Synthetic Low-Energy Monochromatic Image Generation in Single-Energy Computed Tomography System Using a Transformer-Based Deep Learning Model.

While dual-energy computed tomography (DECT) technology introduces energy-specific information in clinical practice, single-energy CT (SECT) is predominantly used, limiting the number of people who can benefit from DECT. This study proposed a novel method to generate synthetic low-energy virtual monochromatic images at 50 keV (sVMI50keV) from SECT images using a transformer-based deep learning model, SwinUNETR. Data were obtained from 85 patients who underwent head and neck radiotherapy. Among these, the model was built using data from 70 patients for whom only DECT images were available. The remaining 15 patients, for whom both DECT and SECT images were available, were used to predict from the actual SECT images. We used the SwinUNETR model to generate sVMI50keV. The image quality was evaluated, and the results were compared with those of the convolutional neural network-based model, Unet. The mean absolute errors from the true VMI50keV were 36.5 ± 4.9 and 33.0 ± 4.4 Hounsfield units for Unet and SwinUNETR, respectively. SwinUNETR yielded smaller errors in tissue attenuation values compared with those of Unet. The contrast changes in sVMI50keV generated by SwinUNETR from SECT were closer to those of DECT-derived VMI50keV than the contrast changes in Unet-generated sVMI50keV. This study demonstrated the potential of transformer-based models for generating synthetic low-energy VMIs from SECT images, thereby improving the image quality of head and neck cancer imaging. It provides a practical and feasible solution to obtain low-energy VMIs from SECT data that can benefit a large number of facilities and patients without access to DECT technology.

Single-Fraction Palliative High-Dose-Rate Brachytherapy for Symptom Management in a 97-Year-Old Patient With Dementia.

Background: Delivering cancer treatment to elderly patients with dementia is often challenging. We describe performing palliative surface mold brachytherapy (SMBT) in an elderly patient with advanced dementia for pain control using music therapy to assist with agitation. Case Description: The patient was a 97-year-old Japanese woman with advanced dementia. Exudate was observed from her tumor, and she complained of Grade 2 severity pain using Support team assessment schedule (STAS), especially when undergoing would dressings. Given her advanced dementia, she was not considered a candidate for radical surgery or external beam radiotherapy. We instead treated her with high-dose-rate (HDR) SMBT. Due to her advanced dementia associated with agitation, she could not maintain her position. She was able to remain calm while listening to traditional Japanese enka music, which enables our team to complete her radiation without using anesthetics or sedating analgesics. Her localized pain severity decreased ≤21 days and the exudate fluid disappeared ≤63 days after HDR-SMBT. Her tumor was locally controlled until her death from intercurrent disease 1 year after HDR-SMBT. Discussion: Single fraction palliative HDR-SMBT was useful for successful treatment of skin cancer in an elderly patient. Traditional Japanese music helped reduce her agitation to complete HDR-SMBT. For elderly patients with agitation associated with dementia, we should consider using music and music therapy to facilitate radiation therapy.

Extracting the gradient component of the gamma index using the Lie derivative method.

Objective. The gamma index (γ) has been extensively investigated in the medical physics and applied in clinical practice. However,γhas a significant limitation when used to evaluate the dose-gradient region, leading to inconveniences, particularly in stereotactic radiotherapy (SRT). This study proposes a novel evaluation method combined withγto extract clinically problematic dose-gradient regions caused by irradiation including certain errors.Approach. A flow-vector field in the dose distribution is obtained when the dose is considered a scalar potential. Using the Lie derivative from differential geometry, we definedL,S, andUto evaluate the intensity, vorticity, and flow amount of deviation between two dose distributions, respectively. These metrics multiplied byγ(γL,γS,γU), along with the threshold valueσ, were verified in the ideal SRT case and in a clinical case of irradiation near the brainstem region using radiochromic films. Moreover, Moran's gradient index (MGI), Bakai's χ factor, and the structural similarity index (SSIM) were investigated for comparisons.Main results. A highL-metric value mainly extracted high-dose-gradient induced deviations, which was supported by highSandUmetrics observed as a robust deviation and an influence of the dose-gradient, respectively. TheS-metric also denotes the measured similarity between the compared dose distributions. In theγdistribution,γLsensitively detected the dose-gradient region in the film measurement, despite the presence of noise. The thresholdσsuccessfully extracted the gradient-error region whereγ> 1 analysis underestimated, andσ= 0.1 (plan) andσ= 0.001 (film measurement) were obtained according to the compared resolutions. However, the MGI, χ, and SSIM failed to detect the clinically interested region.Significance. Although further studies are required to clarify the error details, this study demonstrated that the Lie derivative method provided a novel perspective for the identifying gradient-induced error regions and enabled enhanced and clinically significant evaluations ofγ.

Artificial intelligence-aided lytic spinal bone metastasis classification on CT scans.

PURPOSE: Spinal bone metastases directly affect quality of life, and patients with lytic-dominant lesions are at high risk for neurological symptoms and fractures. To detect and classify lytic spinal bone metastasis using routine computed tomography (CT) scans, we developed a deep learning (DL)-based computer-aided detection (CAD) system. METHODS: We retrospectively analyzed 2125 diagnostic and radiotherapeutic CT images of 79 patients. Images annotated as tumor (positive) or not (negative) were randomized into training (1782 images) and test (343 images) datasets. YOLOv5m architecture was used to detect vertebra on whole CT scans. InceptionV3 architecture with the transfer-learning technique was used to classify the presence/absence of lytic lesions on CT images showing the presence of vertebra. The DL models were evaluated via fivefold cross-validation. For vertebra detection, bounding box accuracy was estimated using intersection over union (IoU). We evaluated the area under the curve (AUC) of a receiver operating characteristic curve to classify lesions. Moreover, we determined the accuracy, precision, recall, and F1 score. We used the gradient-weighted class activation mapping (Grad-CAM) technique for visual interpretation. RESULTS: The computation time was 0.44 s per image. The average IoU value of the predicted vertebra was 0.923 ± 0.052 (0.684-1.000) for test datasets. In the binary classification task, the accuracy, precision, recall, F1-score, and AUC value for test datasets were 0.872, 0.948, 0.741, 0.832, and 0.941, respectively. Heat maps constructed using the Grad-CAM technique were consistent with the location of lytic lesions. CONCLUSION: Our artificial intelligence-aided CAD system using two DL models could rapidly identify vertebra bone from whole CT images and detect lytic spinal bone metastasis, although further evaluation of diagnostic accuracy is required with a larger sample size.

Deep learning-based detection of patients with bone metastasis from Japanese radiology reports.

PURPOSE: Deep learning (DL) is a state-of-the-art technique for developing artificial intelligence in various domains and it improves the performance of natural language processing (NLP). Therefore, we aimed to develop a DL-based NLP model that classifies the status of bone metastasis (BM) in radiology reports to detect patients with BM. MATERIALS AND METHODS: The DL-based NLP model was developed by training long short-term memory using 1,749 free-text radiology reports written in Japanese. We adopted five-fold cross-validation and used 200 reports for testing the five models. The accuracy, sensitivity, specificity, precision, and area under the receiver operating characteristics curve (AUROC) were used for the model evaluation. RESULTS: The developed model demonstrated classification performance with mean ± standard deviation of 0.912 ± 0.012, 0.924 ± 0.029, 0.901 ± 0.014, 0.898 ± 0.012, and 0.968 ± 0.004 for accuracy, sensitivity, specificity, precision, and AUROC, respectively. CONCLUSION: The proposed DL-based NLP model may help in the early and efficient detection of patients with BM.

Patient-specific three-dimensional dose distribution prediction via deep learning for prostate cancer therapy: Improvement with the structure loss.

PURPOSE: Deep learning (DL)-based dose distribution prediction can potentially reduce the cost of inverse planning process. We developed and introduced a structure-focused loss (Lstruct) for 3D dose prediction to improve prediction accuracy. This study investigated the influence of Lstruct on DL-based dose prediction for patients with prostate cancer. The proposed Lstruct, which is similar in concept to dose-volume histogram (DVH)-based optimization in clinical practice, has the potential to provide more interpretable and accurate DL-based optimization. METHODS: This study involved 104 patients who underwent prostate radiotherapy. We used 3D U-Net-based architecture to predict dose distributions from computed tomography and contours of the planning target volume and organs-at-risk. We trained two models using different loss functions: L2 loss and Lstruct. Predicted doses were compared in terms of dose-volume parameters and the Dice similarity coefficient of isodose volume. RESULTS: DVH analysis showed that the Lstruct model had smaller errors from the ground truth than the L2 model. The Lstruct model achieved more consistent dose distributions than the L2 model, with errors close to zero. The isodose Dice score of the Lstruct model was greater than that of the L2 model by >20% of the prescribed dose. CONCLUSIONS: We developed Lstruct using labels of inputted contours for DL-based dose prediction for prostate radiotherapy. Lstruct can be generalized to any DL architecture, thereby enhancing the dose prediction accuracy.

Effective optimization strategy for large optimization volume object, remaining volume at risk (RVR):α-value selection and usage from generalized equivalent uniform dose (gEUD) curve deviation perspective.

Objective.A large optimization volume for intensity-modulated radiation therapy (IMRT), such as the remaining volume at risk (RVR), is traditionally unsuitable for dose-volume constraint control and requires planner-specific empirical considerations owing to the patient-specific shape. To enable less empirical optimization, the generalized equivalent uniform dose (gEUD) optimization is effective; however, the utilization of parametera-values remains elusive. Our study clarifies thea-value characteristics for optimization and to enable effectivea-value use.Approach.The gEUD can be obtained as a function of itsa-value, which is the weighted generalized mean; its curve has a continuous, differentiable, and sigmoid shape, deforming in its optimization state with retained curve characteristics. Using differential geometry, the gEUD curve changes in optimization is considered a geodesic deviation intervened by the forces between deforming and retaining the curve. The curvature and gradient of the curve are radically related to optimization. The vertex point (a=ak) was set and thea-value roles were classified into the following three parts of the curve with respect to thea-value: (i) high gradient and middle curvature, (ii) middle gradient and high curvature, and (iii) low gradient and low curvature. Then, a strategy for multiplea-values was then identified using RVR optimization.Main results.Eleven head and neck patients who underwent static seven-field IMRT were used to verify thea-value characteristics and curvature effect for optimization. The lowera-value (i) (a= 1-3) optimization was effective for the whole dose-volume range; in contrast, the effect of highera-value (iii) (a= 12-20) optimization addressed strongly the high-dose range of the dose volume. The middlea-value (ii) (arounda=ak) showed intermediate but effective high-to-low dose reduction. Thesea-value characteristics were observed as superimpositions in the optimization. Thus, multiple gEUD-based optimization was significantly superior to the exponential constraints normally applied to the RVR that surrounds the PTV, normal tissue objective (NTO), resulting in up to 25.9% and 8.1% improvement in dose-volume indices D2% and V10Gy, respectively.Significance.This study revealed an appropriatea-value for gEUD optimization, leading to favorable dose-volume optimization for the RVR region using fixed multiplea-value conditions, despite the very large and patient-specific shape of the region.

Clinical target volume segmentation based on gross tumor volume using deep learning for head and neck cancer treatment.

Accurate clinical target volume (CTV) delineation is important for head and neck intensity-modulated radiation therapy. However, delineation is time-consuming and susceptible to interobserver variability (IOV). Based on a manual contouring process commonly used in clinical practice, we developed a deep learning (DL)-based method to delineate a low-risk CTV with computed tomography (CT) and gross tumor volume (GTV) input and compared it with a CT-only input. A total of 310 patients with oropharynx cancer were randomly divided into the training set (250) and test set (60). The low-risk CTV and primary GTV contours were used to generate label data for the input and ground truth. A 3D U-Net with a two-channel input of CT and GTV (U-NetGTV) was proposed and its performance was compared with a U-Net with only CT input (U-NetCT). The Dice similarity coefficient (DSC) and average Hausdorff distance (AHD) were evaluated. The time required to predict the CTV was 0.86 s per patient. U-NetGTV showed a significantly higher mean DSC value than U-NetCT (0.80 ± 0.03 and 0.76 ± 0.05) and a significantly lower mean AHD value (3.0 ± 0.5 mm vs 3.5 ± 0.7 mm). Compared to the existing DL method with only CT input, the proposed GTV-based segmentation using DL showed a more precise low-risk CTV segmentation for head and neck cancer. Our findings suggest that the proposed method could reduce the contouring time of a low-risk CTV, allowing the standardization of target delineations for head and neck cancer.

Assessment of using a gamma index analysis for patient-specific quality assurance in Japan.

PURPOSE: The Task Group 218 (TG-218) report was published by the American Association of Physicists in Medicine in 2018, recommending the appropriate use of gamma index analysis for patient-specific quality assurance (PSQA). The paper demonstrates that PSQA for radiotherapy in Japan appropriately applies the gamma index analysis considering TG-218. MATERIALS/METHODS: This survey estimated the acceptance state of radiotherapeutic institutes or facilities in Japan for the guideline using a web-based questionnaire. To investigate an appropriate PSQA of the facility-specific conditions, we researched an optimal tolerance or action level for various clinical situations, including different treatment machines, clinical policies, measurement devices, staff or their skills, and patient conditions. The responded data were analyzed using principal component analysis (PCA) and multidimensional scaling (MDS). The PCA focused on factor loading values of the first contribution over 0.5, whereas the MDS focused on mapped distances among data. RESULTS: Responses were obtained from 148 facilities that use intensity-modulated radiation therapy (IMRT), which accounted for 42.8% of the probable IMRT use in Japan. This survey revealed the appropriate application of the following universal criteria for gamma index analysis from the guideline recommendation despite the facility-specific variations (treatment machines/the number of IMRT cases/facility attributes/responded [representative] expertise or staff): (a) 95% pass rate, (b) 3% dose difference and 2-mm distance-to-agreement, and (c) 10% threshold dose. Conditions (a)-(c) were the principal components of the data by the PCA method and were mapped in a similar distance range, which was easily clustered from other gamma index analytic factors by the MDS method. Conditions (a)-(c) were the universally essential factors for the PSQA in Japan. CONCLUSION: We found that the majority of facilities using IMRT in each region of Japan complied with the guideline and conducted PSQA with deliberation under the individual facility-specific conditions.

Evaluation approach for whole dose distribution in clinical cases using spherical projection and spherical harmonics expansion: spherical coefficient tensor and score method.

Whole dose distribution results from well-conceived treatment plans including patient-specific (location, size and shape of tumor, etc.) and facility-specific (clinical policy and goal, equipment, etc.) information. To evaluate the whole dose distribution efficiently and effectively, we propose a method to apply spherical projection and real spherical harmonics (SH) expansion, thus leading to the expanded coefficients as a rank-2 tensor, SH coefficient tensor, for every patient-specific dose distribution. To verify the feature of this tensor, we introduce Isomap from the manifold learning method and multi-dimensional scaling (MDS). Subsequently, we obtained the MDS distance representing similarity, η, and the SH score, ζ, which is a Frobenius norm of the SH coefficient tensor. These were then validated in the intensity-modulated radiation therapy (IMRT) data sets of: (i) 375 mixing treated regions, (ii) 135 head and neck (HN), and (iii) 132 prostate cases, respectively. The MDS map indicated that the SH coefficient tensor enabled a quantitative feature extraction of whole dose distributions. In particular, the SH score systematically detected irregular cases as the deviation higher than +1.5 standard deviations (SD) from the average case, which matched up with clinically irregular case that required very complicated dose distributions. In summary, the proposed SH coefficient tensor is a useful representation of the whole dose distribution. The SH score from the SH coefficient tensor is a convenient and simple criterion used to characterize the entire dose distributions, which is not dependent on the data set.

Deep learning-based metal artifact reduction using cycle-consistent adversarial network for intensity-modulated head and neck radiation therapy treatment planning.

PURPOSE: To develop a deep learning-based metal artifact reduction (DL-MAR) method using unpaired data and to evaluate its dosimetric impact in head and neck intensity-modulated radiation therapy (IMRT) compared with the water density override method. METHODS: The data set comprised the data of 107 patients who underwent radiotherapy. Fifteen patients with dental fillings were used as the test data set. The computed tomography (CT) images of the remaining 92 patients were divided into two domains: the metal artifact and artifact-free domains. CycleGAN was used for domain translation. The artifact index of the DL-MAR images was compared with that of the original uncorrected (UC) CT images. The dose distributions of the DL-MAR and UC plans were created by comparing the reference clinical plan with the water density override method (water plan) in each dataset. Dosimetric deviation in the oral cavity from the water plan was evaluated. RESULTS: The artifact index of the DL-MAR images was significantly smaller than that of the UC images in all patients (13.2 ± 4.3 vs. 267.3 ± 113.7). Compared with the reference water plan, dose differences of the UC plans were greater than those of the DL-MAR plans. DL-MAR images provided dosimetric results that were more similar to those of the water plan than the UC plan. CONCLUSIONS: We developed a fast DL-MAR method using CycleGAN for head and neck IMRT. The proposed method could provide consistent dose calculation against metal artifact and improve the efficiency of the planning process by eliminating manual delineation.

Robust plan optimization using edge-enhanced intensity for intrafraction organ deformation in prostate intensity-modulated radiation therapy.

This study evaluated a method for prostate intensity-modulated radiation therapy (IMRT) based on edge-enhanced (EE) intensity in the presence of intrafraction organ deformation using the data of 37 patients treated with step-and-shoot IMRT. On the assumption that the patient setup error was already accounted for by image guidance, only organ deformation over the treatment course was considered. Once the clinical target volume (CTV), rectum, and bladder were delineated and assigned dose constraints for dose optimization, each voxel in the CTV derived from the DICOM RT-dose grid could have a stochastic dose from the different voxel location according to the probability density function as an organ deformation. The stochastic dose for the CTV was calculated as the mean dose at the location through changing the voxel location randomly 1000 times. In the EE approach, the underdose region in the CTV was delineated and optimized with higher dose constraints that resulted in an edge-enhanced intensity beam to the CTV. This was compared to a planning target volume (PTV) margin (PM) approach in which a CTV to PTV margin equivalent to the magnitude of organ deformation was added to obtain an optimized dose distribution. The total monitor units, number of segments, and conformity index were compared between the two approaches, and the dose based on the organ deformation of the CTV, rectum, and bladder was evaluated. The total monitor units, number of segments, and conformity index were significantly lower with the EE approach than with the PM approach, while maintaining the dose coverage to the CTV with organ deformation. The dose to the rectum and bladder were significantly reduced in the EE approach compared with the PM approach. We conclude that the EE approach is superior to the PM with regard to intrafraction organ deformation.

A concept for classification of optimal breathing pattern for use in radiotherapy tracking, based on respiratory tumor kinematics and minimum jerk analysis.

PURPOSE: During radiotherapy, maintaining the patient in a relaxed and comfortable state helps ensure respiratory regularity and reproducibility, thereby supports accurate respiratory tracking/gating treatment. Criteria to evaluate respiratory naturalness, regularity, and phase robustness are therefore needed to aid for the treatment system numerically and medical observers visually. This study introduces a new concept of respiratory tumor kinematics that describes the trajectory of tumor motion with respiration, leading to the minimum jerk theory. Using this theory, this study proposes novel respiratory criteria for respiratory naturalness, regularity, and phase robustness. METHODS: According to respiratory tumor kinematics, tumor motion follows the minimum curvature/jerk trajectory in 4D spacetime. Using this theory, the following three respiratory criteria are proposed: (1) respiratory naturalness Us, the residual sum of the squared difference between the normalized average free respiratory wave (single inhalation/exhalation averaged over each 10 phases) and the normalized minimum jerk theoretical respiratory wave; (2) respiratory regularity Cj16, the cumulative jerk squared cost function sampling every 0.2 s with a peak adjustment coefficient, 16; and (3) respiratory phase robustness (LΔ), a second-order partial differential in the respiratory position for regarded Cj16 as the respiratory position function. To verify these respiratory criteria, values obtained from CyberKnife tracking marker log data for 15 patients were compared with regard to the correlation error between the correlation model and the imaged tumor position, as well as with the number of remodels. The Cj16 growth curve was also compared between 15 patients and 15 healthy volunteers. RESULTS: In the 15 patients, data with Us < 1 and Cj16(60 s) < 10 000 satisfied average/maximum correlation errors of less than 1/3 mm. Data with higher Us values (less respiratory naturalness) and higher Cj16(60 s) values (less respiratory regularity) demonstrated more than 3 mm average/5 mm maximum correlation errors and an increased number of remodels. The data for the 15 patients and 15 volunteers demonstrated that the Cj16 growth curve over 120 s from the start of sampling indicated patient-specific respiratory trends and that the distribution of LΔ clearly showed the respiratory phase shift. In 22 of 30 subjects, the degree of change in the Cj growth curve trends from 60 to 120 s was 22% ± 13% (average ± SD). In contrast, the residual data observed when Cj16 > 1000 showed minimum and mean changes of 91% and 180%, respectively. CONCLUSIONS: The authors developed and verified novel respiratory criteria for respiratory naturalness, regularity, and phase robustness obtained using respiratory tumor kinematics and minimum jerk analysis. These criteria should be useful in monitoring respiratory trends on a real-time basis during treatment, as well as in selecting optimal breathing for tracking/gating radiation treatment and defining numerical goals for respiratory training/gating.

Reference respiratory waveforms by minimum jerk model analysis.

PURPOSE: CyberKnife(®) robotic surgery system has the ability to deliver radiation to a tumor subject to respiratory movements using Synchrony(®) mode with less than 2 mm tracking accuracy. However, rapid and rough motion tracking causes mechanical tracking errors and puts mechanical stress on the robotic joint, leading to unexpected radiation delivery errors. During clinical treatment, patient respiratory motions are much more complicated, suggesting the need for patient-specific modeling of respiratory motion. The purpose of this study was to propose a novel method that provides a reference respiratory wave to enable smooth tracking for each patient. METHODS: The minimum jerk model, which mathematically derives smoothness by means of jerk, or the third derivative of position and the derivative of acceleration with respect to time that is proportional to the time rate of force changed was introduced to model a patient-specific respiratory motion wave to provide smooth motion tracking using CyberKnife(®). To verify that patient-specific minimum jerk respiratory waves were being tracked smoothly by Synchrony(®) mode, a tracking laser projection from CyberKnife(®) was optically analyzed every 0.1 s using a webcam and a calibrated grid on a motion phantom whose motion was in accordance with three pattern waves (cosine, typical free-breathing, and minimum jerk theoretical wave models) for the clinically relevant superior-inferior directions from six volunteers assessed on the same node of the same isocentric plan. RESULTS: Tracking discrepancy from the center of the grid to the beam projection was evaluated. The minimum jerk theoretical wave reduced the maximum-peak amplitude of radial tracking discrepancy compared with that of the waveforms modeled by cosine and typical free-breathing model by 22% and 35%, respectively, and provided smooth tracking for radial direction. Motion tracking constancy as indicated by radial tracking discrepancy affected by respiratory phase was improved in the minimum jerk theoretical model by 7.0% and 13% compared with that of the waveforms modeled by cosine and free-breathing model, respectively. CONCLUSIONS: The minimum jerk theoretical respiratory wave can achieve smooth tracking by CyberKnife(®) and may provide patient-specific respiratory modeling, which may be useful for respiratory training and coaching, as well as quality assurance of the mechanical CyberKnife(®) robotic trajectory.

Development and reproducibility evaluation of a Monte Carlo-based standard LINAC model for quality assurance of multi-institutional clinical trials.

Technical developments in radiotherapy (RT) have created a need for systematic quality assurance (QA) to ensure that clinical institutions deliver prescribed radiation doses consistent with the requirements of clinical protocols. For QA, an ideal dose verification system should be independent of the treatment-planning system (TPS). This paper describes the development and reproducibility evaluation of a Monte Carlo (MC)-based standard LINAC model as a preliminary requirement for independent verification of dose distributions. The BEAMnrc MC code is used for characterization of the 6-, 10- and 15-MV photon beams for a wide range of field sizes. The modeling of the LINAC head components is based on the specifications provided by the manufacturer. MC dose distributions are tuned to match Varian Golden Beam Data (GBD). For reproducibility evaluation, calculated beam data is compared with beam data measured at individual institutions. For all energies and field sizes, the MC and GBD agreed to within 1.0% for percentage depth doses (PDDs), 1.5% for beam profiles and 1.2% for total scatter factors (Scps.). Reproducibility evaluation showed that the maximum average local differences were 1.3% and 2.5% for PDDs and beam profiles, respectively. MC and institutions' mean Scps agreed to within 2.0%. An MC-based standard LINAC model developed to independently verify dose distributions for QA of multi-institutional clinical trials and routine clinical practice has proven to be highly accurate and reproducible and can thus help ensure that prescribed doses delivered are consistent with the requirements of clinical protocols.