Comprehensive multimodal deep learning survival prediction enabled by a transformer architecture: A multicenter study in glioblastoma.

Background: This research aims to improve glioblastoma survival prediction by integrating MR images, clinical, and molecular-pathologic data in a transformer-based deep learning model, addressing data heterogeneity and performance generalizability. Methods: We propose and evaluate a transformer-based nonlinear and nonproportional survival prediction model. The model employs self-supervised learning techniques to effectively encode the high-dimensional MRI input for integration with nonimaging data using cross-attention. To demonstrate model generalizability, the model is assessed with the time-dependent concordance index (Cdt) in 2 training setups using 3 independent public test sets: UPenn-GBM, UCSF-PDGM, and Rio Hortega University Hospital (RHUH)-GBM, each comprising 378, 366, and 36 cases, respectively. Results: The proposed transformer model achieved a promising performance for imaging as well as nonimaging data, effectively integrating both modalities for enhanced performance (UCSF-PDGM test-set, imaging Cdt 0.578, multimodal Cdt 0.672) while outperforming state-of-the-art late-fusion 3D-CNN-based models. Consistent performance was observed across the 3 independent multicenter test sets with Cdt values of 0.707 (UPenn-GBM, internal test set), 0.672 (UCSF-PDGM, first external test set), and 0.618 (RHUH-GBM, second external test set). The model achieved significant discrimination between patients with favorable and unfavorable survival for all 3 datasets (log-rank P 1.9 × 10-8, 9.7 × 10-3, and 1.2 × 10-2). Comparable results were obtained in the second setup using UCSF-PDGM for training/internal testing and UPenn-GBM and RHUH-GBM for external testing (Cdt 0.670, 0.638, and 0.621). Conclusions: The proposed transformer-based survival prediction model integrates complementary information from diverse input modalities, contributing to improved glioblastoma survival prediction compared to state-of-the-art methods. Consistent performance was observed across institutions supporting model generalizability.

Improving the safety of radiotherapy treatment processes via incident-driven FMEA feedback loops.

BACKGROUND: Failure mode and effects analysis (FMEA) is a valuable tool for radiotherapy risk assessment, yet its outputs might be unreliable due to failures not being identified or due to a lack of accurate error rates. PURPOSE: A novel incident reporting system (IRS) linked to an FMEA database was tested and evaluated. The study investigated whether the system was suitable for validating a previously performed analysis and whether it could provide accurate error rates to support the expert occurrence ratings of previously identified failure modes. METHODS: Twenty-three pre-identified failure modes of our external beam radiotherapy process, covering the process steps from patient admission to treatment delivery, were proffered on dedicated FMEA feedback and incident reporting terminals generated by the IRS. The clinical setting involved a computed tomography scanner, dosimetry, and five linacs. Incoming reports were used as basis to identify additional failure modes or confirm initial ones. The Kruskal-Wallis H test was applied to compare the risk priorities of the retrospective and prospective failure modes. Wald's sequential probability ratio test was used to investigate the correctness of the experts' occurrence ratings by means of the number of incoming reports. RESULTS: Over a 15-month period, 304 reports were submitted. There were 0.005 (confidence interval [CI], 0.0014-0.0082) reported incidents per imaging study and 0.0006 (CI, 0.0003-0.0009) reported incidents per treatment fraction. Sixteen additional failure modes could be identified, and their risk priorities did not differ from those of the initial failure modes (p = 0.954). One failure mode occurrence rating could be increased, whereas the other 22 occurrence ratings could not be disproved. CONCLUSIONS: Our approach is suitable for validating FMEAs and deducing additional failure modes on a continual basis. Accurate error rates can only be provided if a sufficient number of reports is available.

Principles of artificial intelligence in radiooncology.

PURPOSE: In the rapidly expanding field of artificial intelligence (AI) there is a wealth of literature detailing the myriad applications of AI, particularly in the realm of deep learning. However, a review that elucidates the technical principles of deep learning as relevant to radiation oncology in an easily understandable manner is still notably lacking. This paper aims to fill this gap by providing a comprehensive guide to the principles of deep learning that is specifically tailored toward radiation oncology. METHODS: In light of the extensive variety of AI methodologies, this review selectively concentrates on the specific domain of deep learning. It emphasizes the principal categories of deep learning models and delineates the methodologies for training these models effectively. RESULTS: This review initially delineates the distinctions between AI and deep learning as well as between supervised and unsupervised learning. Subsequently, it elucidates the fundamental principles of major deep learning models, encompassing multilayer perceptrons (MLPs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), transformers, generative adversarial networks (GANs), diffusion-based generative models, and reinforcement learning. For each category, it presents representative networks alongside their specific applications in radiation oncology. Moreover, the review outlines critical factors essential for training deep learning models, such as data preprocessing, loss functions, optimizers, and other pivotal training parameters including learning rate and batch size. CONCLUSION: This review provides a comprehensive overview of deep learning principles tailored toward radiation oncology. It aims to enhance the understanding of AI-based research and software applications, thereby bridging the gap between complex technological concepts and clinical practice in radiation oncology.

Initial needle tracking with the first standalone combined infrared camera - CT system for brachytherapy-analysis of tracking accuracy and uncertainties.

PURPOSE: A prototype infrared camera - cone-beam computed tomography (CBCT) system for tracking in brachytherapy has recently been developed. We evaluated for the first time the corresponding tracking accuracy and uncertainties, and implemented a tracking-based prediction of needles on CBCT scans. METHODS: A marker tool rigidly attached to needles was 3D printed. The precision and accuracy of tool tracking was then evaluated for both static and dynamic scenarios. Euclidean distances between the tracked and CBCT-derived markers were assessed as well. To implement needle tracking, ground truth models of the tool attached to 200 mm and 160 mm needles were matched to the tracked positions in order to project the needles into CBCT scans. Deviations between projected and actual needle tips were measured. Finally, we put our results into perspective with simulations of the system's tracking uncertainties. RESULTS: For the stationary scenario and dynamic movements, we achieved tool-tracking precision and accuracy of 0.04 ± 0.06 mm and 0.16 ± 0.18 mm, respectively. The tracked marker positions differed by 0.52 ± 0.18 mm from the positions determined via CBCT. In addition, the predicted needle tips in air deviated from the actual tip positions by only 1.62 ± 0.68 mm (200 mm needle) and 1.49 ± 0.62 mm (160 mm needle). The simulated tracking uncertainties resulted in tip variations of 1.58 ± 0.91 mm and 1.31 ± 0.69 mm for the 200 mm and 160 mm needles, respectively. CONCLUSION: With the innovative system it was possible to achieve a high tracking and prediction accuracy of marker tool and needles. The system shows high potential for applicator tracking in brachytherapy.

Multicenter privacy-preserving model training for deep learning brain metastases autosegmentation.

OBJECTIVES: This work aims to explore the impact of multicenter data heterogeneity on deep learning brain metastases (BM) autosegmentation performance, and assess the efficacy of an incremental transfer learning technique, namely learning without forgetting (LWF), to improve model generalizability without sharing raw data. MATERIALS AND METHODS: A total of six BM datasets from University Hospital Erlangen (UKER), University Hospital Zurich (USZ), Stanford, UCSF, New York University (NYU), and BraTS Challenge 2023 were used. First, the performance of the DeepMedic network for BM autosegmentation was established for exclusive single-center training and mixed multicenter training, respectively. Subsequently privacy-preserving bilateral collaboration was evaluated, where a pretrained model is shared to another center for further training using transfer learning (TL) either with or without LWF. RESULTS: For single-center training, average F1 scores of BM detection range from 0.625 (NYU) to 0.876 (UKER) on respective single-center test data. Mixed multicenter training notably improves F1 scores at Stanford and NYU, with negligible improvement at other centers. When the UKER pretrained model is applied to USZ, LWF achieves a higher average F1 score (0.839) than naive TL (0.570) and single-center training (0.688) on combined UKER and USZ test data. Naive TL improves sensitivity and contouring accuracy, but compromises precision. Conversely, LWF demonstrates commendable sensitivity, precision and contouring accuracy. When applied to Stanford, similar performance was observed. CONCLUSION: Data heterogeneity (e.g., variations in metastases density, spatial distribution, and image spatial resolution across centers) results in varying performance in BM autosegmentation, posing challenges to model generalizability. LWF is a promising approach to peer-to-peer privacy-preserving model training.

Inter-observer effects in needle reconstruction for temporary prostate brachytherapy: Dosimetric implications and adaptive CBCT-TRUS registration solutions.

PURPOSE: To investigate geometric and dosimetric inter-observer variability in needle reconstruction for temporary prostate brachytherapy. To assess the potential of registrations between transrectal ultrasound (TRUS) and cone-beam computed tomography (CBCT) to support implant reconstructions. METHODS AND MATERIALS: The needles implanted in 28 patients were reconstructed on TRUS by three physicists. Corresponding geometric deviations and associated dosimetric variations to prostate and organs at risk (urethra, bladder, rectum) were analyzed. To account for the found inter-observer variability, various approaches (template-based, probe-based, marker-based) for registrations of CBCT to TRUS were investigated regarding the respective needle transfer accuracy in a phantom study. Three patient cases were examined to assess registration accuracy in-vivo. RESULTS: Geometric inter-observer deviations >1 mm and >3 mm were found for 34.9% and 3.5% of all needles, respectively. Prostate dose coverage (changes up to 7.2%) and urethra dose (partly exceeding given dose constraints) were most affected by associated dosimetric changes. Marker-based and probe-based registrations resulted in the phantom study in high mean needle transfer accuracies of 0.73 mm and 0.12 mm, respectively. In the patient cases, the marker-based approach was the superior technique for CBCT-TRUS fusions. CONCLUSION: Inter-observer variability in needle reconstruction can substantially affect dosimetry for individual patients. Especially marker-based CBCT-TRUS registrations can help to ensure accurate reconstructions for improved treatment planning.

Optimized raw data selection for artifact reduction of breathing controlled four-dimensional sequence scanning.

Background and purpose: Even with most breathing-controlled four-dimensional computed tomography (4DCT) algorithms image artifacts caused by single significant longer breathing still occur, resulting in negative consequences for radiotherapy. Our study presents first phantom examinations of a new optimized raw data selection and binning algorithm, aiming to improve image quality and geometric accuracy without additional dose exposure. Materials and methods: To validate the new approach, phantom measurements were performed to assess geometric accuracy (volume fidelity, root mean square error, Dice coefficient of volume overlap) for one- and three-dimensional tumor motion trajectories with and without considering motion hysteresis effects. Scans without significantly longer breathing cycles served as references. Results: Median volume deviations between optimized approach and reference of at maximum 1% were obtained considering all movements. In comparison, standard reconstruction yielded median deviations of 9%, 21% and 12% for one-dimensional, three-dimensional, and hysteresis motion, respectively. Measurements in one- and three-dimensional directions reached a median Dice coefficient of 0.970 ± 0.013 and 0.975 ± 0.012, respectively, but only 0.918 ± 0.075 for hysteresis motions averaged over all measurements for the optimized selection. However, for the standard reconstruction median Dice coefficients were 0.845 ± 0.200, 0.868 ± 0.205 and 0.915 ± 0.075 for one- and three-dimensional as well as hysteresis motions, respectively. Median root mean square errors for the optimized algorithm were 30 ± 16 HU2 and 120 ± 90 HU2 for three-dimensional and hysteresis motions, compared to 212 ± 145 HU2 and 130 ± 131 HU2 for the standard reconstruction. Conclusions: The algorithm was proven to reduce 4DCT-related artifacts due to missing projection data without further dose exposure. An improvement in radiotherapy treatment planning due to better image quality can be expected.

First implementation of an innovative infra-red camera system integrated into a mobile CBCT scanner for applicator tracking in brachytherapy-Initial performance characterization.

PURPOSE: To enable a real-time applicator guidance for brachytherapy, we used for the first time infra-red tracking cameras (OptiTrack, USA) integrated into a mobile cone-beam computed tomography (CBCT) scanner (medPhoton, Austria). We provide the first description of this prototype and its performance evaluation. METHODS: We performed assessments of camera calibration and camera-CBCT registration using a geometric calibration phantom. For this purpose, we first evaluated the effects of intrinsic parameters such as camera temperature or gantry rotations on the tracked marker positions. Afterward, calibrations with various settings (sample number, field of view coverage, calibration directions, calibration distances, and lighting conditions) were performed to identify the requirements for achieving maximum tracking accuracy based on an in-house phantom. The corresponding effects on camera-CBCT registration were determined as well by comparing tracked marker positions to the positions determined via CBCT. Long-term stability was assessed by comparing tracking and a ground-truth on a weekly basis for 6 weeks. RESULTS: Robust tracking with positional drifts of 0.02 ± 0.01 mm was feasible using the system after a warm-up period of 90 min. However, gantry rotations affected the tracking and led to inaccuracies of up to 0.70 mm. We identified that 4000 samples and full coverage were required to ensure a robust determination of marker positions and camera-CBCT registration with geometric deviations of 0.18 ± 0.03 mm and 0.42 ± 0.07 mm, respectively. Long-term stability showed deviations of more than two standard deviations from the initial calibration after 3 weeks. CONCLUSION: We implemented for the first time a standalone combined camera-CBCT system for tracking in brachytherapy. The system showed high potential for establishing corresponding workflows.

A comprehensive quality assurance protocol for electromagnetic tracking in brachytherapy.

BACKGROUND: Electromagnetic tracking (EMT) systems have proven to be a valuable source of information regarding the location and geometry of applicators in patients undergoing brachytherapy (BT). As an important element of an enhanced and individualized pre-treatment verification, EMT can play a pivotal role in detecting treatment errors and uncertainties to increase patient safety. PURPOSE: The purpose of this study is two-fold: to design, develop and test a dedicated measurement protocol for the use of EMT-enabled afterloaders in BT and to collect and compare the data acquired from three different radiation oncology centers in different clinical environments. METHODS: A novel quality assurance (QA) phantom composed of a scaffold with supports to fix the field generator, different BT applicators, and reference sensors (sensor verification tools) was used to assess the precision (jitter error) and accuracy (relative distance errors and target registration error) of the EMT sensor integrated into an afterloader prototype. Measurements were repeated in different environments where EMT measurements are likely to be performed, namely an electromagnetically clean laboratory, a BT suite, an operating room, and, if available, a CT suite and an MRI suite dedicated to BT. RESULTS: The mean positional jitter was consistently under 0.1 mm across all measurement points, with a slight trend of increased jitter at greater distances from the field generator. The mean variability of sensor positioning in the tested tandem and ring gynecological applicator was also below 0.1 mm. The tracking accuracy close to the center of the measurement volume was higher than at its edges. The relative distance error at the center was 0.2-0.3 mm with maximum values reaching 1.2-1.8 mm, but up to 5.5 mm for measurement points close to the edges. In general, similar accuracy results were obtained in the clinical environments and in all investigated institutions (median distance error 0.1-0.4 mm, maximum error 1.0-2.0 mm), however, errors were found to be larger in the CT suite (median distance error up to 1.0 mm, maximum error up to 3.6 mm). CONCLUSION: The presented quality assessment protocol for EMT systems in BT has demonstrated that EMT offers a high-accuracy determination of the applicator/implant geometry even in clinical environments. In addition to that, it has provided valuable insights into the performance of EMT-enabled afterloaders across different radiation oncology centers.

"sCT-Feasibility" - a feasibility study for deep learning-based MRI-only brain radiotherapy.

BACKGROUND: Radiotherapy (RT) is an important treatment modality for patients with brain malignancies. Traditionally, computed tomography (CT) images are used for RT treatment planning whereas magnetic resonance imaging (MRI) images are used for tumor delineation. Therefore, MRI and CT need to be registered, which is an error prone process. The purpose of this clinical study is to investigate the clinical feasibility of a deep learning-based MRI-only workflow for brain radiotherapy, that eliminates the registration uncertainty through calculation of a synthetic CT (sCT) from MRI data. METHODS: A total of 54 patients with an indication for radiation treatment of the brain and stereotactic mask immobilization will be recruited. All study patients will receive standard therapy and imaging including both CT and MRI. All patients will receive dedicated RT-MRI scans in treatment position. An sCT will be reconstructed from an acquired MRI DIXON-sequence using a commercially available deep learning solution on which subsequent radiotherapy planning will be performed. Through multiple quality assurance (QA) measures and reviews during the course of the study, the feasibility of an MRI-only workflow and comparative parameters between sCT and standard CT workflow will be investigated holistically. These QA measures include feasibility and quality of image guidance (IGRT) at the linear accelerator using sCT derived digitally reconstructed radiographs in addition to potential dosimetric deviations between the CT and sCT plan. The aim of this clinical study is to establish a brain MRI-only workflow as well as to identify risks and QA mechanisms to ensure a safe integration of deep learning-based sCT into radiotherapy planning and delivery. DISCUSSION: Compared to CT, MRI offers a superior soft tissue contrast without additional radiation dose to the patients. However, up to now, even though the dosimetrical equivalence of CT and sCT has been shown in several retrospective studies, MRI-only workflows have still not been widely adopted. The present study aims to determine feasibility and safety of deep learning-based MRI-only radiotherapy in a holistic manner incorporating the whole radiotherapy workflow. TRIAL REGISTRATION: NCT06106997.

Process failure mode and effects analysis for external beam radiotherapy: Introducing a literature-based template and a novel action priority.

PURPOSE: The first aim of the study was to create a general template for analyzing potential failures in external beam radiotherapy, EBRT, using the process failure mode and effects analysis (PFMEA). The second aim was to modify the action priority (AP), a novel prioritization method originally introduced by the Automotive Industry Action Group (AIAG), to work with different severity, occurrence, and detection rating systems used in radiation oncology. METHODS AND MATERIALS: The AIAG PFMEA approach was employed in combination with an extensive literature survey to develop the EBRT-PFMEA template. Subsets of high-risk failure modes found through the literature survey were added to the template where applicable. Our modified AP for radiation oncology (RO AP) was defined using a weighted sum of severity, occurrence, and detectability. Then, Monte Carlo simulations were conducted to compare the original AIAG AP, the RO AP, and the risk priority number (RPN). The results of the simulations were used to determine the number of additional corrective actions per failure mode and to parametrize the RO AP to our department's rating system. RESULTS: An EBRT-PFMEA template comprising 75 high-risk failure modes could be compiled. The AIAG AP required 1.7 additional corrective actions per failure mode, while the RO AP ranged from 1.3 to 3.5, and the RPN required 3.6. The RO AP could be parametrized so that it suited our rating system and evaluated severity, occurrence, and detection ratings equally to the AIAG AP. CONCLUSIONS: An adjustable EBRT-PFMEA template is provided which can be used as a practical starting point for creating institution-specific templates. Moreover, the RO AP introduces transparent action levels that can be adapted to any rating system.

Implementation and validation of 2-D-array based tests in routine linac quality assurance.

PURPOSE: In medical linac quality assurance (QA), to replace film dosimetry with low-resolution 2-D ionization chamber array measurements, to validate the procedures, and to perform a comprehensive sensitivity analysis. METHODS: A 2-D ionization chamber array with a spatial resolution of 7.62 mm was deployed to perform the following tests: Junction tests, MLC transmission test, beam profile constancy vs. gantry angle test, beam profile constancy vs. low dose delivery test, and beam energy constancy vs. low dose delivery test. Test validation and sensitivity analyses based on short- and long-term statistics of the test results were performed. RESULTS: All selected mechanical and dosimetry tests could be successfully performed with a 2-D array. Considering the tolerance limits recommended by the AAPM Task Group 142 report (2009), sensitivities of 99.0% or better and specificities ranging from 99.5% to 99.9% could be achieved in all tests when the proper metrics were chosen. CONCLUSIONS: The results showed that a low-resolution 2-D ionization chamber array could replace film dosimetry without having to sacrifice high test sensitivity. Its implementation in the routine clinical linac QA program may involve considerable QA time savings.

Quality requirements for MRI simulation in cranial stereotactic radiotherapy: a guideline from the German Taskforce "Imaging in Stereotactic Radiotherapy".

Accurate Magnetic Resonance Imaging (MRI) simulation is fundamental for high-precision stereotactic radiosurgery and fractionated stereotactic radiotherapy, collectively referred to as stereotactic radiotherapy (SRT), to deliver doses of high biological effectiveness to well-defined cranial targets. Multiple MRI hardware related factors as well as scanner configuration and sequence protocol parameters can affect the imaging accuracy and need to be optimized for the special purpose of radiotherapy treatment planning. MRI simulation for SRT is possible for different organizational environments including patient referral for imaging as well as dedicated MRI simulation in the radiotherapy department but require radiotherapy-optimized MRI protocols and defined quality standards to ensure geometrically accurate images that form an impeccable foundation for treatment planning. For this guideline, an interdisciplinary panel including experts from the working group for radiosurgery and stereotactic radiotherapy of the German Society for Radiation Oncology (DEGRO), the working group for physics and technology in stereotactic radiotherapy of the German Society for Medical Physics (DGMP), the German Society of Neurosurgery (DGNC), the German Society of Neuroradiology (DGNR) and the German Chapter of the International Society for Magnetic Resonance in Medicine (DS-ISMRM) have defined minimum MRI quality requirements as well as advanced MRI simulation options for cranial SRT.

Investigating the impact of breast positioning control on physical treatment parameters in multi-catheter breast brachytherapy.

PURPOSE: To assess the effects of a workflow for reproducible patient and breast positioning on implant stability during high-dose-rate multi-catheter breast brachytherapy. METHODS: Thirty patients were treated with our new positioning control workflow. Implant stability was evaluated based on a comparison of planning-CTs to control-CTs acquired halfway through the treatment. To assess geometric stability, button-button distance variations as well as Euclidean dwell position deviations were evaluated. The latter were also quantified within various separated regions within the breast to investigate the location-dependency of implant alterations. Furthermore, dosimetric variations to target volume and organs at risk (ribs, skin) as well as isodose volume changes were analyzed. Results were compared to a previously treated cohort of 100 patients. RESULTS: With the introduced workflow, the patient fraction affected by button-button distance variations > 5 mm and by dwell position deviations > 7 mm were reduced from 37% to 10% and from 30% to 6.6%, respectively. Implant stability improved the most in the lateral to medial breast regions. Only small stability enhancements were observed regarding target volume dosimetry, but the stability of organ at risk exposure became substantially higher. D0.2ccm skin dose variations > 12.4% and D0.1ccm rib dose variations > 6.7% were reduced from 11% to 0% and from 16% to 3.3% of all patients, respectively. CONCLUSION: Breast positioning control improved geometric and dosimetric implant stability for individual patients, and thus enhanced physical plan validity in these cases.

Toward a deep learning-based magnetic resonance imaging only workflow for postimplant dosimetry in I-125 seed brachytherapy for prostate cancer.

BACKGROUND AND PURPOSE: The current standard imaging-technique for creating postplans in seed prostate brachytherapy is computed tomography (CT), that is associated with additional radiation exposure and poor soft tissue contrast. To establish a magnetic resonance imaging (MRI) only workflow combining improved tissue contrast and high seed detectability, a deep learning-approach for automatic seed segmentation on MRI-scans was developed. MATERIAL AND METHODS: Patients treated with I-125 seed brachytherapy received a postplan-CT and a 1.5 T MRI-scan on nominal day 30 after implantation. For MRI-based seed visualization, DIXON-sequences were acquired and deep learning-based quantitative susceptibility maps (QSM) were generated from 3D-gradient-echo-sequences from 20 patients. Seed segmentations created on CT served as ground truth. For automatic seed segmentation on MRI, a 3D nnU-net model was trained using QSM and DIXON, both solely and combined. RESULTS: Of the implanted seeds 94.8 ± 2.4% were detected with deep learning automatic segmentation entrained on both QSM and DIXON data. Models trained on the individual sequence data-sets performed worse with detection rates of 87.5 ± 2.6% or 88.6 ± 7.5% for QSM and DIXON respectively. The seed centers identified on CT versus QSM and DIXON were on average 1.8 ± 1.3 mm apart. Postimplant dosimetry for evaluation of positioning inaccuracies revealed only small variations of up to 0.4 ± 4.26 Gy in D90 (dose 90% of the prostate receives) between the standard CT-approach and our MRI-only workflow. CONCLUSION: The proposed deep learning-based MRI-only workflow provided a promisingly accurate and robust seed localization and thus has the potential to compete with current state-of-the-art CT-based postimplant dosimetry in the future.

Benchmarking ChatGPT-4 on a radiation oncology in-training exam and Red Journal Gray Zone cases: potentials and challenges for ai-assisted medical education and decision making in radiation oncology.

Purpose: The potential of large language models in medicine for education and decision-making purposes has been demonstrated as they have achieved decent scores on medical exams such as the United States Medical Licensing Exam (USMLE) and the MedQA exam. This work aims to evaluate the performance of ChatGPT-4 in the specialized field of radiation oncology. Methods: The 38th American College of Radiology (ACR) radiation oncology in-training (TXIT) exam and the 2022 Red Journal Gray Zone cases are used to benchmark the performance of ChatGPT-4. The TXIT exam contains 300 questions covering various topics of radiation oncology. The 2022 Gray Zone collection contains 15 complex clinical cases. Results: For the TXIT exam, ChatGPT-3.5 and ChatGPT-4 have achieved the scores of 62.05% and 78.77%, respectively, highlighting the advantage of the latest ChatGPT-4 model. Based on the TXIT exam, ChatGPT-4's strong and weak areas in radiation oncology are identified to some extent. Specifically, ChatGPT-4 demonstrates better knowledge of statistics, CNS & eye, pediatrics, biology, and physics than knowledge of bone & soft tissue and gynecology, as per the ACR knowledge domain. Regarding clinical care paths, ChatGPT-4 performs better in diagnosis, prognosis, and toxicity than brachytherapy and dosimetry. It lacks proficiency in in-depth details of clinical trials. For the Gray Zone cases, ChatGPT-4 is able to suggest a personalized treatment approach to each case with high correctness and comprehensiveness. Importantly, it provides novel treatment aspects for many cases, which are not suggested by any human experts. Conclusion: Both evaluations demonstrate the potential of ChatGPT-4 in medical education for the general public and cancer patients, as well as the potential to aid clinical decision-making, while acknowledging its limitations in certain domains. Owing to the risk of hallucinations, it is essential to verify the content generated by models such as ChatGPT for accuracy.

Deep Learning and Registration-Based Mapping for Analyzing the Distribution of Nodal Metastases in Head and Neck Cancer Cohorts: Informing Optimal Radiotherapy Target Volume Design.

We introduce a deep-learning- and a registration-based method for automatically analyzing the spatial distribution of nodal metastases (LNs) in head and neck (H/N) cancer cohorts to inform radiotherapy (RT) target volume design. The two methods are evaluated in a cohort of 193 H/N patients/planning CTs with a total of 449 LNs. In the deep learning method, a previously developed nnU-Net 3D/2D ensemble model is used to autosegment 20 H/N levels, with each LN subsequently being algorithmically assigned to the closest-level autosegmentation. In the nonrigid-registration-based mapping method, LNs are mapped into a calculated template CT representing the cohort-average patient anatomy, and kernel density estimation is employed to estimate the underlying average 3D-LN probability distribution allowing for analysis and visualization without prespecified level definitions. Multireader assessment by three radio-oncologists with majority voting was used to evaluate the deep learning method and obtain the ground-truth distribution. For the mapping technique, the proportion of LNs predicted by the 3D probability distribution for each level was calculated and compared to the deep learning and ground-truth distributions. As determined by a multireader review with majority voting, the deep learning method correctly categorized all 449 LNs to their respective levels. Level 2 showed the highest LN involvement (59.0%). The level involvement predicted by the mapping technique was consistent with the ground-truth distribution (p for difference 0.915). Application of the proposed methods to multicenter cohorts with selected H/N tumor subtypes for informing optimal RT target volume design is promising.

Plan robustness analysis for threshold determination of SGRT-based intrafraction motion control in 3DCRT breast cancer radiation therapy.

PURPOSE: The goal of this study was to obtain maximum allowed shift deviations from planning position in six degrees of freedom (DOF), that can serve as threshold values in surface guided radiation therapy (SGRT) of breast cancer patients. METHODS: The robustness of conformal treatment plans of 50 breast cancer patients against 6DOF shifts was investigated. For that, new dose distributions were calculated on shifted computed tomography scans and evaluated with respect to target volume and spinal cord dose. Maximum allowed shift values were identified by imposing dose constraints on the target volume dose coverage for 1DOF, and consecutively, for 6DOF shifts using an iterative approach and random sampling. RESULTS: Substantial decreases in target dose coverage and increases of spinal cord dose were observed. Treatment plans showed highly differing robustness for different DOFs or treated area. The sensitivity was particularly high if clavicular lymph nodes were irradiated, for shifts in lateral, vertical, roll or yaw direction, and showed partly pronounced asymmetries. Threshold values showed similar properties with an absolute value range of 0.8 mm to 5 mm and 1.4° to 5°. CONCLUSION: The robustness analysis emphasized the necessity of taking differences between DOFs and asymmetrical sensitivities into account when evaluating the dosimetric impact of position deviations. It also highlighted the importance of rotational shifts, especially if clavicular lymph nodes were irradiated. A practical approach of determining 6DOF shift limits was introduced and a set of threshold values applicable for SGRT based patient motion control was identified.

Development and clinical implementation of a digital system for risk assessments for radiation therapy.

Before introducing new treatment techniques, an investigation of hazards due to unintentional radiation exposures is a reasonable activity for proactively increasing patient safety. As dedicated software is scarce, we developed a tool for risk assessment to design a quality management program based on best practice methods, i.e., process mapping, failure modes and effects analysis and fault tree analysis. Implemented as a web database application, a single dataset was used to describe the treatment process and its failure modes. The design of the system and dataset allowed failure modes to be represented both visually as fault trees and in a tabular form. Following the commissioning of the software for our department, previously conducted risk assessments were migrated to the new system after being fully re-assessed which revealed a shift in risk priorities. Furthermore, a weighting factor was investigated to bring risk levels of the migrated assessments into perspective. The compensation did not affect high priorities but did re-prioritize in the midrange of the ranking. We conclude that the tool is suitable to conduct multiple risk assessments and concomitantly keep track of the overall quality management activities.