Accurate needle placement is essential for prostate biopsy. Recently, transperineal prostate biopsies are receiving renewed interest due to concern over infection from conventional transrectal biopsies. However, accurate needle placement is more challenging in the transperineal approach than in the transrectal approach due to the long insertion distance leading to a large targeting error and repeated insertion attempts. Improved procedure planning tools that can predict the deviation of the needle can potentially reduce the targeting error and number of insertion attempts. Prediction of deflection magnitude requires a model of biopsy needle deflection, which in turn requires information about tissue material properties. However, material properties of tissue in patients cannot be easily obtained. Accounting for this uncertainty in patient tissue properties requires a model capable of quantifying uncertainty in needle deflection as a function of a distribution of tissue properties. A Monte Carlo uncertainty quantification requires 1000s of samples, but it is not possible to obtain this many samples in a short enough time for intraoperative procedure planning using published needle deflection predictionmodels. This work seeks to develop a model of needle deflection fast enough for use in intraoperative procedure planning, validate this model against experimental results, and integrate it into a Monte Carlo uncertainty quantificationmodel. This work used a mechanics-based model of biopsy needle deflection to train a Fourier feature neural network (FFNN) model in order to make predictions with a low computational cost. Both models were validated against experimental data. The neural network model was used in a Monte Carlo uncertainty quantification model to quantify uncertainty in needle deflection arising from uncertain tissue mechanicalproperties. This work (1) implemented a mechanics-based model and a FFNN model. Both models were validated against previously published experiments carried out with tissue phantoms. Both models showed close agreement with the experimental data. (2) We showed that our FFNN model was more accurate than a baseline ordinary least squares model, introducing only about 0.3-mm tip deflection error compared to the mechanics-based model. We also showed that our FFNN model makes unbiased predictions with respect to the amount of deflection. (3) We demonstrated a Monte Carlo uncertainty quantification model of needle deflection with a low computational cost of about 20 CPU s. We used our uncertainty quantification model to show how the depth, stiffness, and magnitude of uncertainty in a layer of tissue affect needle deflection. In addition, we showed a simple clinical example of the use of our model. This work demonstrates a Monte Carlo uncertainty quantification model of needle deflection with a low computational cost. This method shows promise for future applications in procedure planning for prostate biopsies as well as other transperineal procedures conducted with flexible needles such as cryoablation andbrachytherapy.

BackgroundDiagnostic ultrasound is rapidly evolving, and the increasing complexity of ultrasound systems underscores the importance of robust quality assurance (QA) and quality control (QC) methods. Current methods typically rely on manual image acquisition and subjective evaluation, making them operator‐dependent, poorly reproducible, and challenging for longitudinal tracking. Efforts have been made to develop automated ultrasound QA/QC methods for more quantitative and reproducible outcomes. One such approach utilized the Random Hypoechoic Sphere Phantom (RHSP) to evaluate the detectability of ultrasound systems by measuring the human observer‐related Lesion Signal‐to‐Noise Ratio (LSNR). A practical manual scanning method was developed for the RHSP, which eliminated the need for a mechanical guide to control transducer translation. However, this approach requires maintaining a uniform translation speed and a small translation distance between successive images. Broader adoption of the RHSP also requires easily accessible automated analysis software.PurposeTo improve the practicality of using the RHSP for ultrasound performance evaluation in clinical settings, we developed an automated analysis method tailored for freehand scanning. We validated the method by evaluating ultrasound system performance under various acquisition settings, identifying sources of variability, and analyzing their impact on LSNR measurements.MethodsThe RHSP consisted of 2 mm diameter spheres at 20% volume fraction. It was scanned by moving the transducer by freehand during a cine‐loop acquisition. Volumetric images were generated by stacking the elevational frames. After denoising and homogeneity correction, the spheres were segmented in 3D using depth‐adaptive thresholds from a Gaussian‐Mixture Model. LSNR was computed for each segmented sphere, and LSNRs at similar depth were averaged to generate an LSNR vs. depth curve.The algorithm was validated by computing the LSNR vs. depth and sphere‐count vs. depth curves for various acquisition settings, with the following factors changing one at a time: nominal frequency, imaging modes and compound techniques, transmit power, gain, dynamic range, and transmit focal depth. Variability of the algorithm was assessed using repeated scans performed by different operators and at various transducer translation speeds.ResultsAll LSNR results closely matched visual assessments and followed expectations when changing acquisition settings. LSNR values improved with higher nominal frequency, harmonic imaging with compound, higher transmit power, and decreased dynamic range; gain had minimal effect. The best LSNR also aligned with the transmit focal depths set by the operator. The variabilities of LSNR were from 2.5% to 6.9% intra‐operator, 5.2% inter‐operator, and 5.5% due to transducer translation speed. These variabilities of the LSNR were much smaller than those of the corresponding sphere counts, demonstrating the proposed algorithms’ robustness against potential variations produced by freehand scanning.ConclusionsThe automated analysis method for the RHSP with freehand scanning provides an accurate and stable quantitative assessment of lesion detectability for evaluating ultrasound performance, and is feasible in clinical settings. It demonstrated robustness by achieving approximately 5% variability in LSNR across various transducer translation speeds and different operators. With such robustness, this approach simplifies ultrasound performance evaluation and holds promise for integration into routine ultrasound QA/QC in a clinical environment.

Processing DICOM datasets for research and education is challenging due to the format's complexity and frequent patient-specific workflow exceptions. Proper handling demands substantial technical expertise and meticulous care to ensure data fidelity in downstream applications. We developed MedicalDataHandler to streamline the reading, visualization, and processing of DICOM data. By consolidating essential tasks into a user-friendly environment, it minimizes reliance on advanced coding skills and promotes reproducible data handling without custom scripting. Implemented in Python with the third-party Dear PyGui toolkit, MedicalDataHandler organizes DICOM files by patient identifiers and groups each patient's radiation therapy (RT) images, structure sets, plans, and doses based on mutual unique identifiers (UIDs). A comprehensive table of patient data enables metadata inspection, data visualization, and data processing. The GUI supports interactive visualization in axial, coronal, and sagittal views, with intuitive scrolling, zooming, panning, and window width/level adjustments. Segmentation labels, colors, and data orientation can be modified on the fly, and hovering over a voxel reveals its image/dose values and relevant segmented structures. Multithreading and multiprocessing enable rapid data reading and conversion to the deep-learning-friendly Nearly Raw Raster Data (NRRD) format. Additional features include metadata inspection, voxel grid resampling, Hounsfield-Unit-to-Relative-Electron-Density mapping, plan-sum dose creation, and partial or bulk data saving options. We validated MedicalDataHandler with an end-to-end testing approach. DICOM data from 61 radiotherapy patients were processed, and the resulting dataset was used to train a deep-learning-based dose prediction model. MedicalDataHandler streamlined the workflow by eliminating the need for complex, patient-specific code and accelerating the preparation of a research-ready dataset. MedicalDataHandler streamlines DICOM data management and accelerates preprocessing, serving as a valuable tool for researchers and trainees. Its intuitive interface, flexible editing, and rapid data conversion empower a broader audience to manage DICOM data efficiently and consistently in research and education settings.

Automatic radiotherapy planning based on deep learning-based dose prediction shows promise in improving planning efficiency, consistency, and quality. However, most existing studies have focused on scenario-specific dose prediction models, thereby limiting the widespread clinical application of such algorithms. This study aims to develop a generalizable dose prediction model capable of predicting dose distributions across various tumor sites, delivery techniques, and prescription doses, with high accuracy. Furthermore, we seek to demonstrate the feasibility of universal automatic planning guided by this generalizable dose prediction model. Model development and evaluation were conducted using the GDP-HMM dataset, comprising 3234 plans encompassing different tumor sites (head and neck cancer, lung cancer), delivery techniques (IMRT, VMAT), and prescription doses. The proposed dose prediction model, based on the 3D MedNeXt architecture, was trained on a subset of 2878 plans. Physics-informed priors were incorporated by designing model inputs such as normalized distance-aware beam plates and mass density maps to enhance performance. Large-kernel convolutions, combined with the UpKern initialization strategy, were employed to expand the model's receptive field and improve prediction accuracy. Model generalizability and accuracy were assessed on the remaining 356 plans. Ablation studies were performed to evaluate the impact of physics-informed priors and large-kernel convolutions. To demonstrate the feasibility of universal automatic planning, predicted dose distributions for four representative test cases were converted into deliverable plans using fallback planning process in the RayStationTM (RaySearch Laboratories, Stockholm, Sweden) treatment planning system. The proposed model accurately predicted dose distributions for plans with varying tumor sites, delivery techniques, and prescription doses, achieving median prediction errors within 2 percentage points (pp) for all assessed dose-volume histogram (DVH) metrics. The mean absolute prediction errors across all DVH metrics were 2.13 pp for head and neck plans and 1.65 pp for lung plans. Ablation studies revealed that incorporating mass density maps and normalized distance-aware beam plates as model inputs significantly reduced body mean absolute errors (MAE) by 2.1% (p<0.001) and 5.8% (p<0.001), respectively. Increasing the convolution kernel size from 3 to 5 with the UpKern initialization strategy further reduced body MAE by 2.1% (p<0.001). The model achieved second place in the AAPM GDP-HMM Challenge. Deliverable plans generated from predicted dose distributions for four representative cases of various clinical scenarios demonstrated quality comparable to reference plans. We have developed a generalizable and accurate dose prediction model capable of predicting dose distributions for diverse tumor sites, delivery techniques, and prescription doses. The integration of physics-informed priors and large-kernel convolutions significantly enhanced prediction accuracy. The feasibility of universal automatic planning based on the generalizable dose prediction model has been demonstrated, highlighting its potential to facilitate broader clinical application of automatic planning and to improve planning efficiency, consistency, and quality. The code and trained models are publicly available at https://github.com/tyxiong123/RTPDosePred.

Magnetoacoustic tomography with magnetic induction (MAT-MI) is a promising noninvasive, radiation-free imaging technique capable of millimeter-level spatial resolution for mapping tissue conductivity. However, its application to luminal tissues (e.g., digestive tract, vasculature) is significantly hindered by low image quality due to acoustic wave attenuation and inherent tissue properties. This study aimed to overcome the image quality limitation in MAT-MI imaging of luminal structures by utilizing liquid metal (LM) as a novel conductive contrast agent and implementing M-sequence coded excitation to enhance signal strength and acquisition efficiency. The LM contrast agent used was a biocompatible Ga67In20.5Sn12.5 alloy (σ=3.1×10⁶ S/m). MAT-MI experiments were conducted on gel phantoms containing LM in various shapes and simulated luminal structures (PVC tubes, ex vivo mouse intestines), ex vivo mouse stomach tissues infused with LM or PBS (control), and in vivo mouse stomachs. The MAT-MI system employed 0.34 T static magnetic field, a pulsed excitation coil (1MHz center frequency), and an ultrasound transducer. We applied M-sequence coded excitation (up to 31-bit) and processed the signals using pulse compression followed by filtered back-projection reconstruction. Safety was assessed via long-term biocompatibility studies, blood elemental analysis, and histological examination (H&E staining) of gastrointestinal tissues. Integrating LM into the imaging target greatly increased the MAT-MI signal intensity. In vivo imaging of mouse stomachs demonstrated an approximate 28dB increase in image quality after LM infusion compared to pre-infusion imaging. LM enabled clear visualization within fine luminal structures (down to 0.5mm inner diameter) in phantoms and ex vivo tissues. M-sequence coding further improved image clarity and reduced total imaging time by approximately 84% relative to single-pulse excitation (from 6360s to 1020s for comparable image quality). High-resolution imaging (approximately 2mm spatial resolution) of LM distribution was achieved. The LM was safely excreted, with no significant toxicity observed in blood analysis or histology over 180 days. LM proved to be an effective and safe contrast agent for MAT-MI, significantly enhancing image quality and enabling high-quality imaging of luminal tissues, including the first successful in vivo visualization of the stomach in a live animal model. Combined with M-sequence coded excitation, this approach overcomes key limitations of conventional MAT-MI and could broaden its diagnostic utility in cardiovascular and gastrointestinal imaging.

Low-dose computed tomography (LDCT) has been widely adopted in clinical imaging to reduce radiation exposure. However, the inherent quantum noise and streaking artifacts in LDCT markedly degrade image quality, thereby compromising diagnostic accuracy. While conventional model-based iterative denoising (MBIR) approaches effectively mitigate noise via rigorous physical modeling, their adoption is hindered by substantial computational overhead. Deep learning-based approaches demonstrate strong denoising capabilities but face challenges in generalizing across imaging scanners and protocols. In this study, we propose the Dual-Interactive Fusion Network framework (DIFNet) for LDCT images, integrating the Dual-Phase Denoising Architecture (DPDA), Context-Aware Training Strategy (CATS), and a combined dual-phase loss function. On two in‑house LDCT datasets acquired with different Philips scanners, our approach reduces noise and artifacts while preserving essential anatomical detail and outperforms established denoising methods such as REDCNN, EDCNN, DDPM, and CTformer in both qualitative evaluation and quantitative metrics. Evaluation on the public Mayo-2016 benchmark, collected using a Siemens scanner, confirms DIFNet's robust and competitive performance. Ablation experiments further validate the effectiveness of our overall network design and the contribution of its core components. Our findings highlight the potential of DIFNet for real-world clinical applications, improving diagnostic reliability and patient care. The proposed framework advances LDCT denoising by balancing performance, robustness, and computational efficiency.

Implicit neural representation (INR), particularly in combination with hash encoding, has recently emerged as a promising approach for computed tomography (CT) image reconstruction. However, directly applying INR techniques to 3D dental cone-beam computed tomography (CBCT) with a truncated field of view (FOV) is challenging. During the training process, if the FOV does not fully encompass the patient's head, a discrepancy arises between the measured projections and the forward projections computed within the truncated domain. This mismatch leads the network to estimate attenuation values inaccurately, producing severe artifacts in the reconstructedimages. This study aims to develop a computationally efficient INR-based reconstruction framework that leverages multi-resolution hash encoding for 3D dental CBCT with a truncatedFOV. To mitigate truncation artifacts, we train the network over an expanded reconstruction domain that fully encompasses the patient's head. For computational efficiency, we adopt an adaptive training strategy that uses a multi-resolution grid: finer resolution levels and denser sampling inside the truncated FOV, and coarser resolution levels with sparser sampling outside. To maintain consistent input dimensionality of the network across spatially varying resolutions, we introduce an adaptive hash encoder that selectively activates the lower-level features of the hash hierarchy for points outside the truncatedFOV. The proposed method with an extended FOV effectively mitigates truncation artifacts. Compared with a naive domain extension, using fixed resolution levels and a fixed sampling rate, the adaptive strategy reduces computational time by 60% for an image volume of , while preserving the peak signal-to-noise ratio (PSNR) within the truncatedFOV. We propose an INR-based reconstruction framework for 3D dental CBCT with a truncated FOV, effectively reducing truncation artifacts and trainingcost.

BackgroundCoronary artery calcium (CAC) scoring requires a non‐contrast scan in addition to contrast‐enhanced coronary CT angiography (CCTA), increasing radiation dose and scan time. Virtual non contrast (VNC) methods from spectral CT may obviate the CAC CT scan, but available methods underestimate CAC scores and show low detectability. Although VNC‐based CAC scoring is feasible via conversion factors, it misses low‐density calcifications, requiring improved VNC methods.PurposeThis study evaluates a self‐developed CAC‐preserving VNC (TrueCa) method to improve detectability and quantification of CAC on CCTA.MethodsAn artificial hollow coronary artery containing five cylindrical calcifications of varying density calcium hydroxyapatite (75–800 mg/cc) was filled with 0%, 50%, 100% (400 Hounsfield units) and 150% of a clinical iodine concentration. The phantom was scanned on dual‐layer spectral CT with a CAC scoring and CCTA protocol with five repetitions. CAC CT reconstruction followed Agatston protocol. CCTA scans were reconstructed at thick (3 mm) and thin (0.67 mm) slices using (1) commercially available VNC and (2) self‐developed TrueCa algorithm, generated via linear separation of spectral data. Non‐overlapping 95% confidence intervals between Agatston scores (AS) from the CAC protocol and CCTA protocol indicated significant differences. Volume scores (VS) were compared to physical calcification volume.ResultsFor AS, thick‐slice VNC CCTA underestimated AS by 30%–65% across all CAC densities, failing to detect CAC below 400 mg/cc. TrueCa CCTA detected all calcifications without significant difference with CAC CT for all CAC densities. Thin‐slice compared to thick‐slice TrueCa significantly increased AS in low‐density CAC.For VS, TrueCa CCTA showed less underestimation than VNC CCTA across all densities except for highest‐density calcifications. VS accuracy improved by up to 40% using TrueCa over VNC in medium to lowest‐density calcifications.ConclusionIn this phantom study, TrueCa CCTA demonstrated superior performance in calcium detection and quantification across most CAC densities, showing potential as a reliable alternative to CAC CT.

Accurate dose calculation is essential for securing the therapeutic benefit and safety of carbon ion radiotherapy. However, the uncertainty associated with tissue parameter estimation methods for Monte Carlo-based dose calculation has not been systematically evaluated, potentially limiting the reliability of quality assurance in clinical practice. To support more accurate treatment planning and quality assurance for carbon ion radiotherapy, this study aims to systematically assess how upstream uncertainties in tissue parameter estimation affect Monte Carlo-based carbon ion dose calculations. By comparing three elemental decomposition approaches, their impact on the consistency and reliability of dose distributions were quantitatively evaluated. Relevant sources of uncertainties in physical density and elemental composition were analyzed and propagated based on three methods: single-energy CT (SECT), parameterized dual-energy CT (PA-DECT), and machine learning-based DECT (ML-DECT). For validation, the ICRP 110 phantom was perturbed accordingly, and dose distributions were simulated using FLUKA with 10 samples for each method. Both physical and biological doses were evaluated. The voxel-wise uncertainty, gamma passing rate and range uncertainty were examined. The ML-DECT method consistently yielded the lowest uncertainties across all tissue parameters, reducing the uncertainties for C, N, and O density by up to 77.4% compared to SECT. In dose simulations, ML-DECT achieved lower average relative uncertainties (∼5% physical, ∼7%-9% biological), a higher gamma passing rate (97.96±0.28%), and reduced range uncertainty (0.5%) compared to other methods. By more accurately modeling the relationship between CT numbers and tissue parameters, ML-DECT substantially reduces uncertainty in both input estimation and subsequent dose calculation. These findings support its integration into Monte Carlo-based carbon ion treatment planning workflows, contributing to more quantitative and uncertainty-aware quality assurance.