Objective. Accurate segmentation of the prostate and dominant intraprostatic lesions (DILs) on magnetic resonance imaging (MRI) is important for prostate cancer radiation therapy treatment planning and targeted dose escalation. However, DIL segmentation remains challenging due to small datasets, institutional bias, and variable imaging protocols. Although the segment anything model (SAM) has shown promise in medical image segmentation, most prior work depends on manual prompts. This study developed a fully automated pipeline that combines localization with a fine-tuned SAM model to segment the prostate and DIL.Approach. Two datasets were utilized: the PI-CAI dataset, comprising 1476 patients, and the cancer imaging archive dataset, comprising 803 patients. The pipeline consisted of two stages: (1) a reinforcement learning-based localization network predicted bounding boxes as segmentation inputs, and (2) a fine-tuned SAM model performed segmentation. Model performance was evaluated using the dice similarity coefficient (DSC), intersection over union (IoU), and detection rates, with additional analysis based on lesion volumes.Main results. The proposed method achieved a mean and median DSC of 0.896 ± 0.070 and 0.915, and an IoU of 0.818 ± 0.100 and 0.844 for prostate segmentation. For DIL segmentation, the mean and median DSC were 0.592 ± 0.192 and 0.636, IoU of 0.446 ± 0.190 and 0.466, with a detection rate of 89%. Four DIL groups were created based on lesion volume percentile. The mean/median DSC and IoU for each volume group are as follows: 0.5-1.0 cubic centimeters (cc): 0.555 ± 0.201/0.562 & 0.414 ± 0.205/0.391; 1.0-1.8 cc: 0.603 ± 0.185/0.660 & 0.454 ± 0.180/0.492; 1.8-4.0 cc: 0.588 ± 0.183/0.627 & 0.439 ± 0.174/0.456; >4.0 cc: 0.621 ± 0.197/0.669 & 0.477 ± 0.197/0.503.Significance. This study presented a fully automated prostate and DIL segmentation framework on MRI by integrating a localization network with fine-tuned SAM. The method achieved robust performance across large multi-institutional datasets and diverse lesion shapes. It shows strong potential for application to clinical workflows for prostate cancer radiation therapy planning and treatment.

Purpose Source step size is a key parameter in brachytherapy (BT) treatment planning, influencing dwell time and dose distribution. This study aims to evaluate the impact of different source step sizes on dosimetry and organ-at-risk (OAR) sparing in high-dose-rate (HDR) brachytherapy for gynecological (GYN) cancers. Materials and Methods Patients with various GYN cancers underwent HDR brachytherapy with either a single-channel cylinder or a tandem and ovoids (T&O) applicator. Cylinder diameters of 20, 25, 30, and 35 mm were used, with activation lengths of 25 mm and 40 mm, ensuring coverage while minimizing mucosal surface dose and improving dose homogeneity. For T&O, applicator size was selected based on patient anatomy, with a representative configuration of a 70 mm tandem and 30 mm ovoids analyzed.Treatment planning was performed using CT-based localization in Oncentra (Nucletron), generating 40 plans across different cylinder sizes and T&O, with source step sizes ranging from 2 mm to 5 mm. Dose calculations were performed using a 1 mm grid size. Cylinder plans were optimized using point-based optimization with a 7.0 Gy prescription at 5 mm from the applicator surface, while T&O plans were optimized using Point A normalization. OAR dose constraints (D2cc, D1cc, and D0.1cc) for the bladder, rectum, and bowel were analyzed across different step sizes. Percentage differences were calculated with reference to 2 mm step size plan for D2cc and averaged over different cylinder sizes. We calculated standard deviation (std) of percentage difference between plans to evaluate step size effect. Results For cylinder applicators, dose constraints were higher for the 40 mm activation length compared to the 25 mm activation length. Overall, step size had minimal impact on OAR sparing in cylinder-based brachytherapy. However, for the largest cylinder size, increasing step size was associated with a decrease in bowel dose but a slight increase in rectum dose, while bladder dose remained relatively unchanged. Bladder D2cc were relatively stable across step sizes for all cylinder diameters and std were 0.61% and 0.34% over the step sizes for 25- and 40-mm activation length. Rectum D2cc showed a slight increase with larger step sizes, and std of 0.78% and 0.93% for 25- and 40-mm activation length respectively, respectively. Bowel dose D2cc were relatively consistent for 25 cm activation length with increasing step size, especially in larger cylinders 35 mm, and decreased with larger step sizes for 40 mm activation length. The std were 0.04% and 1.16% for 25- and 40-mm activation length, respectively.For T&O, step size had a similar effect on OARs, with minimal differences in bladder and rectum doses. Across all step sizes, HR-CTV coverage remained consistent, with std less than 1% between different steps size. Bladder, rectum and bowel doses increased slightly with std of 0.24%, 1.10%, and 0.76% for 70 mm tandem length and 1.28%, 1.01%, 1.29% for 40 mm tandem length, respectively. Conclusion Step size variations had limited impact on organ sparing and HR-CTV coverage in HDR GYN brachytherapy. While larger step sizes may slightly affect specific OAR doses in certain cases, overall dosimetric differences were minimal, supporting flexibility in step size selection based on clinical workflow and efficiency considerations.

7 Tesla (7T) apparent diffusion coefficient (ADC) maps derived from diffusion-weighted imaging (DWI) demonstrate improved image quality and spatial resolution over 3 Tesla (3T) ADC maps. However, 7T magnetic resonance imaging (MRI) currently suffers from limited clinical unavailability, higher cost, and increased susceptibility to artifacts. To address these issues, we propose a hybrid CNN-transformer model to synthesize high-resolution 7T ADC maps from multi-modal 3T MRI. The Vision CNN-Transformer (VCT), composed of both Vision Transformer (ViT) blocks and convolutional layers, is proposed to produce high-resolution synthetic 7T ADC maps from 3T ADC maps and 3T T1-weighted (T1w) MRI. ViT blocks enabled global image context while convolutional layers efficiently captured fine detail. The VCT model was validated on the publicly available Human Connectome Project Young Adult dataset, comprising 3T T1w, 3T DWI, and 7T DWI brain scans. The Diffusion Imaging in the Python library was used to compute ADC maps from the DWI scans. A total of 171 patient cases were randomly divided: 130 training cases, 20 validation cases, and 21 test cases. The synthetic ADC maps were evaluated by comparing their similarity to the ground truth volumes with the following metrics: peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and mean squared error (MSE). The results are as follows: PSNR: 27.0+-0.9 dB, SSIM: 0.945+-0.010, and MSE: 2.0+-0.4E-3. Our predicted images demonstrate better spatial resolution and contrast compared to 3T MRI and prediction results made by ResViT and pix2pix. These high-quality synthetic 7T MR images could be beneficial for disease diagnosis and intervention, especially when 7T MRI scanners are unavailable.

BackgroundRadiation Oncology (RO) residents are required to pass board exams covering radiation physics, radiation biology and clinical knowledge prior to completion of residency and board certification. Radiation physics material is commonly delivered in didactic format covering material based on recommendations from the American Society for Radiation Oncology (ASTRO) and American Board of Radiology (ABR). While multiple studies highlight the benefits of active learning (AL) techniques in a classroom setting (e.g. quizzes, open discussions, and case-based review), AL strategies have not been readily implemented into curriculum. This project's scope is to incorporate AL into the resident's radiation physics curriculum to help augment resident learning and retention.MethodsAn AL workgroup comprised of a team of medical physicists and a RO resident was created to transform the current traditional lecture-based physics curriculum into a "flipped classroom" model using AL techniques. The flipped classroom model required residents to review provided physics materials as well as self-guided questions pertinent to the covered topic(s) prior to weekly lectures. ABR exam styled multiple choice questions were imbedded into lecture slides with answer choices linked to polling software (polleverywhere.com) used to engage learners. After each question, resident responses were assessed, and a content slide was reviewed to explain and discuss the tested concept. After the initial block of physics lectures (6 out of 32), a 5-point Likert scale and free-form content feedback survey was given to current residents to assess the flipped-classroom format and implementation. Additional survey assessments included the efficient and effective coverage of course content, alignment of out-of-class material with topics covered in class, and resident comprehension.ResultsBased on resident Likert scale responses (n=10), after the initial AL lectures, residents felt more confident in the radiation physics material (3.9/5). Additionally, by reviewing content prior to lectures, and then actively participating in the AL lectures, residents felt they better understood and retained the lecture material (3.7/5).DiscussionBased on the survey results, the implementation of an AL learning environment (e.g. flipped classroom with interactive polling) was an overall positive experience. Residents' perceived comprehension of radiation physics lecture material, in-class engagement, and retention was improved. The AL format for teaching high-level curriculum to residents could effectively be expanded to other subjects pertaining to RO resident training. Beyond the use of in-class polling software, other AL methods such as open discussions, workshops, and case-based review could be implemented by educators to expand upon in-class engagement.

Purpose/Objective(s) Radiation Oncology (RO) residents are required to pass American Board of Radiology (ABR) qualifying exams covering medical physics, radiation and cancer biology, and clinical radiation oncology prior to completion of residency and board certification. Medical physics material is commonly delivered in didactic format covering material based on recommendations from the American Society for Radiation Oncology (ASTRO) and ABR. While multiple studies highlight the benefits of active learning (AL) techniques in a classroom setting (e.g. quizzes, open discussions, and case-based review), AL strategies have not been readily implemented into the RO medical physics curriculum. This project's scope is to incorporate AL into residents’ medical physics curriculum to help augment resident learning and retention. Materials/Methods An AL workgroup comprised of a team of medical physicists and a RO resident was created to transform the current traditional lecture-based medical physics curriculum into a “flipped classroom” model using AL techniques. The flipped classroom model required RO residents to review lecturer curated physics materials as well as self-guided questions pertinent to the covered topic(s) prior to weekly lectures. ABR exam styled multiple choice questions were imbedded into lecture slides with answer choices linked to polling software (polleverywhere.com) used to engage learners. After each question, polled resident responses were assessed, and a content slide was reviewed to explain and discuss the tested concept. After the first block of 6 medical physics lectures (6 out of 32), a 5-point Likert scale and free-form content feedback survey was given to current residents to assess the flipped-classroom format and implementation. Additional survey assessments included the efficient and effective coverage of course content, alignment of out-of-class material with topics covered in class, and resident comprehension. Results Based on resident Likert scale responses (n = 10 out of 10), after the initial AL lectures, residents felt more confident in the medical physics material (3.9/5). Additionally, by reviewing content prior to lectures, and then actively participating in the AL lectures, residents felt they better understood and retained the lecture material (3.7/5). Conclusion Based on the initial survey results, the implementation of an AL learning environment (e.g. flipped classroom with interactive polling) was an overall positive experience. Residents’ perceived comprehension of medical physics lecture material, in-class engagement, and retention was improved. The AL format for teaching high-level curriculum to residents could effectively be expanded to other subjects pertaining to RO resident training. Beyond the use of in-class polling software, other AL methods such as open discussions, workshops, and case-based review could be implemented by educators to expand upon in-class engagement.

Purpose/Objective(s) Peer review among Radiation Oncology physicians is an essential step in the quality management of a patient’s treatment. Recently, our multi-site department converted from a centralized chart rounds setting, to separate disease site specific sessions, including head and neck (HN). This allowed for a more focused review of patient tumor volumes and treatment plans by a quorum of departmental disease site experts, at times resulting in subsequent plan adjustments. As one of our facilities houses a large HN practice, the local dosimetry planning load is directly affected by the number of replans that are needed post plan review. This study aimed to quantify edit requests for HN tumor volumes obtained through physician peer review and corresponding additional planning by dosimetry. Materials/Methods Data was collected on the number of tumor volume change requests originating from physician peer review feedback. Patient contours were reviewed at various stages in treatment planning – physician contours review only, plan complete, plan under treatment. Additionally, this study documented any additional time allocated for treatment planning amongst the dosimetry team. Results Within a two-month period, a total 214 patients were planned by the dosimetry team across all disease sites. Within this group, a total of 35 head and neck cases were presented at chart rounds. A total of 10 were reviewed at the contouring phase with one case requiring contour revisions (10%) resulting in no additional dosimetry work burden. The remaining 25 were not reviewed at the contouring phase, but within the planning stage. A total of 8 required changes to tumor volumes which resulted in “replan” generation by dosimetry, yielding a replan rate of 32% of cases at planning phase and total replan rate of 25.7% over all cases. With the overall time documented for replans averaging 5 hours, this added an additional 40-hour work burden on the dosimetry team. Conclusion Prospective contour review should be incorporated into our practice to aid in minimizing the need for HN replans. Future plans are to incorporate a hard stop MD task for tumor volume review within the patient care path to reduce the additional planning burden on the dosimetry team created by peer review change requests.

AbstractStereotactic radiotherapy (SRT) methods have become common for the treatment of small tumors in various parts of the body. Small field dosimetry has a unique set of challenges when it comes to the pre‐treatment validation of a radiotherapy plan that involves film dosimetry or high‐resolution detectors. Comparison of commercial quality assurance (QA) devices to the film dosimetry method for pre‐treatment evaluation of stereotactic radiosurgery (SRS), fractionated SRT, and stereotactic body radiation therapy treatment plans have been evaluated in this study. Forty stereotactic QA plans were measured using EBT‐XD film, IBA Matrixx Resolution, SNC ArcCHECK, Varian aS1200 EPID, SNC SRS MapCHECK, and IBA myQA SRS. The results of the commercial devices are compared to the EBT‐XD film dosimetry results for each gamma criteria. Treatment plan characteristics such as modulation factor and target volume were investigated for correlation with the passing rates. It was found that all detectors have greater than 95% passing rates at 3%/3 mm. Passing rates decrease rapidly for ArcCHECK and the Matrixx as criteria became more strict. In contrast, EBT‐XD film, SNC SRS MapCHECK, and IBA myQA SRS passing rates do not decline as rapidly when compared to Matrix Resolution, ArcCHECK, and the EPID. EBT‐XD film, SNC SRS MapCHECK, and IBA myQA SRS maintain greater than 90% passing rate at 2%/1 mm and greater than 80% at 1%/1 mm. Additionally, the ability of these devices to detect changes in dose distribution due to MLC positioning errors was investigated. Ten VMAT SBRT/SRS treatment plans were created with 6 MV FFF or 10 MV FFF beam energies using Eclipse 15.6. A MATLAB script was used to create two MLC positioning error scenarios from the original treatment plan. It was found that errors in MLC positioning were most reliably detected at 2%/1 mm for high‐resolution detectors and that lower‐resolution detectors did not consistently detect MLC positioning errors.

We have identified several features that correlate strongly with HN replans. A fine-tuned planning protocol is being developed to reduce the BHI for all HN plans. Our future work is to build a deep-learning model based on our current features and others such as patients' clinical tumor information, planning image, plan conformity, etc., to predict the probability of a replan before treatment.

Using a recombinant protein antigen for antibody testing shows a sum of antibody responses to multiple different immune epitopes existing in the protein antigen. In contrast, the antibody testing to an immunogenic peptide epitope reflects a singular antibody response to the individual peptide epitope. Therefore, using a panel of peptide epitopes provides an advantage for profiling multiple singular antibody responses with potential to estimate recent malaria exposure in human infections. However, transitioning from malaria immune epitope peptide-based ELISA to an all peptide bead-based multiplex Luminex assay presents some challenges including variation in the ability of different peptides to bind beads. The aim of this study was to develop a peptide coupling method while demonstrating the utility of these peptide epitopes from multiple stage antigens of Plasmodium falciparum for measuring antibodies. Successful coupling of peptide epitopes to beads followed three steps: 1) development of a peptide tag appended to the C-terminus of each peptide epitope consisting of beta-alanine-lysine (x 4)--cysteine, 2) bead modification with a high concentration of adipic acid dihydrazide, and 3) use of the peptide epitope as a blocker in place of the traditional choice, bovine serum albumin (BSA). This new method was used to couple 12 peptide epitopes from multiple stage specific antigens of P. falciparum, 1 Anopheles mosquito salivary gland peptide, and 1 Epstein-Barr virus peptide as an assay control. The new method was applied to testing of IgG in pooled samples from 30 individuals with previously repeated malaria exposure in western Kenya and IgM and IgG in samples from 37 U.S. travelers with recent exposure to malaria. The new peptide-bead coupling method and subsequent multiplex Luminex assay showed reliable detection of IgG to all 14 peptides in Kenyan samples. Among 37 samples from U.S. travelers recently diagnosed with malaria, IgM and IgG to the peptide epitopes were detected with high sensitivity and variation. Overall, the U.S. travelers had a much lower positivity rates of IgM than IgG to different peptide epitopes, ranging from a high of 62.2% positive for one epitope to a low of only 5.4% positive for another epitope. In contrast, the travelers had IgG positive rates from 97.3% to 91.9% to various peptide epitopes. Based on the different distribution in IgM and IgG positivity to overall number of peptide epitopes and to the number of pre-erythrocytic, erythrocytic, gametocytic, and salivary stage epitopes at the individual level, four distinct patterns of IgM and IgG responses among the 37 samples from US travelers were observed. Independent peptide-bead coupling and antibody level readout between two different instruments also showed comparable results. Overall, this new coupling method resolves the peptide-bead coupling challenge, is reproducible, and can be applied to any other immunogenic peptide epitopes. The resulting all peptide bead-based multiplex Luminex assay can be expanded to include other peptide epitopes of P. falciparum, different malaria species, or other diseases for surveillance, either in US travelers or endemic areas.

On-board cone-beam CT (CBCT) is commonly used during radiation therapy for patient positioning and other purposes. In particular, 4D CBCT is desirable for motion-resolved imaging of moving tumors e.g., for lung, pancreatic and liver cancer patients; limited-angle or sparse-view CBCT can monitor treatment delivery, e.g., via triggered imaging. Mathematically these can be categorized as sparse-data image reconstruction problem that is still unsolved. This work will develop a new reconstruction method titled AirNet for high-quality sparse-data CBCT image reconstruction. AirNet is designed to incorporate the benefits from analytical reconstruction method (AR), iterative reconstruction method (IR), and deep neural networks (DNN). AirNet built upon fused analytical and iterative reconstruction (AIR) that synergizes AR and IR via the optimization framework of modified proximal forward-backward splitting (PFBS). By unrolling PFBS into IR updates of CT data fidelity and DNN regularization with residual learning, AirNet utilizes AR such as FBP during the data fidelity, introduces dense connectivity into DNN regularization, and learns PFBS coefficients and DNN parameters that minimize the loss function during the training stage. With trained parameters AirNet can then be used for end-to-end image reconstruction. Additionally, two different strategies are developed for 4D-AirNet: prior-guided (PG) and all-phase (AP). PG-AirNet utilizes phase-by-phase training and reconstruction, while PG-AirNet uses a prior image reconstructed with all ten-phase projection data. CT libraries of prostate and lung scans was used to validate the AirNet in comparison with state-of-art DNN-based post-processing and image reconstruction methods. The validation loss in AirNet had the fastest decreasing rate, owing to inherited fast convergence from AIR. AirNet was robust to noise in projection data and content differences between the training set and the images to be reconstructed. In addition, AP-AirNet provided the best reconstruction quality overall for 4D CBCT. A new image reconstruction AirNet is developed for high-quality sparse-data CBCT image reconstruction. AirNet achieved the best image reconstruction quality among all methods under comparison for all sparse-data scenarios (sparse-view, limited-angle, and 4D CBCT).

OBJECTIVEThe optimal margin size in postoperative stereotactic radiosurgery (SRS) for brain metastases is unknown. Herein, the authors investigated the effect of SRS planning target volume (PTV) margin on local recurrence and symptomatic radiation necrosis postoperatively.METHODSRecords of patients who received postoperative LINAC-based SRS for brain metastases between 2006 and 2016 were reviewed and stratified based on PTV margin size (1.0 or > 1.0 mm). Patients were treated using frameless and framed SRS techniques, and both single-fraction and hypofractionated dosing were used based on lesion size. Kaplan-Meier and cumulative incidence models were used to estimate survival and intracranial outcomes, respectively. Multivariate analyses were also performed.RESULTSA total of 133 patients with 139 cavities were identified; 36 patients (27.1%) and 35 lesions (25.2%) were in the 1.0-mm group, and 97 patients (72.9%) and 104 lesions (74.8%) were in the > 1.0-mm group. Patient characteristics were balanced, except the 1.0-mm cohort had a better Eastern Cooperative Group Performance Status (grade 0: 36.1% vs 19.6%), higher mean number of brain metastases (1.75 vs 1.31), lower prescription isodose line (80% vs 95%), and lower median single fraction-equivalent dose (15.0 vs 17.5 Gy) (all p < 0.05). The median survival and follow-up for all patients were 15.6 months and 17.7 months, respectively. No significant difference in local recurrence was noted between the cohorts. An increased 1-year rate of symptomatic radionecrosis was seen in the larger margin group (20.9% vs 6.0%, p = 0.028). On multivariate analyses, margin size > 1.0 mm was associated with an increased risk for symptomatic radionecrosis (HR 3.07, 95% CI 1.13-8.34; p = 0.028), while multifraction SRS emerged as a protective factor for symptomatic radionecrosis (HR 0.13, 95% CI 0.02-0.76; p = 0.023).CONCLUSIONSExpanding the PTV margin beyond 1.0 mm is not associated with improved local recurrence but appears to increase the risk of symptomatic radionecrosis after postoperative SRS.

This study's purpose is to develop a learning-based approach to improve cone beam CT (CBCT) image quality for quantitative analysis during CBCT-guided adaptive prostate cancer radiotherapy (RT). We propose to integrate auto-context model and anatomical features into a machine learning framework to iteratively predict the corrected CBCT (CCBCT) with high image quality. After some preprocessing, we partition a given CBCT image into a set of patches. The most informative and salient anatomical features are extracted to train random forests. For each patch, we use the random forest to directly predict a CCBCT patch as an output. Moreover, we utilize an auto-context model to iteratively refine the prediction. Finally, we combine all of the predicted CCBCT patches to obtain the final CCBCT image. Our CCBCT and original CBCT (OCBCT) were registered to planning CT images for generating CCBCT-based and OCBCT-based treatment plans. Mean absolute error (MAE) and normalized cross-correlation (NCC) were used to quantify the differences between the CCBCT and planning CT images as well as the OCBCT and planning CT images. Clinically-relevant dose volume histogram (DVH) metrics were extracted from OCBCT-based, CCBCT-based and CT-based treatment plans for quantitative dosimetric evaluation. Gamma analysis was performed for the comparison of absorbed dose distributions among these three plans of each patient. This learning-based correction algorithm was tested using 10 prostate cancer patients with 16 treatment plans (mean prescription dose: 44.99±14.07 Gy), each of whom has planning CT and CBCT. The mean MAE and NCC between OCBCT/CCBCT and planning CT were 45.47±12.26/5.44±12.27 HU and 0.94±0.01/0.95±0.01 for all patients' data. The mean dose differences between the PTVs on OCBCT/CCBCT-based and CT-based plans were -0.16±1.19%/-0.11±0.68%, 0.81±0.71%/0.09±0.41%, 0.67±0.64%/0.04±0.38%, 0.55±0.57%/0.04±0.34%, 0.68±0.64%/0.05±0.38% and 1.33±1.46%/0.46±1.03% for Dmin, D10, D50, D95, Dmean and Dmax, respectively. The mean dose differences between bladder, rectum and femur head on OCBCT/CCBCT-based and CT-based plans ranged from -0.32%/-0.25% to 1.37%/0.84% for all DVH metrics. There were significant improvements after our CBCT correction in the D10, D50, D95, Dmean and Dmax of the PTV, bladder and femur head, and in the Dmean of rectum (p < 0.05). The average pass rate of gamma analysis after our correction was over 99% with 3%/3 mm (dose difference/distance to agreement) acceptance criteria in all plans. We have developed a novel learning-based method to improve CBCT imaging for quantitative analysis during prostate cancer RT. We have demonstrated that the reported method is capable of reliably improving CBCT images quality and providing comparable dose accuracy to the standard CT for prostate cancer treatment planning. The proposed learning-based CBCT correction method has great potential in CBCT-guided adaptive prostate cancer RT.