monitoring accuracy.
OBJECTIVE:Proton arc (PAT) therapy combines the dosimetric advantages of protons with the efficiency of arc delivery, but current planning algorithms are based on a static delivery sequence. The discrepancy between the static plan and the actual delivery can result in dosimetric deviations during stereotactic radiosurgery (SRS), where precision is critical.

Approach: A dynamic arc delivery sequencing optimization framework was developed, consisting of three steps: (1) static irradiation and dynamic arc delivery time calculation, (2) incorporation of timing information into static control points, and (3) spot-weighting fine-tuning. Eight multi-metastatic brain cases were retrospectively selected to validate the framework. Plan quality, delivery accuracy, and efficiency were evaluated by reconstructing the delivered dose from virtual logfiles and comparing dosimetric parameters and treatment times.

Main results: Sequencing optimization maintained nominal plan quality, with no significant differences in target coverage (D98) or normal brain sparing (V12, V8) compared with static-control-point PAT. Delivery accuracy improved substantially. For the total gross tumor volume, mean absolute D98 deviation decreased from 62.9 ± 70.1 cGyE (3.4% ± 3.8%) with static-control-point PAT to 11.0 ± 7.4 cGyE (0.6% ± 0.6%) with sequencing optimization. For the worst metastasis, deviations were reduced from 116.4 ± 89.1 cGyE (6.3% ± 5.3%) to 79.4 ± 69.2 cGyE (3.6% ± 3.0%). Delivery efficiency was preserved, with minimal changes in spot number, energy layers, and total treatment time.

Significance: Dynamic sequencing optimization significantly improves the dosimetric fidelity of PAT therapy for brain SRS while maintaining efficiency. By addressing machine-specific timing and mechanical constraints, this framework bridges the gap between nominal planning and clinical delivery, representing an essential step toward routine PAT implementation.
.
Positron Emission Tomography (PET) is a molecular imaging technique that creates an image of radiopharmaceutical distribution using acquired sinogram data. Accurate PET reconstruction requires a balance between data fidelity and regularization. The choice of penalty type and the adjustment of the regularization parameters have a critical effect on image quality, including noise suppression, edge preservation, and contrast recovery. However, manually determining optimal regularization strength is challenging due to their dependence on data properties and clinical needs. It often requires multiple reconstructions and considerable time to achieve satisfying results. To address this issue, we propose SINORES, a supervised deep learning approach to predict the optimal regularization parameters for the modified Block Sequential Regularized Expectation Maximization (BSREM) algorithm. The prediction is based on the raw sinogram data and the scaling coefficients that encode acquisition-related properties. By learning from a synthetic dataset of 2D sinograms and scaling coefficients paired with their corresponding optimal parameters, SINORES rapidly identifies suitable parameter values and avoids the need for manual setting. This work presents a proof of concept that demonstrates the feasibility of the proposed framework in the context of 2D PET imaging. The proposed method achieves consistent parameter estimation across different phantom types and effectively determines suitable parameters for reconstructing real PET data, leading to improved reconstruction quality and reliability in practical settings.
Objective. Linear energy transfer (LET) measurements ofα-particles are increasingly important for micro-dosimetric evaluation in targetedα-radionuclide therapy. However, direct LET measurements remain challenging due to the extremely short range ofα-particles in matter. In this study, we propose a method to estimate LET distributions using quenching effects observed inα-particle trajectory images.Approach. α-particle trajectories were imaged using an ultra-high-resolution imaging system consisting of a Gd3Al2Ga3O12scintillator, an optical magnification unit, and an electron-multiplying charge-coupled device camera.α-particles from three radionuclides were measured: Am-241 (5.5 MeV), Po-213 (8.4 MeV) and Po-212 (8.8 MeV). LET distributions were estimated by taking the ratio of simulated dose profiles (without quenching) to measured depth profiles ofα-particle trajectory images (including quenching). The estimated LET distributions were then compared with those obtained from simulations.Main results. The relative LET distributions along the depth direction were successfully estimated from the measured depth profiles for allα-particles with different energies. The estimated relative LET distributions differed from the simulated distributions by no more than 11%-24% over the depth range from 5µm to the Bragg peak position. Approximate absolute LET distributions were also evaluated.Significance. The proposed method enables a simple and effective estimation of LET distributions ofα-particles from measured depth profiles of trajectory images. It is applicable toα-particles with a wide range of energies and has potential for advancing micro-dosimetry in targetedα-radionuclide therapy.
Objective.Multi-sweep freehand 3D ultrasound is affected by two types of trajectory degradation: inter-sweep rigid bias and intra-sweep local tracking jitter, both of which hinder high-fidelity volumetric reconstruction from tracked 2D ultrasound slices. Although recent neural radiance field methods enable joint optimization of pose and neural representation in optical imaging, they are not directly transferable to ultrasound since limited slice overlap and strong view-dependent speckle of ultrasound make photometric supervision alone ill posed.Approach.This paper presents GLA-NeRF, a global-local aligned neural radiance field framework for joint optimization of pose and neural representation in multi-sweep freehand ultrasound. Within this framework, a unified pose model decouples sweep-wise global rigid bias from frame-wise local tracking jitter, allowing all slices to be mapped into a common canonical space. A global registration module is introduced to correct inter-sweep misalignment through appearance-based image retrieval, hierarchical matching, geometric inlier verification, and uncertainty-weighted Mahalanobis distance. A local denoising module is further incorporated to suppress intra-sweep tracking jitter through statistical sensor noise priors and cumulative Lie-group B-splines, which provide variance-reduced pseudo-labels and physically plausible kinematic constraints.Main results.Experiments on in vivo ultrasound datasets demonstrate that GLA-NeRF stabilizes joint optimization, reduces tracking errors at both global (mm/accuracy regardless of initialization) and local (frommm/tomm/) scales, and improves the anatomical fidelity of synthesized ultrasound views.Significance.Joint optimization of pose and implicit neural representations allow the reconstructed representation to become both more geometrically coherent and more stable for ultrasound novel view synthesis.
