Medical physics最新文献

{"title":"Inference-specific learning for improved medical image segmentation.","authors":"Yizheng Chen, Sheng Liu, Mingjie Li, Bin Han, Lei Xing","doi":"10.1002/mp.17883","DOIUrl":"https://doi.org/10.1002/mp.17883","url":null,"abstract":"<p><strong>Background: </strong>Deep learning networks map input data to output predictions by fitting network parameters using training data. However, applying a trained network to new, unseen inference data resembles an interpolation process, which may lead to inaccurate predictions if the training and inference data distributions differ significantly.</p><p><strong>Purpose: </strong>This study aims to generally improve the prediction accuracy of deep learning networks on the inference case by bridging the gap between training and inference data.</p><p><strong>Methods: </strong>We propose an inference-specific learning strategy to enhance the network learning process without modifying the network structure. By aligning training data to closely match the specific inference data, we generate an inference-specific training dataset, enhancing the network optimization around the inference data point for more accurate predictions. Taking medical image auto-segmentation as an example, we develop an inference-specific auto-segmentation framework consisting of initial segmentation learning, inference-specific training data deformation, and inference-specific segmentation refinement. The framework is evaluated on public abdominal, head-neck, and pancreas CT datasets comprising 30, 42, and 210 cases, respectively, for medical image segmentation.</p><p><strong>Results: </strong>Experimental results show that our method improves the organ-averaged mean Dice by 6.2% (p-value = 0.001), 1.5% (p-value = 0.003), and 3.7% (p-value < 0.001) on the three datasets, respectively, with a more notable increase for difficult-to-segment organs (such as a 21.7% increase for the gallbladder [p-value = 0.004]). By incorporating organ mask-based weak supervision into the training data alignment learning, the inference-specific auto-segmentation accuracy is generally improved compared with the image intensity-based alignment. Besides, a moving-averaged calculation of the inference organ mask during the learning process strengthens both the robustness and accuracy of the final inference segmentation.</p><p><strong>Conclusions: </strong>By leveraging inference data during training, the proposed inference-specific learning strategy consistently improves auto-segmentation accuracy and holds the potential to be broadly applied for enhanced deep learning decision-making.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144016013","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Field dispersion in uniformly-excited radial parallel plate waveguides for a compact proton accelerator design.","authors":"Morgan J Maher, Christopher M Lund, Julien Bancheri, David G Cooke, Jan Seuntjens","doi":"10.1002/mp.17868","DOIUrl":"https://doi.org/10.1002/mp.17868","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Background: &lt;/strong&gt;Proton therapy (PT) is a beneficial modality for treating certain cancers but remains under utilized due in part to the high cost of existing PT devices. Dielectric wall accelerators (DWAs) are a proposed class of coreless induction accelerators that may present a suitable option for compact and affordable PT. To realize a compact device, acceleration modules must be designed to achieve field strengths approaching 100 MV/m delivered as pulses on the order of nanoseconds.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Purpose: &lt;/strong&gt;Here, we examine pulse injection into radial parallel plate waveguides as a means of producing high-intensity, pulsed accelerating fields. We present an approach for understanding the impact of waveguide properties on electromagnetic dispersion as well as a means of accounting for this dispersion to produce suitable accelerating fields.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Methods: &lt;/strong&gt;Geometric and material properties for a set of waveguides were identified based on existing literature and commonly available materials. An analytic model is presented to describe how waveguide geometry and material affect electromagnetic dispersion in a waveguide. Simulations performed in COMSOL Multiphysics are used to calculate a transfer function for the set of waveguides, which provide a means of determining the waveguides output for arbitrary inputs and vice versa. The simulation results are compared to the analytic solution and used to explore alternate matching conditions at the beampipe of the accelerator.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;Overall, radial waveguides provide a passive enhancement of the injected pulse, with enhancement of high-frequency components found to be proportional to the square root of the ratio of outer radius to inner radius of the waveguide. Dispersion in the waveguide caused by the radial propagation of the pulse depends on multiple waveguide properties (outer radius, inner radius, material) and leads to reduced enhancement at lower frequencies. The field enhancement in the waveguides reduces the peak voltage required to achieve the desired accelerating field strength. However, dispersion alters the temporal profile of the applied pulse, resulting in a distorted field at the inner radius. Using the transfer function, it is possible to determine the shape of the pulse required to achieve a suitable accelerating field for a given waveguide design.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Conclusions: &lt;/strong&gt;Passive field enhancement occurred in all waveguides and across all frequencies studied in this work. As such, radial parallel plate waveguides could help to reduce the high voltages required from upstream switching networks. The analytic model can be used to select waveguide parameters that provide a suitable enhancement of the upstream voltage pulse to achieve the high field strengths required for a compact accelerator. However, pulse dispersion must be accounted for. If upstream pulse shaping can be achieved to account for electromagnetic ","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144052527","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Impact of tracer uptake rate on quantification accuracy of myocardial blood flow in PET: A simulation study.","authors":"Xiaotong Hong, Amirhossein Sanaat, Yazdan Salimi, René Nkoulou, Hossein Arabi, Lijun Lu, Habib Zaidi","doi":"10.1002/mp.17871","DOIUrl":"https://doi.org/10.1002/mp.17871","url":null,"abstract":"&lt;p&gt;&lt;strong&gt;Background: &lt;/strong&gt;Cardiac perfusion PET is commonly used to assess ischemia and cardiovascular risk, which enables quantitative measurements of myocardial blood flow (MBF) through kinetic modeling. However, the estimation of kinetic parameters is challenging due to the noisy nature of short dynamic frames and limited sample data points.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Purpose: &lt;/strong&gt;This work aimed to investigate the errors in MBF estimation in PET through a simulation study and to evaluate different parameter estimation approaches, including a deep learning (DL) method.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Materials and methods: &lt;/strong&gt;Simulated studies were generated using digital phantoms based on cardiac segmentations from 55 clinical CT images. We employed the irreversible 2-tissue compartmental model and simulated dynamic &lt;sup&gt;13&lt;/sup&gt;N-ammonia PET scans under both rest and stress conditions (220 cases each). The simulations covered a rest K&lt;sub&gt;1&lt;/sub&gt; range of 0.6 to 1.2 and a stress K&lt;sub&gt;1&lt;/sub&gt; range of 1.2 to 3.6 (unit: mL/min/g) in the myocardium. A transformer-based DL model was trained on the simulated dataset to predict parametric images (PIMs) from noisy PET image frames and was validated using 5-fold cross-validation. We compared the DL method with the voxel-wise nonlinear least squares (NLS) fitting applied to the dynamic images, using either Gaussian filter (GF) smoothing (GF-NLS) or a dynamic nonlocal means (DNLM) algorithm for denoising (DNLM-NLS). Two patients with coronary CT angiography (CTA) and fractional flow reserve (FFR) were enrolled to test the feasibility of applying DL models on clinical PET data.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Results: &lt;/strong&gt;The DL method showed clearer image structures with reduced noise compared to the traditional NLS-based methods. In terms of mean absolute relative error (MARE), as the rest K&lt;sub&gt;1&lt;/sub&gt; values increased from 0.6 to 1.2 mL/min/g, the overall bias in myocardium K&lt;sub&gt;1&lt;/sub&gt; estimates decreased from approximately 58% to 45% for the NLS-based methods while the DL method showed a reduction in MARE from 42% to 18%. For stress data, as the stress K&lt;sub&gt;1&lt;/sub&gt; decreased from 3.6 to 1.2 mL/min/g, the MARE increased from 30% to 70% for the GF-NLS method. In contrast, both the DNLM-NLS (average: 42%) and the DL methods (average: 20%) demonstrated significantly smaller MARE changes as stress K&lt;sub&gt;1&lt;/sub&gt; varied. Regarding the regional mean bias (±standard deviation), the GF-NLS method had a bias of 6.30% (±8.35%) of rest K&lt;sub&gt;1&lt;/sub&gt;, compared to 1.10% (±8.21%) for DNLM-NLS and 6.28% (±14.05%) for the DL method. For the stress K&lt;sub&gt;1&lt;/sub&gt;, the GF-NLS showed a mean bias of 10.72% (±9.34%) compared to 1.69% (±8.82%) for DNLM-NLS and -10.55% (±9.81%) for the DL method.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Significance: &lt;/strong&gt;This study showed that an increase in the tracer uptake rate (K&lt;sub&gt;1&lt;/sub&gt;) corresponded to improved accuracy and precision in MBF quantification, whereas lower tracer uptake resulted in higher noise in dynamic PET ","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144056871","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Learned high resolution energy-integrating detector CT angiography: Harnessing the power of ultra-high-resolution photon counting detector CT.","authors":"Emily K Koons, Shaojie Chang, Andrew D Missert, Hao Gong, Jamison E Thorne, Safa Hoodeshenas, Prabhakar Shantha Rajiah, Cynthia H McCollough, Shuai Leng","doi":"10.1002/mp.17874","DOIUrl":"https://doi.org/10.1002/mp.17874","url":null,"abstract":"<p><strong>Background: </strong>Coronary computed tomography angiography (cCTA) is a widely used noninvasive diagnostic exam to assess patients for coronary artery disease (CAD). However, the spatial resolution of most CT scanners is limited due to the use of energy-integrating detectors (EIDs).</p><p><strong>Purpose: </strong>To develop a convolutional neural network (Improved LUMEN visualization through Artificial super-resoluTion imagEs (ILUMENATE)) informed by photon-counting-detector (PCD)-CT to improve EID-CT image resolution and determine its impact on cCTA.</p><p><strong>Materials and methods: </strong>With IRB approval, 30 patients undergoing clinically indicated cCTA were scanned with EID-CT (SOMATOM Force, Siemens Healthineers, Forchheim, Germany) and subsequently with ultra-high-resolution (UHR) PCD-CT (NAEOTOM Alpha, Siemens Healthineers) on the same day. ILUMENATE was trained on eight patient PCD-CT datasets (67,890 patch pairs with 90% for training (61,101), 10% reserved for validation (6,789)) and applied to 22 unseen EID-CT cases. Spatial resolution was evaluated using line profiles and percent diameter stenosis quantified with a severity score assigned. Two experienced radiologists, blinded to image type, selected preferred series and scored images for overall quality, sharpness, and noise comparing original EID-CT and ILUMENATE output.</p><p><strong>Results: </strong>Visual assessment and line profiles showed substantial resolution improvement with ILUMENATE. Percent diameter stenosis was significantly reduced (mean ± standard deviation: 4.42% ± 4.82%) using ILUMENATE (p < 0.001) with nine lesions shifting down in severity score. Readers preferred ILUMENATE images in 22/22 cases and scored ILUMENATE superiorly for overall quality, sharpness, and noise (p < 0.05).</p><p><strong>Conclusions: </strong>ILUMENATE enhanced image resolution, resulting in improved overall image quality, reduced calcium blooming artifacts, and improved lumen visibility in cCTA exams performed using EID-CT. This could potentially allow for improved accessibility to UHR image quality, allowing for more accurate assessment of CAD.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144016096","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Implementation of a novel pencil beam scanning Bragg peak FLASH technique to a commercial treatment planning system.","authors":"Alexander Bookbinder, Miriam Krieger, Pierre Lansonneur, Anthony Magliari, Xingyi Zhao, J Isabelle Choi, Charles B Simone, Haibo Lin, Michael Folkerts, Minglei Kang","doi":"10.1002/mp.17876","DOIUrl":"https://doi.org/10.1002/mp.17876","url":null,"abstract":"<p><strong>Background: </strong>Ultra-high dose rate, or FLASH, radiotherapy has shown promise in preclinical experiments of sparing healthy tissue without compromising tumor control. This \"FLASH effect\" can compound with dosimetric sparing of the proton Bragg peak (BP) using a method called Single Energy Pristine Bragg Peak (SEPBP) FLASH. However, this and other proposed FLASH techniques are constrained by lack of familiar treatment planning systems (TPSs). Creating modules to implement SEPBP FLASH into a commercial TPS opens up the possibility of more widespread investigation of FLASH and lays the groundwork for future clinical translation.</p><p><strong>Purpose: </strong>To implement, investigate, and benchmark the capacity of a commercial TPS research extension for BP FLASH SBRT treatment planning by studying the dosimetric properties and FLASH ratio for critical organs-at-risk (OARs) at several sites.</p><p><strong>Methods: </strong>A 250 MeV clinical proton beam model was commissioned in the Eclipse TPS (Varian Medical Systems, Palo Alto, USA). BP FLASH fields were single-layer maximum-energy beams with a universal range shifter (URS) and field-specific range compensators (RCs). RCs for each beam angle were included as contours within the structure set, while the URS was modeled in the PBS beamline. Spotmaps were created using Lloyd's algorithm with minimum monitor units (MU)-based spacing to ensure plan quality and preserve FLASH coverage for critical OARs. Inverse optimization while preserving minimum MU constraints was done with scorecard-based optimization. Fifteen SBRT cases from three anatomical sites (liver, lung, base-of-skull [BOS]) previously treated at the New York Proton Center were re-optimized using this method, and dosimetric characteristics of BP plans were compared to clinically treated plans. FLASH ratios for critical OARs were evaluated for BP FLASH plans.</p><p><strong>Results: </strong>The dose distributions, including target uniformity, conformity index (CI), and DVHs, showed no significant difference in clinically-used metrics between BP FLASH and clinically delivered plans across all anatomical sites. Mean 40 Gy/s FLASH ratios for critical OARs were above 84% for all but one OAR with 2 Gy threshold and above 98% for all OARs with 5 Gy threshold. D_max for liver and BOS cases was 111.3 ± 2.68 and 112.88 ± 1.29, respectively, and D_2% for lung cases was 112.04 ± 1.09. All D_max remained below 115%.</p><p><strong>Conclusions: </strong>Inverse planning using a single-energy BP FLASH technique based on sparse spots and ultra-high minimum MU/spot can achieve intensity-modulated proton therapy (IMPT)-equivalent quality and sufficient FLASH coverage. This successful prototype brings us closer to commercial implementation and may increase the availability of proton FLASH dosimetry studies.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144056873","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Advancing the evaluation of radiation-induced vaginal toxicity using ultrasound radiomics: Phantom validation and pilot clinical study.","authors":"Jing Wang, Joseph Shelton, Boran Zhou, Deborah C Marshall, Himanshu Joshi, Emi J Yoshida, Xiaofeng Yang, Tian Liu","doi":"10.1002/mp.17864","DOIUrl":"https://doi.org/10.1002/mp.17864","url":null,"abstract":"<p><strong>Background: </strong>Radiation-induced long-term toxicities, such as vaginal stenosis, severely impact the quality of life for patients undergoing pelvic radiotherapy (RT) for gynecologic (GYN) malignancies. However, current methods for assessing these toxicities rely on subjective physical examinations and patient-reported symptoms, leading to inconsistencies in grading and suboptimal management.</p><p><strong>Purpose: </strong>This pilot study investigates the potential of ultrasound-based radiomics, specifically gray level co-occurrence matrix (GLCM) texture metrics, as objective and quantitative biomarkers for evaluating long-term radiation-induced vaginal toxicity.</p><p><strong>Methods: </strong>A two-phase study was conducted. First, a phantom study was performed to identify robust GLCM texture features with low variability [coefficient of variance (COV) < 10%] across ultrasound brightness settings. In a subsequent clinical pilot study, 22 female participants were recruited: 10 had received pelvic radiotherapy (RT) with follow-up times ranging from 8 to 23 months, while 12 served as non-RT controls. All participants underwent transvaginal ultrasound imaging, and GLCM texture features were extracted for analysis. A Mann-Whitney U test was used to assess between-group differences of distribution, with a p value < 0.05 identified as statistically significance. Cohen's d values were calculated to quantify effect sizes, with a value of greater than 0.8 indicating large effects.</p><p><strong>Results: </strong>Seventeen GLCM features demonstrated robustness (COVs < 10%) across brightness settings in the phantom study, including two with COVs < 1%, 10 with COVs between 1% and 5%, and five with COVs between 5% and 10%. In the clinical study, four texture features showed significant differences between the treated group and controls (p < 0.05). Specifically, the treated group exhibited a 15.5% increase in correlation (p = 0.03), a 35.8% decrease in contrast (p = 0.03), a 10.1% decrease in difference entropy (p = 0.04), and a 17.9% decrease in dissimilarity (p = 0.07).</p><p><strong>Conclusion: </strong>This phantom and pilot study demonstrated that ultrasound GLCM features can serve as reliable quantitative biomarkers for assessing radiation-induced vaginal toxicity in female patients receiving pelvic RT for GYN cancers. Implementing these biomarkers in clinical practice could enhance the objectivity of toxicity evaluations, leading to more consistent grading and better-informed follow-up care for patients.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144033619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Small-field output factor dependence on the field size definition in MR-Linac.","authors":"Indra J Das, Ahtesham U Khan, Poonam Yadav","doi":"10.1002/mp.17857","DOIUrl":"https://doi.org/10.1002/mp.17857","url":null,"abstract":"<p><strong>Background: </strong>Radiation beam characteristics are primarily evaluated based on field size. However, in small fields, especially with magnetic fields used in new technology (MR-Linac), the field size definition is altered. Typically, field size is defined by two methods: geometric and dosimetric, which are evaluated in this study.</p><p><strong>Purpose: </strong>Small field size definitions are distorted due to lateral electron disequilibrium and the presence of magnetic fields. MR-Linac systems, which combine an MR imaging system and a linear accelerator on a single gantry, require precise evaluations of field size definitions and beam parameters, particularly for small fields. which is investigated in this study.</p><p><strong>Methods: </strong>A 0.35 T MRIdian Viewray system was evaluated using beam profiles and field output factors (FOF) with various MR-compatible microdetectors, such as ion chamber, microDiamond, microSilicon, and plastic scintillators. Validity of geometric field size (S) and dosimetric field size (S_clin) is investigated with measurements performed with MR compatible scanning water phantom at 85 cm source-to-surface distance (SSD) at a depth of 5 cm. Measured FOF data was compared with treatment planning systems (TPS) and independent Monte Carlo simulations.</p><p><strong>Results: </strong>The measured S_clin data is detector and machine-dependent, while S is machine-dependent only. The FOF was found to be a smooth function of S within experimental uncertainties, showing higher reproducibility compared to S_clin which exhibited erratic behavior.</p><p><strong>Conclusions: </strong>It is concluded that geometric field size (S) provides accurate beam characterization data, whereas S_clin may not be a reliable parameter in MR-Linac systems.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144039278","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A segmentation method for oral CBCT image based on Segment Anything Model and semi-supervised teacher-student model.","authors":"Jianhong Gan, Runqing Kang, Xun Deng, Tongli He, Nie Yu, Yuling Gan, Peiyang Wei, Xiangyi Chen, Xiaoli Peng, Zhibin Li","doi":"10.1002/mp.17854","DOIUrl":"https://doi.org/10.1002/mp.17854","url":null,"abstract":"<p><strong>Background: </strong>Accurate segmentation of oral cone beam computed tomography (CBCT) images is essential for research and clinical diagnosis. However, irregular and blurred tooth boundaries in CBCT images complicate the labeling of oral tissues, and insufficient labeled samples further limit the generalization ability of segmentation models. The Segment Anything Model (SAM) demonstrates strong generalization and segmentation accuracy across diverse tasks as a vision foundation model. The Teacher-Student (TS) model has proven effective in semi-supervised learning approaches.</p><p><strong>Purpose: </strong>To accurately segment various parts of oral CBCT, such as enamel, pulp, bone, blood vessels, air, etc., an improved segmentation method named SAM-TS is proposed, which combines SAM with the TS model. SAM-TS leverages Low-Rank Adaptation (LoRA) to fine-tune the SAM model on oral CBCT images with fewer parameters.</p><p><strong>Methods: </strong>To efficiently utilize numerous unlabeled images for training models, the LoRA strategy is improved to fine-tune the SAM. The fine-tuned SAM and teacher models collaboratively generate pseudo-labels on unlabeled images, which are filtered and utilized to train the student model. Then, a data augmentation-based Mean Intersection over Union (MIoU) method is proposed to filter out unreliable or spurious pseudo-labels. Finally, the Exponential Moving Average (EMA) method is used to transfer the student model's parameters to the teacher model. After repeating this process, the final optimized student model for segmentation is obtained. The experimental results demonstrate that incorporating unlabeled data into model training through SAM-TS significantly enhances the model's generalization ability and segmentation accuracy.</p><p><strong>Results: </strong>Compared to the baseline algorithm, the proposed method achieves an overall improvement of over 6.48% in MIoU. In the tooth segmentation task, the minimum MIoU and maximum MIoU increased by at least 10% and 27.32%, respectively. In the bone segmentation task, the minimum MIoU and maximum MIoU increased by 7.9% and 32.44%, respectively. Additionally, for overall segmentation, the Hausdorff distance (HD) decreased by 5.1 mm, and the Dice coefficient increased by 2.87%.</p><p><strong>Conclusion: </strong>SAM-TS outperforms existing semi-supervised methods, offering a more competitive and efficient approach to CBCT image segmentation. This method addresses the data annotation bottleneck and opens new avenues for semi-supervised learning applications in medical imaging.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144061233","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An efficient method to evaluate the mechanical accuracy of MR-Linac configured with an MV panel.","authors":"Zhihui Hu, Hui Yan, Ke Zhang, Peng Huang, Yuan Xu, Jianrong Dai, Kuo Men","doi":"10.1002/mp.17843","DOIUrl":"https://doi.org/10.1002/mp.17843","url":null,"abstract":"<p><strong>Background: </strong>The mechanical accuracy of magnetic resonance linear accelerators (MR-Linac) is crucial in terms of the accuracy of magnetic resonance-guided radiotherapy. Current clinical quality assurance procedures, which involve individual measurements of the accelerator's mechanical components, have high time costs.</p><p><strong>Purpose: </strong>This study developed an efficient method to evaluate the mechanical accuracy of the MR-Linac by measuring multiple mechanical components simultaneously using a single phantom with only one setup.</p><p><strong>Methods: </strong>The measurements were performed using an MR-to-MV phantom with an Elekta Unity MR-Linac. The phantom contains regularly arranged ceramic ball bearings (BB) that are visible in megavoltage (MV) images. MV projection images of the phantom were acquired at various gantry angles, and a software program was developed to detect the radiation field edges and the positions of the BBs, thereby enabling effective and efficient measurement of the radiation isocenter size, field size accuracy, couch position accuracy, and gantry angle accuracy. The accuracy, reproducibility and robustness of the proposed method were evaluated through tests with different gantry angles and phantom offsets.</p><p><strong>Results: </strong>The entire measurement procedure was completed in 6.3 ± 0.2 min, and the obtained results were consistent with those of conventional methods. The proposed method can detect angular uncertainties as small as 0.1°. The measurement results exhibited excellent inter-operator reproducibility, with the intraclass correlation coefficient >0.9 and standard deviations within 0.1 mm and 0.1°. In the robustness test, introducing a 2 mm phantom setup error resulted in an average deviation of 0.02 mm in the measured radiation isocenter size and a maximum deviation of approximately 0.1° in the gantry angle.</p><p><strong>Conclusion: </strong>The proposed is a simple, robust, and accurate tool to measure the mechanical accuracy of an MR-Linac. By enabling simultaneous measurement of multiple mechanical parameters using a single phantom, the proposed method reduces the time costs of quality assurance procedures considerably.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144040234","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Modeling inter-reader variability in clinical target volume delineation for soft tissue sarcomas using diffusion model.","authors":"Yafei Dong, Thibault Marin, Yue Zhuo, Elie Najem, Arnaud Beddok, Laura Rozenblum, Maryam Moteabbed, Kira Grogg, Fangxu Xing, Jonghye Woo, Yen-Lin E Chen, Ruth Lim, Xiaofeng Liu, Chao Ma, Georges El Fakhri","doi":"10.1002/mp.17865","DOIUrl":"https://doi.org/10.1002/mp.17865","url":null,"abstract":"<p><strong>Background: </strong>Accurate delineation of the clinical target volume (CTV) is essential in the radiotherapy treatment of soft tissue sarcomas. However, this process is subject to inter-reader variability due to the need for clinical assessment of risk and extent of potential microscopic spread. This can lead to inconsistencies in treatment planning, potentially impacting treatment outcomes. Most existing automatic CTV delineation methods do not account for this variability and can only generate a single CTV for each case.</p><p><strong>Purpose: </strong>This study aims to develop a deep learning-based technique to generate multiple CTV contours for each case, simulating the inter-reader variability in the clinical practice.</p><p><strong>Methods: </strong>We employed a publicly available dataset consisting of fluorodeoxyglucose positron emission tomography (FDG-PET), x-ray computed tomography (CT), and pre-contrast T1-weighted magnetic resonance imaging (MRI) scans from 51 patients with soft tissue sarcoma, along with an independent validation set containing five additional patients. An experienced reader drew a contour of the gross tumor volume (GTV) for each patient based on multi-modality images. Subsequently, two additional readers, together with the first one, were responsible for contouring three CTVs in total based on the GTV. We developed a diffusion model-based deep learning method that is capable of generating arbitrary number of different and plausible CTVs to mimic the inter-reader variability in CTV delineation. The proposed model incorporates a separate encoder to extract features from the GTV masks, leveraging the critical role of GTV information in accurate CTV delineation.</p><p><strong>Results: </strong>The proposed diffusion model demonstrated superior performance with the highest Dice Index (0.902 compared to values below 0.881 for state-of-the-art models) and the best generalized energy distance (GED) (0.209 compared to values exceeding 0.221 for state-of-the-art models). It also achieved the second-highest recall and precision metrics among the compared ambiguous image segmentation models. Results from both datasets exhibited consistent trends, reinforcing the reliability of our findings. Additionally, ablation studies exploring different model structures and input configurations highlighted the significance of incorporating prior GTV information for accurate CTV delineation.</p><p><strong>Conclusions: </strong>The proposed diffusion model successfully generates multiple plausible CTV contours for soft tissue sarcomas, effectively capturing inter-reader variability in CTV delineation.</p>","PeriodicalId":94136,"journal":{"name":"Medical physics","volume":" ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-05-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144013547","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}