To evaluate site-specific performance and complexity-dependent risk in patient-specific quality assurance for robotic radiosurgery using a two-dimensional array detector. A retrospective analysis was performed on 137 CyberKnife stereotactic treatment plans. PSQA was conducted using the PTW 1600 SRS 2D array detector. Gamma analysis was evaluated at multiple dose–distance criteria (3
Specialized language models can improve performance in expert domains; however, in radiological technology, evaluation datasets and models tailored to radiological technologists’ practice remain limited and are not widely available. In this study, RadTechBERT was developed by pretraining a Bidirectional Encoder Representations from Transformers on a radiological technology–specific corpus from the initial pretraining stage, without relying on continued pretraining of existing general models. To enable systematic evaluation, a new cloze question dataset was constructed from items derived from the Japanese national examination for radiological technologists, and subject labels were added to support subject-level analyses. To develop the model, Unigram and Byte Pair Encoding (BPE) tokenizers were compared with vocabulary sizes of 32 K, 50 K, and 100 K. RadTechBERT was trained under two settings: using only the domain-specific corpus and using a mixed corpus that additionally included Wikipedia. For benchmarking, baseline models pretrained on general corpora such as Wikipedia, as well as existing and medical textbook–based models, were also evaluated. Performance was assessed using Top-5 accuracy on the cloze task, both overall and by subject. RadTechBERT with BPE_32K outperformed baselines in many subjects, with more than a twofold improvement in radiation safety management and radiation measurement relative to the strongest baseline. In contrast, gains were smaller in subjects having substantial overlap with general medicine, and Wikipedia mixing did not yield consistent improvements. The optimal tokenizer and vocabulary size were subject-dependent.
To investigate whether three-dimensional image metric maps can describe the extent and characteristics of local image changes associated with AI-based motion correction using CLEAR Motion in coronary CT. This retrospective single-center study included 24 coronary CT cases reconstructed from the same raw data with and without CLEAR Motion. Three-dimensional maps of structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), and deformation vector field (DVF) magnitude were generated after resampling to 0.5-mm isotropic voxels and intensity normalization. Without a true motion-free reference, the anatomical correctness of motion correction could not be directly verified. Therefore, the maps were assessed using spatial congruence analysis with Precision and Recall, patch-wise Spearman correlation analysis with bootstrap confidence intervals, and visual assessment by two readers using a 5-point scale. The three-dimensional image metric maps depicted local image changes predominantly near the coronary arteries, in a distribution consistent with the intended design of CLEAR Motion. Under the main analysis condition, Precision was 90.1
Age-related iron accumulation in the brain is linked to neurodegenerative processes, contributing to neuronal damage and functional decline. Quantitative Susceptibility Mapping (QSM), an advanced MRI technique, provides superior sensitivity for assessing iron deposition in vivo compared to traditional methods like R2* and susceptibility-weighted imaging (SWI). This systematic review evaluates QSM’s ability to detect age-related iron changes in healthy aging populations, focusing on technical and methodological considerations. Following PRISMA guidelines, we searched Embase, MEDLINE, Scopus, and Web of Science for studies (2015–2025) using QSM to assess brain iron in healthy aging. Included studies reported susceptibility changes in brain nuclei. Data on study characteristics, QSM values, reference regions, and processing methods were extracted. Quality was assessed using the Newcastle–Ottawa Scale. A narrative synthesis was conducted due to methodological heterogeneity. From 110 records, 12 studies with 2,178 participants were included. Consistent increases in magnetic susceptibility, indicating iron accumulation, were observed in the caudate nucleus and putamen. The red nucleus, substantia nigra, and dentate nucleus showed increased susceptibility in most studies, while the hippocampus and thalamus exhibited variable, age-dependent patterns. Methodological diversity in QSM acquisition and processing (phase unwrapping, background field removal, dipole inversion) contributed to variability. QSM could effectively detect age-related cerebral iron deposition, especially in deep gray matter nuclei, with implications for understanding brain aging and neurodegenerative risk. Standardized protocols and longitudinal studies are needed to improve comparability and clarify temporal dynamics. QSM’s sensitivity makes it a valuable biomarker for distinguishing normal aging from pathological processes, informing future diagnostic and therapeutic strategies.
Parotid T2 mapping is used for quantitative evaluation, for example to assess early radiation‑induced changes in the parotid gland. However, the effects of pre-scan meals on parotid gland T2 values have not yet been reported. We investigated the temporal changes in parotid T2 values induced by meal ingestion. This prospective, single-center repeated-measures study included sixteen healthy volunteers. 3.0 T axial multi-slice multi-echo TSE T2 mapping was performed with and without fat suppression. Measurements were performed at fasting (baseline) and then alternated between 10 and 30 min after meal ingestion. T2 values at each time point were calculated by averaging the measurements obtained from six ROIs, which were drawn to avoid Stensen’s duct and major vessels. Repeated-measures ANOVA with Greenhouse–Geisser correction and Bonferroni adjustment was applied. Without fat suppression, the mean parotid T2 value increased from 63.4 ± 4.63 ms at baseline to 65.8 ± 4.52 ms at 26 min after the start of meal ingestion, with no significant differences from baseline in the pairwise comparisons at any time point. With fat suppression, the mean T2 value increased from 55.7 ± 3.10 ms to 58.6 ± 3.01 ms; significant differences were observed from 16 min onward and persisted through the final time point (p < 0.05). These findings suggest the importance of standardizing pre-scan meal status for quantitative parotid T2 mapping.
In this work, we developed a 3D computational framework to evaluate the beam transport performance of a static gantry proposed in previous studies. This static gantry enables magnetically driven azimuthal angle selection via tailored fields rather than mechanical rotation, yet its practical feasibility requires rigorous simulation-based evaluation. Our approach utilizes full 3D magnetic field modeling in Opera3D coupled with a custom phase-ellipse tracking program. This integrated workflow captures realistic 3D fringe-field effects and quantitatively evaluates the optical limits driven by momentum spread and boundary edge-focusing. By identifying these primary limitations, this platform provides the requirements and a practical foundation to guide the development and optimization of future static gantry designs.
This study investigated whether irradiation-induced chemical changes in sucrose solutions could be utilized to control magnetic resonance imaging (MRI) relaxation properties and to propose a novel design strategy for quantitative MRI phantoms. Aqueous sucrose solutions at concentrations of 5–40
A measurable fraction of examinations in retrospective imaging archives lack a recorded body weight, limiting metadata completeness for tasks such as cohort auditing, weight-stratified analysis, and the assembly of machine-learning training datasets. We developed and compared three approaches for body-weight prediction from a single middle-slice CT image: a LightGBM model on 19 CT-derived tabular features (water equivalent diameter, effective diameter, body composition, scanner parameters), an ImageNet-pretrained ResNet-18 on images alone, and a ResNet-18 with late fusion of the same 19 tabular features into the 512-dimensional image embedding. Models were trained with 5-fold patient-disjoint cross-validation on a combined train+validation development pool of 30,857 examinations from 16,169 patients; the best-performing fold of each model family was selected by within-fold validation mean absolute error (MAE) and evaluated, without further tuning, on a held-out test partition of 3,479 examinations from 1,796 patients (patient-disjoint from training and validation). On the test set, the fusion ResNet-18 achieved MAE 4.06 kg (95
MRI provides essential insights into tissue microstructure, and high angular resolution diffusion imaging (HARDI) enables detailed assessment of complex white matter architecture through fiber orientation distribution (FOD) analysis. However, HARDI requires high b-values and multiple diffusion directions, leading to reduced signal-to-noise ratio (SNR) and long scan times. Conventional zero-fill interpolation processing (ZIP) is widely used for super-resolution but is limited by edge blurring and artifacts. This study evaluated an AI-based reconstruction method, Precise IQ Engine (PIQE). Specifically, low-resolution diffusion data were reconstructed to standard resolution and compared with standard-resolution acquisitions. Ten healthy volunteers underwent HARDI at 3T, and FOD were estimated using constrained spherical deconvolution. Quantitative comparisons with reference data demonstrated that PIQE exhibited higher distributional similarity (lower Jensen–Shannon divergence) and directional agreement (higher angular correlation coefficient) compared with ZIP+Advanced Intelligent Clear-IQ Engine (AiCE), with statistically significant similarity observed. These findings indicate potential usefulness in advanced diffusion MRI applications.
Bilateral renal tumors have proven challenging to precisely segment out of abdominopelvic CT scans. In this regard, the present paper explores a hybridization technique based on spine anchoring using active contour modelling and spatial prioritization for enhanced renal volumetric segmentation. To begin with, the presented model is built upon the spine detection that helps define an anatomically sound anchor for subsequent processing, which involves spatially constrained localization of renal areas using active contour modelling. The pre-processing step can be considered a spatial attention mechanism that limits the search space of the target region and provides additional anatomical context. Next, a 3D-UNet network is utilized for the precise segmentation of nephric tumors with independent coding of left/right kidneys to account for morphological variations. The model’s evaluation was performed on the KiTS19 dataset where the Dice coefficient was equal to 90.39
To compare surface dose at the mammary gland level between 70- and 120-kVp non-contrast neonatal chest computed tomography (CT) under matched volume CT dose index (CTDIvol) conditions, a neonatal anthropomorphic phantom was scanned using a 256-row CT scanner. Real-time dosimeters were placed bilaterally on the anterior chest wall at the mammary gland level and on the posterior chest wall at the same axial level. For the 70-kVp protocol, the noise index was adjusted to match the console-displayed CTDIvol of the 120-kVp protocol. Each protocol was repeated 15 times, and image noise was measured on axial images. Anterior surface dose was significantly higher with 70 kVp than with 120 kVp (median [IQR], 5.67 [5.17–5.86] vs. 5.29 [4.84–5.43] mGy; p < 0.01). Posterior surface dose and image noise did not differ significantly. Under matched CTDIvol conditions, 70-kVp neonatal chest CT increased anterior mammary surface dose relative to 120-kVp CT in this phantom.
Accurate delineation of cervical cancer clinical target volume (CTV) remains labor-intensive and variable. This retrospective single-center study internally evaluated a SAM-Med3D baseline initialized from public pretrained weights and fine-tuned on the study training cohort for prompted cervical cancer CTV segmentation on planning CT. The eligible cohort comprised 182 cases split into training ( n=130 ), validation ( n=17 ), and temporally subsequent independent test ( n=35 ) sets. CT scans were resampled, normalized, and prepared as label-centered 128 × 128 × 128 patches. The SAM-Med3D model was initialized from public pretrained weights and fine-tuned on the training cohort without architectural modification, then assessed with simulated/oracle-guided prompts at 1, 3, 5, 7, 9, and 11 clicks using Dice, HD95, 3-mm surface Dice, and volume consistency. Mean Dice increased from 0.827 at 1 click to a peak of 0.835 at 7 clicks and was 0.833 at 11 clicks. HD95 decreased from 12.46 mm at 1 click to 9.48 mm at 9 clicks and 10.18 mm at 11 clicks, and 3-mm surface Dice increased from 0.707 to 0.721 at 7 clicks. Thus, multi-click prompting produced modest improvements that plateaued at later clicks. A supplementary risk-guided prompt-selection pilot was performed on the same test set but was treated only as exploratory. Because patch extraction and prompt generation used reference-contour information, all analyses represent controlled upper-bound evidence. This study provides an internal feasibility baseline for promptable cervical cancer CTV segmentation; label-free ROI selection, full-volume inference, external validation, and clinician-in-the-loop testing remain necessary before clinical deployment.
The aim of this study was to establish an optimal treatment strategy for functional liver–sparing radiotherapy by incorporating technetium-99 m galactosyl human serum albumin (99mTc-GSA) single-photon emission computed tomography (SPECT) into radiotherapy planning for recurrent hepatocellular carcinoma (rHCC). Three irradiation techniques—volumetric-modulated arc therapy (VMAT), intensity-modulated radiotherapy (IMRT), and three-dimensional conformal radiotherapy (3D-CRT)—were systematically compared. Six patients with rHCC who underwent stereotactic body radiotherapy (SBRT) were included. Eligible patients had previously received SBRT to a different hepatic segment and had undergone 99mTc-GSA SPECT before the current course of radiotherapy. For treatment planning, planning computed tomography (CT) images were rigidly registered with 99mTc-GSA SPECT images, and functional liver regions were manually contoured by radiation oncologists. Using identical functional liver contours, three treatment plans—VMAT, IMRT, and 3D-CRT—were independently generated by medical physicists. The prescribed dose was 40 Gy, delivered in four fractions. These treatment plans were retrospectively created for research purposes and were separate from actual clinical treatments. Under identical planning target volume dose-coverage conditions, the dose–volume histogram indices (V2–V40) of the functional liver were compared among the three techniques. VMAT consistently demonstrated the best preservation of functional liver regions, particularly in the low-dose range, followed by IMRT and 3D-CRT. These findings demonstrate the feasibility of 99mTc-GSA SPECT-guided function-avoidance radiotherapy planning for the re-irradiation of rHCC and suggest that VMAT provides an optimal balance between target coverage and functional liver preservation.
A high-fidelity electron-source Monte Carlo model of a mobile C-arm fluoroscopy system was developed using Particle and Heavy Ion Transport code system (PHITS) to evaluate scattered radiation and calculation acceleration. Electrons were injected into the X-ray tube target, and energy spectra, dose profiles, and scattered doses around the tube head and in the room were calculated. Simulated spectra Simulated spectra agreed with the measured spectra with a root mean square error of 0.06 or less, and the simulation-to-measurement ratios of the scattered air kerma from the electron-source simulations were within 20
To evaluate the impact of five automated field-in-field (Auto FIF) priority levels in RayStation on dosimetric parameters in left-sided whole-breast radiotherapy. Ten women with early-stage left-sided breast cancer were planned with standard tangential fields and Auto FIF at priority levels 1–5 in deep inspiration breath hold. The evaluated metrics were CTV_eval D98
In radiology practice, patient body weight is important for contrast media and radiopharmaceutical dosing, radiation dose management, and examination-related workflows, yet it is not consistently available in routine clinical settings. The purpose of this study was to investigate the feasibility of estimating adult body weight using diagnostic CT dose report metrics across CT systems. This retrospective single-center study included 2496 consecutive adults who underwent diagnostic CT on three scanners. Measured body weight served as the reference standard. LightGBM regression was evaluated in three settings: (1) a baseline model using dose report–derived patient size and exposure metrics, including water-equivalent diameter and sex; (2) an extended model additionally incorporating body region and CT system; and (3) scanner-wise cross-validation to assess cross-scanner generalizability. For settings 1 and 2, data were split into training, validation, and test sets (75
Dose-Volume Histograms (DVH) discard spatial information by summarizing 3D distributions into 1D curves. This study quantifies this loss in Gamma Knife radiosurgery and introduces TopoGK, a novel unsupervised deep learning framework capturing full 3D dose geometry. Ninety-six vestibular schwannoma plans were analyzed. Dose grids were tumor-centered, resampled to 64^3 voxels, and normalized to D_max . A 3D Convolutional Variational Autoencoder compressed each dose-mask volume into a 64dimensional spatial embedding. Spatial information loss was quantified by comparing pairwise DVH and latent distances both globally and within volume-stratified subgroups. Physical validation employed hotspot center-of-mass displacement, gradient anisotropy, and a novel Spatial Discordance Index (SDI). Five-fold cross-validation ensured generalizability. A linear PCA baseline was included for comparison. The correlation between DVH and spatial similarity was weak ( =0.132;r=0.109 within volume-matched pairs). Among DVH-matched pairs, the hotspot exhibited a median physical displacement of 2.81 mm (90th percentile: 4.69 mm), and 61.8–76.5 R^2=0.621 ; volume-adjusted R^2=0.487 ) confirmed that the learned embedding is driven by spatial metrics-hotspot displacement ( =0.47 ) and anisotropy ( =0.18 ) rather than DVH ( =0.10 ). Held-out reconstruction SSIM reached 0.943± 0.018 . TopoGK outperformed PCA in spatial correlation ( r_COM= 0.451 vs. 0.287). DVH similarity does not guarantee spatial dose equivalence. The proposed framework provides a physically validated spatial fingerprint capturing geometric variations invisible to conventional plan evaluation. As a proof-of-concept in single-fraction vestibular schwannoma radiosurgery, these results motivate further investigation toward spatially-aware quality assurance in stereotactic radiosurgery.
Three-dimensional time-of-flight magnetic resonance angiography (3D TOF-MRA) is widely used for follow-up after treatment of cerebral aneurysms, but signal loss around metallic devices often limits evaluation of the parent vessel. We clinically evaluated slab-selective three-dimensional magnetic resonance angiography using radial stack-of-stars acquisition and UTE sampling (STARS UTE-MRA) in 89 patients after coil embolization and surgical clipping. STARS UTE-MRA was compared with conventional 3D TOF-MRA using full width at half-minimum (FWHM) of signal-intensity profiles and visual scores assigned by four readers. STARS UTE-MRA showed significantly smaller FWHM values and higher visual scores than 3D TOF-MRA for coils 10–15 mm in long-axis length and for all clip-length subgroups, whereas no clear advantage was observed in smaller coil cases. STARS UTE-MRA may be a useful supplementary sequence for postoperative follow-up, particularly in aneurysms treated with larger coil or surgical clips.
This study evaluated organ-specific absorbed doses and sex-dependent dosimetric variations in 15-year-old patients undergoing panoramic radiography across different units technologies. ICRP 156 mesh-type reference computational phantoms (MRCPs) were coupled with the PHITS Monte Carlo code for this analysis. Three panoramic units with distinct kinematics and exposure protocols were simulated: Unit A (fixed isocentric rotation, constant parameters) and Units B and C (sliding rotation axis, dynamic kV/mA modulation). Absolute absorbed doses were estimated using experimentally derived air kerma-area product conversion factors. Salivary glands and oral mucosa received the highest absorbed doses, ranging from 270 to 640 µGy, depending mainly on the tube voltage protocols. Sex-specific dosimetric disparities appeared due to anatomical variations interacting with beam geometry. The female MRCP absorbed 26
This study presents a comprehensive dosimetric comparison between Collaborative Ocular Melanoma Study (COMS) eye plaque containing 125I seeds and a novel seed-less design plaque of similar shape and radioactive strength using high-precision Monte Carlo Technique. Monte Carlo-based dose distributions such as central axis depth dose, lateral off-axis dose profiles and 2-dimensional isodose distributions for three different plaque diameters 10, 14 and 20 mm are calculated and compared with the dose distributions of analogous seed-less plaques. This study also evaluates the influence of different seed Models on the dose distributions across 10, 14, and 20 mm plaque diameters. The results demonstrate that the dosimetric advantage of the seed-less design is independent of the specific seed model and dependent on the diameter of the plaques. Seed-less plaque of 10 mm diameter produces comparable dose distribution with the corresponding multi-seed plaque design. Seed-less plaques of 14 mm and 20 mm diameter deliver dose rates approximately twice as high as the corresponding multi-seed plaques. For large tumours, the seed-less design demonstrates superior dose retention at the tumour apex, maintaining 23.7
