A comprehensive lung CT landmark pair dataset for evaluating deformable image registration algorithms.

PURPOSE: Deformable image registration (DIR) is a key enabling technology in many diagnostic and therapeutic tasks, but often does not meet the required robustness and accuracy for supporting clinical tasks. This is in large part due to a lack of high-quality benchmark datasets by which new DIR algorithms can be evaluated. Our team was supported by the National Institute of Biomedical Imaging and Bioengineering to develop DIR benchmark dataset libraries for multiple anatomical sites, comprising of large numbers of highly accurate landmark pairs on matching blood vessel bifurcations. Here we introduce our lung CT DIR benchmark dataset library, which was developed to improve upon the number and distribution of landmark pairs in current public lung CT benchmark datasets. ACQUISITION AND VALIDATION METHODS: Thirty CT image pairs were acquired from several publicly available repositories as well as authors' institution with IRB approval. The data processing workflow included multiple steps: (1) The images were denoised. (2) Lungs, airways, and blood vessels were automatically segmented. (3) Bifurcations were directly detected on the skeleton of the segmented vessel tree. (4) Falsely identified bifurcations were filtered out using manually defined rules. (5) A DIR was used to project landmarks detected on the first image onto the second image of the image pair to form landmark pairs. (6) Landmark pairs were manually verified. This workflow resulted in an average of 1262 landmark pairs per image pair. Estimates of the landmark pair target registration error (TRE) using digital phantoms were 0.4 mm ± 0.3 mm. DATA FORMAT AND USAGE NOTES: The data is published in Zenodo at https://doi.org/10.5281/zenodo.8200423. Instructions for use can be found at https://github.com/deshanyang/Lung-DIR-QA. POTENTIAL APPLICATIONS: The dataset library generated in this work is the largest of its kind to date and will provide researchers with a new and improved set of ground truth benchmarks for quantitatively validating DIR algorithms within the lung.

Technical note: Minimizing CIED artifacts on a 0.35 T MRI-Linac using deep learning.

BACKGROUND: Artifacts from implantable cardioverter defibrillators (ICDs) are a challenge to magnetic resonance imaging (MRI)-guided radiotherapy (MRgRT). PURPOSE: This study tested an unsupervised generative adversarial network to mitigate ICD artifacts in balanced steady-state free precession (bSSFP) cine MRIs and improve image quality and tracking performance for MRgRT. METHODS: Fourteen healthy volunteers (Group A) were scanned on a 0.35 T MRI-Linac with and without an MR conditional ICD taped to their left pectoral to simulate an implanted ICD. bSSFP MRI data from 12 of the volunteers were used to train a CycleGAN model to reduce ICD artifacts. The data from the remaining two volunteers were used for testing. In addition, the dataset was reorganized three times using a Leave-One-Out scheme. Tracking metrics [Dice similarity coefficient (DSC), target registration error (TRE), and 95 percentile Hausdorff distance (95% HD)] were evaluated for whole-heart contours. Image quality metrics [normalized root mean square error (nRMSE), peak signal-to-noise ratio (PSNR), and multiscale structural similarity (MS-SSIM) scores] were evaluated. The technique was also tested qualitatively on three additional ICD datasets (Group B) including a patient with an implanted ICD. RESULTS: For the whole-heart contour with CycleGAN reconstruction: 1) Mean DSC rose from 0.910 to 0.935; 2) Mean TRE dropped from 4.488 to 2.877 mm; and 3) Mean 95% HD dropped from 10.236 to 7.700 mm. For the whole-body slice with CycleGAN reconstruction: 1) Mean nRMSE dropped from 0.644 to 0.420; 2) Mean MS-SSIM rose from 0.779 to 0.819; and 3) Mean PSNR rose from 18.744 to 22.368. The three Group B datasets evaluated qualitatively displayed a reduction in ICD artifacts in the heart. CONCLUSION: CycleGAN-generated reconstructions significantly improved both tracking and image quality metrics when used to mitigate artifacts from ICDs.

A deep learning approach to remove contrast from contrast-enhanced CT for proton dose calculation.

PURPOSE: Non-Contrast Enhanced CT (NCECT) is normally required for proton dose calculation while Contrast Enhanced CT (CECT) is often scanned for tumor and organ delineation. Possible tissue motion between these two CTs raises dosimetry uncertainties, especially for moving tumors in the thorax and abdomen. Here we report a deep-learning approach to generate NCECT directly from CECT. This method could be useful to avoid the NCECT scan, reduce CT simulation time and imaging dose, and decrease the uncertainties caused by tissue motion between otherwise two different CT scans. METHODS: A deep network was developed to convert CECT to NCECT. The network receives a 3D image from CECT images as input and generates a corresponding contrast-removed NCECT image patch. Abdominal CECT and NCECT image pairs of 20 patients were deformably registered and 8000 image patch pairs extracted from the registered image pairs were utilized to train and test the model. CTs of clinical proton patients and their treatment plans were employed to evaluate the dosimetric impact of using the generated NCECT for proton dose calculation. RESULTS: Our approach achieved a Cosine Similarity score of 0.988 and an MSE value of 0.002. A quantitative comparison of clinical proton dose plans computed on the CECT and the generated NCECT for five proton patients revealed significant dose differences at the distal of beam paths. V100% of PTV and GTV changed by 3.5% and 5.5%, respectively. The mean HU difference for all five patients between the generated and the scanned NCECTs was ∼4.72, whereas the difference between CECT and the scanned NCECT was ∼64.52, indicating a ∼93% reduction in mean HU difference. CONCLUSIONS: A deep learning approach was developed to generate NCECTs from CECTs. This approach could be useful for the proton dose calculation to reduce uncertainties caused by tissue motion between CECT and NCECT.

Respiratory motion management using a single rapid MRI scan for a 0.35 T MRI-Linac system.

BACKGROUND: MRI has a rapidly growing role in radiation therapy (RT) for treatment planning, real-time image guidance, and beam gating (e.g., MRI-Linac). Free-breathing 4D-MRI is desirable in respiratory motion management for therapy. Moreover, high-quality 3D-MRIs without motion artifacts are needed to delineate lesions. Existing MRI methods require multiple scans with lengthy acquisition times or are limited by low spatial resolution, contrast, and signal-to-noise ratio. PURPOSE: We developed a novel method to obtain motion-resolved 4D-MRIs and motion-integrated 3D-MRI reconstruction using a single rapid (35-45 s scan on a 0.35 T MRI-Linac. METHODS: Golden-angle radial stack-of-stars MRI scans were acquired from a respiratory motion phantom and 12 healthy volunteers (n = 12) on a 0.35 T MRI-Linac. A self-navigated method was employed to detect respiratory motion using 2000 (acquisition time = 5-7 min) and the first 200 spokes (acquisition time = 35-45 s). Multi-coil non-uniform fast Fourier transform (MCNUFFT), compressed sensing (CS), and deep-learning Phase2Phase (P2P) methods were employed to reconstruct motion-resolved 4D-MRI using 2000 spokes (MCNUFFT2000) and 200 spokes (CS200 and P2P200). Deformable motion vector fields (MVFs) were computed from the 4D-MRIs and used to reconstruct motion-corrected 3D-MRIs with the MOtion Transformation Integrated forward-Fourier (MOTIF) method. Image quality was evaluated quantitatively using the structural similarity index measure (SSIM) and the root mean square error (RMSE), and qualitatively in a blinded radiological review. RESULTS: Evaluation using the respiratory motion phantom experiment showed that the proposed method reversed the effects of motion blurring and restored edge sharpness. In the human study, P2P200 had smaller inaccuracy in MVFs estimation than CS200. P2P200 had significantly greater SSIMs (p < 0.0001) and smaller RMSEs (p < 0.001) than CS200 in motion-resolved 4D-MRI and motion-corrected 3D-MRI. The radiological review found that MOTIF 3D-MRIs using MCNUFFT2000 exhibited the highest image quality (scoring > 8 out of 10), followed by P2P200 (scoring > 5 out of 10), and then motion-uncorrected (scoring < 3 out of 10) in sharpness, contrast, and artifact-freeness. CONCLUSIONS: We have successfully demonstrated a method for respiratory motion management for MRI-guided RT. The method integrated self-navigated respiratory motion detection, deep-learning P2P 4D-MRI reconstruction, and a motion integrated reconstruction (MOTIF) for 3D-MRI using a single rapid MRI scan (35-45 s) on a 0.35 T MRI-Linac system.

Implementation of a knowledge-based decision support system for treatment plan auditing through automation.

BACKGROUND: Independent auditing is a necessary component of a comprehensive quality assurance (QA) program and can also be utilized for continuous quality improvement (QI) in various radiotherapy processes. Two senior physicists at our institution have been performing a time intensive manual audit of cross-campus treatment plans annually, with the aim of further standardizing our planning procedures, updating policies and guidelines, and providing training opportunities of all staff members. PURPOSE: A knowledge-based automated anomaly-detection algorithm to provide decision support and strengthen our manual retrospective plan auditing process was developed. This standardized and improved the efficiency of the assessment of our external beam radiotherapy (EBRT) treatment planning across all eight campuses of our institution. METHODS: A total of 843 external beam radiotherapy plans for 721 lung patients from January 2020 to March 2021 were automatically acquired from our clinical treatment planning and management systems. From each plan, 44 parameters were automatically extracted and pre-processed. A knowledge-based anomaly detection algorithm, namely, "isolation forest" (iForest), was then applied to the plan dataset. An anomaly score was determined for each plan using recursive partitioning mechanism. Top 20 plans ranked with the highest anomaly scores for each treatment technique (2D/3D/IMRT/VMAT/SBRT) including auto-populated parameters were used to guide the manual auditing process and validated by two plan auditors. RESULTS: The two auditors verified that 75.6% plans with the highest iForest anomaly scores have similar concerning qualities that may lead to actionable recommendations for our planning procedures and staff training materials. The time to audit a chart was approximately 20.8 min on average when done manually and 14.0 min when done with the iForest guidance. Approximately 6.8 min were saved per chart with the iForest method. For our typical internal audit review of 250 charts annually, the total time savings are approximately 30 hr per year. CONCLUSION: iForest effectively detects anomalous plans and strengthens our cross-campus manual plan auditing procedure by adding decision support and further improve standardization. Due to the use of automation, this method was efficient and will be used to establish a standard plan auditing procedure, which could occur more frequently.

Deep learning-based motion compensation for four-dimensional cone-beam computed tomography (4D-CBCT) reconstruction.

BACKGROUND: Motion-compensated (MoCo) reconstruction shows great promise in improving four-dimensional cone-beam computed tomography (4D-CBCT) image quality. MoCo reconstruction for a 4D-CBCT could be more accurate using motion information at the CBCT imaging time than that obtained from previous 4D-CT scans. However, such data-driven approaches are hampered by the quality of initial 4D-CBCT images used for motion modeling. PURPOSE: This study aims to develop a deep-learning method to generate high-quality motion models for MoCo reconstruction to improve the quality of final 4D-CBCT images. METHODS: A 3D artifact-reduction convolutional neural network (CNN) was proposed to improve conventional phase-correlated Feldkamp-Davis-Kress (PCF) reconstructions by reducing undersampling-induced streaking artifacts while maintaining motion information. The CNN-generated artifact-mitigated 4D-CBCT images (CNN enhanced) were then used to build a motion model which was used by MoCo reconstruction (CNN+MoCo). The proposed procedure was evaluated using in-vivo patient datasets, an extended cardiac-torso (XCAT) phantom, and the public SPARE challenge datasets. The quality of reconstructed images for XCAT phantom and SPARE datasets was quantitatively assessed using root-mean-square-error (RMSE) and normalized cross-correlation (NCC). RESULTS: The trained CNN effectively reduced the streaking artifacts of PCF CBCT images for all datasets. More detailed structures can be recovered using the proposed CNN+MoCo reconstruction procedure. XCAT phantom experiments showed that the accuracy of estimated motion model using CNN enhanced images was greatly improved over PCF. CNN+MoCo showed lower RMSE and higher NCC compared to PCF, CNN enhanced and conventional MoCo. For the SPARE datasets, the average (± standard deviation) RMSE in mm-1 for body region of PCF, CNN enhanced, conventional MoCo and CNN+MoCo were 0.0040 ± 0.0009, 0.0029 ± 0.0002, 0.0024 ± 0.0003 and 0.0021 ± 0.0003. Corresponding NCC were 0.84 ± 0.05, 0.91 ± 0.05, 0.91 ± 0.05 and 0.93 ± 0.04. CONCLUSIONS: CNN-based artifact reduction can substantially reduce the artifacts in the initial 4D-CBCT images. The improved images could be used to enhance the motion modeling and ultimately improve the quality of the final 4D-CBCT images reconstructed using MoCo.

Contour interpolation by deep learning approach.

Purpose: Contour interpolation is an important tool for expediting manual segmentation of anatomical structures. The process allows users to manually contour on discontinuous slices and then automatically fill in the gaps, therefore saving time and efforts. The most used conventional shape-based interpolation (SBI) algorithm, which operates on shape information, often performs suboptimally near the superior and inferior borders of organs and for the gastrointestinal structures. In this study, we present a generic deep learning solution to improve the robustness and accuracy for contour interpolation, especially for these historically difficult cases. Approach: A generic deep contour interpolation model was developed and trained using 16,796 publicly available cases from 5 different data libraries, covering 15 organs. The network inputs were a 128 × 128 × 5 image patch and the two-dimensional contour masks for the top and bottom slices of the patch. The outputs were the organ masks for the three middle slices. The performance was evaluated on both dice scores and distance-to-agreement (DTA) values. Results: The deep contour interpolation model achieved a dice score of 0.95 ± 0.05 and a mean DTA value of 1.09 ± 2.30 mm , averaged on 3167 testing cases of all 15 organs. In a comparison, the results by the conventional SBI method were 0.94 ± 0.08 and 1.50 ± 3.63 mm , respectively. For the difficult cases, the dice score and DTA value were 0.91 ± 0.09 and 1.68 ± 2.28 mm by the deep interpolator, compared with 0.86 ± 0.13 and 3.43 ± 5.89 mm by SBI. The t-test results confirmed that the performance improvements were statistically significant ( p < 0.05 ) for all cases in dice scores and for small organs and difficult cases in DTA values. Ablation studies were also performed. Conclusions: A deep learning method was developed to enhance the process of contour interpolation. It could be useful for expediting the tasks of manual segmentation of organs and structures in the medical images.

Real-time B0 compensation during gantry rotation in a 0.35-T MRI-Linac.

BACKGROUND: Rotation of the ferromagnetic gantry of a low magnetic field MRI-Linac was previously demonstrated to cause large center frequency offsets of ±400 Hz. The B0 off-resonances cause image artifacts and imaging isocenter shifts that would preclude MRI-guided arc therapy. PURPOSE: The purpose of this study was to measure and compensate for center frequency offsets in real time during gantry rotation on a 0.35-T MRI-Linac using a free induction decay (FID) navigator. METHODS: A nonselective FID navigator was added before each 2D balanced steady-state free precession cine image acquisition on a 0.35-T MRI-Linac. Images were acquired at 7.3 frames per second. Phase data from the initial FID navigator (while the gantry was stationary) was used as a reference. The phase data from each subsequent FID navigator was used to calculate the real-time B0 off-resonance. The transmitter/receiver phase and the phase accrual over the adjacent image acquisition were adjusted to correct for the center frequency offset. Measurements were performed using an MRI-Linac dynamic phantom prior to and while the gantry rotated clockwise and counterclockwise. Image quality and signal-to-noise ratio (SNR) were compared between uncorrected and B0 -corrected MRIs using a reference image acquired while the gantry was stationary. Four targets in the phantom were manually contoured on the first image frame, and an active contouring algorithm was used retrospectively on each subsequent frame to assess image variations and calculate Dice coefficients. Additionally, three healthy volunteers were imaged using the same pulse sequences with and without real-time B0 compensation during gantry rotation. Normalized root mean square errors (nRMSEs) were calculated for the phantom and in vivo to assess the efficacy of the B0 compensation on image quality. The measured center frequency offsets from the volunteer and MRI dynamic phantom navigator data were also compared. The sinusoidal behavior of the center frequency offsets was modeled based on the gantry layout and long-time constant eddy currents resulting from gantry rotation. RESULTS: The duration of the FID navigator and processing was 4.5 ms. The FID navigator resulted in a ≤11% drop in SNR in the phantom and in vivo (liver). Dice coefficients from the MRI-guided radiation therapy (MR-IGRT) phantom contour measurements remained above 0.8 with B0 compensation. Without B0 compensation, the Dice coefficients dropped below 0.8 for up to 21% of the time depending on the contour. Real-time B0 compensation resulted in mean reductions in nRMSE of 51% and 16% for the MR-IGRT phantom and in vivo, respectively. Peak-to-peak center frequency offsets ranged from 757 to 773 Hz in the phantom and 760 to 871 Hz in vivo. CONCLUSION: Dynamic real-time B0 compensation significantly improved image quality and reduced artifacts during gantry rotation in the phantom and in vivo. However, the FID navigator resulted in a small drop in the imaging duty cycle and SNR.

A reconstruction approach for proton computed tomography by modeling the integral depth dose of the scanning proton pencil beam.

PURPOSE: To present a proton computed tomography (pCT) reconstruction approach that models the integral depth dose (IDD) of the clinical scanning proton beam into beamlets. Using a multilayer ionization chamber (MLIC) as the imager, the proposed pCT system and the reconstruction approach can minimize extra ambient neutron dose and simplify the beamline design by eliminating an additional collimator to confine the proton beam. METHODS: Monte Carlo simulation was applied to digitally simulate the IDDs of the exiting proton beams detected by the MLIC. A forward model was developed to model each IDD into a weighted sum of percentage depth doses of the constituent beamlets separated laterally by 1 mm. The water equivalent path lengths (WEPLs) of the beamlets were determined by iteratively minimizing the squared L2-norm between the forward projected and simulated IDDs. The final WEPL values were reconstructed to pCT images, that is, proton stopping power ratio (SPR) maps, through simultaneous algebraic reconstruction technique with total variation regularization. The reconstruction process was tested with a digital cylindrical water-based phantom and an ICRP adult reference computational phantom. The mean of SPR within regions of interest (ROIs) and the WEPL along a 4 mm-wide beam ( WEP L 4 mm ${\rm{WEP}}{{\rm{L}}_{4{\rm{mm}}}}$ ) were compared with the reference values. The spatial resolution was analyzed at the edge of a cortical insert of the cylindrical phantom. RESULTS: The percentage deviations from reference SPR were within ±1% in all selected ROIs. The mean absolute error of the reconstructed SPR was 0.33%, 0.19%, and 0.27% for the cylindrical phantom, the adult phantom at the head and lung region, respectively. The corresponding percentage deviations from reference WEP L 4 mm ${\rm{WEP}}{{\rm{L}}_{4{\rm{mm}}}}$ were 0.48 ± 0.64%, 0.28 ± 0.48%, and 0.22 ± 0.49%. The full width at half maximum of the line spread function (LSF) derived from the radial edge spread function (ESF) of a cortical insert was 0.13 cm. The frequency at 10% of the modulation transfer function (MTF) was 6.38 cm-1 . The mean signal-to-noise ratio (SNR) of all the inserts was 2.45. The mean imaging dose was 0.29 and 0.25 cGy at the head and lung region of the adult phantom, respectively. CONCLUSION: A new pCT reconstruction approach was developed by modeling the IDDs of the uncollimated scanning proton beams in the pencil beam geometry. SPR accuracy within ±1%, spatial resolution of better than 2 mm at 10% MTF, and imaging dose at the magnitude of mGy were achieved. Potential side effects caused by neutron dose were eliminated by removing the extra beam collimator.

Technical Note: Effects of rotating gantry on magnetic field and eddy currents in 0.35 T MRI-guided radiotherapy (MR-IGRT) system.

PURPOSE: The purpose of this study was to identify the cause of severe image artifacts that occurred during gantry rotation in a 0.35 T MRI-Linac by comparing measurements of eddy currents, center frequency, and field inhomogeneities made with the gantry in motion and stationary. METHODS: Gradient and B0 eddy currents were calculated from the free induction decays (FIDs) resulting from selective excitation at a temporal resolution of 200 ms/measurement. B0 eddy currents were also calculated from FIDs acquired with nonselective excitation at a temporal resolution of 100 ms/measurement. Center frequencies and B0 inhomogeneities were measured by acquiring FIDs with a repetition time (TR) of 290 ms. Cartesian and radial 2D true fast imaging with steady-state precession (TrueFISP) pulse sequences used in real-time MRI-guided radiation therapy (MR-IGRT) were acquired. To assess artifact severity, the normalized root mean square error (nRMSE) was calculated between a reference MRI (static gantry) and MRIs acquired during gantry rotation for each serial acquisition. Image artifacts were qualitatively graded as nominal, minor, or severe. Measurements were conducted while the gantry was rotated through its entire range for both clockwise and counterclockwise. Measurements during gantry rotation were compared to measurements with a stationary gantry (every 30°). RESULTS: Severe image artifacts were observed 22-35% of the time while the gantry was rotating. Short time constant eddy currents were not affected by gantry rotation. The peak to peak center frequency and FWHM rose by factors of 13.2-14.5 and 1.1-1.6, respectively, for the rotating versus stationary gantry. The magnitude of the center frequency offset and field inhomogeneities depended on the direction of the gantry rotation. CONCLUSIONS: Image artifacts during gantry rotation were primarily caused by center frequency variations and field inhomogeneities. Therefore, dynamic B0 compensation techniques should be able to reduce artifacts during gantry rotation.

Development of a storage phosphor imaging system for proton pencil beam spot profile determination.

PURPOSE: Accurate two-dimensional (2D) profile measurements at submillimeter precision are necessary for proton beam commissioning and periodic quality assurance (QA) purposes and are currently performed at our institution with a commercial scintillation detector (Lynx PT) with limited means for independent checks. The purpose of this work was to create an independent dosimetry system consisting of an in-house optical scanner and a BaFBrI:Eu2+ storage phosphor dosimeter by: (a) determining the optimal settings for the optical scanner, (b) measuring 2D proton spot profiles with the storage phosphors, and (c) comparing them to similar measurements using a commercial scintillation detector. METHODS: An in-house 2D laboratory optical scanner was constructed and spatially calibrated for accurate 2D photostimulated luminescence (PSL) dosimetry. Square 5 × 5 cm2 BaFBrI:Eu2+ dosimeter samples were uniformly irradiated with line scans performed to determine the physical and electronic scanner settings resulting in the highest signal-to-noise ratios (SNR) at a sub-millimeter spatial resolution. The resultant spatial resolution of the scanner was then quantitatively assessed by measuring (a) line pairs on a standard X-ray lead bar phantom and (b) modulation transfer functions. Following this, 2D proton spot profiles from a Mevion S250i Hyperscan proton unit were obtained at 1, 10, 20, 30, 40, and 50 monitor unit (MU) settings at maximum energy (E0  = 227.1 MeV) and compared to baseline profiles from a commercial scintillation detector, where 1 MU is calibrated to deliver 1 Gy absolute proton dose-to-water under reference conditions, that is, 41 × 41 proton spots uniformly spaced by 0.25 cm within a 10 × 10 cm2 square field size at maximum energy (227.1 MeV) in water at depth of 5 cm at isocenter. The dosimetric system's sensitivities to (a) ±1 mm positional shifts and (b) ±0.3 mm beam lateral spread changes were quantitatively evaluated through a Gaussian fitting of the crossline and inline plots of the respective artificially shifted beam profiles. RESULTS: The physical scanner settings of (a) Δτ = 27 ms time interval between data samples, (b) vx  = 1.235 cm/s scanning speed, (c) 1% laser transmission (0.02 mW power) and (d) (Δx, Δy) = (0.33, 0.50 mm) pixel sizes with electronic settings of (a) 300 microseconds time constant, (b) normal dynamic reserve, (c) 24 dB/oct low pass filter slope, and (d) 160 Hz chopping frequency resulted in the highest SNR while maintaining sub-millimeter spatial resolution. The BaFBr0.85 I0.15 :Eu2+ storage phosphor dosimeters were linear from 1 to 50 MU and their profiles did not saturate up to 150 MU. The scanner was able to detect lateral displacements of ±1 mm in both the crossline and inline directions and ±0.3 mm beam spread changes that were artificially introduced by varying the incident proton energy. Specific to our proton unit, proton energy changes of ±1 MeV can also be detected indirectly via beam spread measurements. CONCLUSION: Our combined dosimetric system including an in-house laboratory optical scanner and reusable BaFBr0.85 I0.15 :Eu2+ storage phosphors demonstrated a sufficient spatial resolution and dosimetric accuracy to support its use as an independent proton spot measurement dosimeter system. Its wide dynamic range allows for other versatile applications such as proton halo measurements.

Effects of B0 eddy currents on imaging isocenter shifts in 0.35-T MRI-guided radiotherapy (MR-IGRT) system.

PURPOSE: The purpose of this study was to measure gantry angle-related eddy currents in a 0.35-T MRI-Linac and determine if B0 (zeroth order) eddy currents are the primary cause of gantry angle-dependent imaging isocenter shifts vs other potential causes like B0 inhomogeneities and gradient (first order) eddy currents. For conventional Cartesian acquisitions, B0 eddy currents can cause imaging isocenter shifts along both phase encode and readout directions. Gradient eddy currents can cause spatial distortion along both the phase encode and readout directions. Center frequency offsets can cause imaging isocenter shifts along the readout direction that vary with readout gradient polarity. METHODS: MRI-related eddy currents and imaging isocenter shifts were measured on a 0.35-T MRI-Linac at gantry angles from 0° to 330° in increments of 30° . All measurements were made after gradient shimming and center frequency tuning at each planned gantry angle. Eddy current and field homogeneity measurements were conducted using a 24-cm diameter spherical phantom. Gradient and B0 eddy currents were calculated from the free induction decays (FIDs) resulting from selective excitation of slices located ±5 cm from isocenter. B0 eddy currents were also calculated from FIDs acquired with nonselective excitation and compared with B0 eddy current values derived using selective excitation. B0 inhomogeneities and center frequency offsets were measured by acquiring FIDs with nonselective excitation. Imaging isocenter shifts were measured using a 33x33x10.5 cm3 uniformity linearity (grid) phantom and a 3D true fast imaging with steady-state precession (TrueFISP) sequence used in MRI-guided radiation therapy. Eddy currents were compared to vendor specifications and correlated with the imaging isocenter shifts. Measurements were conducted before and after the MRI-Linac's waveguide was replaced with an updated design to reduce eddy currents. RESULTS: B0 eddy currents were highly correlated (r = 0.986, P << 0.001) for measurements made with vs without selective excitation. Transverse (X and Y) axis B0 eddy currents before and after the waveguide upgrade were out of specification (specification: ≤0.1 µT m/mT for delays < 10 ms) for most of the measured gantry angles. Gradient eddy currents before and after the upgrade were within specifications for the measured gantry angles (≤0.1% for delays < 10 ms). B0 eddy currents and imaging isocenter shifts were highly correlated (r = 0.965, P << 0.001). After the Linac waveguide upgrade, root mean square (RMS) peak B0 and gradient eddy currents dropped 45% and 11%, respectively, for delays <10 ms, while imaging isocenter shifts dropped 53%. Isocenter shifts were observed in both phase encode and readout directions. Center frequency offsets were <26 Hz while B0 inhomogeneities were <33 Hz full width at half maximum (FWHM). CONCLUSIONS: Imaging isocenter shifts measured in a 0.35-T MRI-Linac were highly correlated with B0 eddy currents. The eddy currents and imaging isocenter shifts decreased after the MRI-Linac's waveguide was replaced.

Dual-storage phosphor proton therapy dosimetry: Simultaneous quantification of dose and linear energy transfer.

PURPOSE: To investigate the feasibility of using the high Zeff storage phosphor material BaFBrI:Eu2+ in conjunction with the low Zeff storage phosphor material KCl:Eu2+ for simultaneous proton dose and linear energy transfer (LET) measurements by (a) measuring the fundamental optical and dosimetric properties of BaFBrI:Eu2+ , (b) evaluating its compatibility in being readout simultaneously with KCl:Eu2+ dosimeters, and (c) modeling and validating its LET dependence under elevated proton LET irradiation. METHODS: A commercial BaFBrI:Eu2+ storage phosphor detector (Model ST-VI, Fujifilm) was characterized with energy dispersive x-ray spectroscopy (EDS) analysis to obtain its elemental composition. The dosimeters were irradiated using both a Mevion S250 proton therapy unit (at the center of a spread-out Bragg peak, SOBP) and a Varian Clinac iX linear accelerator with the latter being a low LET irradiation. The photostimulated luminescence (PSL) emission spectra, excitation spectra, and luminescent lifetimes of the detector were measured after proton and photon irradiations. Dosimetric properties including dose linearity, dose rate dependence, radiation hardness, temporal, and readout stabilities were studied using a laboratory optical reader after proton irradiations. In addition, its proton energy dependence was analytically modeled and experimentally validated by irradiating the detectors at various depths within the SOBP (Range: 15.0 g/cm2 , Modulation: 10.0 g/cm2 ). RESULTS: The active detector composition for the high Zeff storage phosphor detector was found to be BaFBr0.85 I0.15 :Eu2+ . The BaFBr0.85 I0.15 :Eu2+ material's excitation and emission spectra were in agreement under proton and photon irradiations, with peaks of 586 ± 1 nm and 400 ± 1 nm, respectively, with a full width at half maximum (FWHM) of 119 ± 3 nm and 30 ± 2 nm, respectively. As dosimeter response under photon irradiation is generally believed to be free from LET effect, these results suggest LET independence of charge storage center types resulted from ionizing radiations. There is sufficient spectral overlaps with KCl:Eu2+ dosimeters allowing both dosimeters to be readout under equivalent readout conditions, that is, 594 nm stimulation and 420 nm detection wavelengths. Its PSL characteristic lifetime was found to be less than 5 microseconds which would make it suitable for fast 2D readout post irradiation. Its 420 nm emission band intensity was found to be linear up to 10 Gy absolute proton dose under the same irradiation conditions, dose rate independent, stable in time and under multiple readouts, and with high radiation hardness under cumulative proton dose histories up to 200 Gy as tested in this study. BaFBr0.85 I0.15 :Eu2+ showed significant proton energy-dependent dose under-response in regions of high LET which could be modeled by stopping power ratio calculations with an accuracy of 3% in low LET regions and a distance-to-agreement (DTA) of 1 mm in high LET regions (>5 keV/µm). CONCLUSION: We have proven the feasibility of dual-storage phosphor proton dosimetry for simultaneous proton dose and LET measurements. BaFBr0.85 I0.15 :Eu2+ has shown equally excellent dosimetry performance as its low Zeff complement KCl:Eu2+ with distinctive LET dependence merely as a result of its higher Zeff . These promising results pave the way for future studies involving simultaneous proton dose and LET measurements using this novel approach.

GroupRegNet: a groupwise one-shot deep learning-based 4D image registration method.

Accurate deformable four-dimensional (4D) (three-dimensional in space and time) medical images registration is essential in a variety of medical applications. Deep learning-based methods have recently gained popularity in this area for the significantly lower inference time. However, they suffer from drawbacks of non-optimal accuracy and the requirement of a large amount of training data. A new method named GroupRegNet is proposed to address both limitations. The deformation fields to warp all images in the group into a common template is obtained through one-shot learning. The use of the implicit template reduces bias and accumulated error associated with the specified reference image. The one-shot learning strategy is similar to the conventional iterative optimization method but the motion model and parameters are replaced with a convolutional neural network and the weights of the network. GroupRegNet also features a simpler network design and a more straightforward registration process, which eliminates the need to break up the input image into patches. The proposed method was quantitatively evaluated on two public respiratory-binned 4D-computed tomography datasets. The results suggest that GroupRegNet outperforms the latest published deep learning-based methods and is comparable to the top conventional method pTVreg. To facilitate future research, the source code is available at https://github.com/vincentme/GroupRegNet.

Factorization Theorem Connecting the Light-Cone Distribution Amplitudes of Heavy-Flavor Mesons in QCD and Heavy-Quark Effective Theory.

The light-cone distribution amplitude (LCDA) of a heavy-light meson defined in heavy quark effective theory (HQET) is a fundamental nonperturbative input to account for innumerable B meson exclusive decay and production processes. On the other hand, the conventional heavy-flavored meson LCDA defined in QCD also ubiquitously enters the factorization formula for hard exclusive B production processes. Inspired by the observation that these two LCDAs exhibit the identical infrared behaviors, yet differ in the ultraviolet scale of order m_{b} or greater, we propose a novel factorization theorem for the heavy-light mesons, that the LCDA defined in QCD can be further expressed as a convolution between the LCDA in HQET and a perturbatively calculable coefficient function thanks to asymptotic freedom. This refactorization program can be invoked to fully disentangle the effects from three disparate scales Q, m_{b}, and Λ_{QCD} for a hard exclusive B production process, particularly to facilitate the resummation of logarithms of type lnQ/m_{b} and lnm_{b}/Λ_{QCD} in a systematic fashion.

Technical Note: Automatic segmentation of CT images for ventral body composition analysis.

PURPOSE: Body composition is known to be associated with many diseases including diabetes, cancers, and cardiovascular diseases. In this paper, we developed a fully automatic body tissue decomposition procedure to segment three major compartments that are related to body composition analysis - subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), and muscle. Three additional compartments - the ventral cavity, lung, and bones - were also segmented during the segmentation process to assist segmentation of the major compartments. METHODS: A convolutional neural network (CNN) model with densely connected layers was developed to perform ventral cavity segmentation. An image processing workflow was developed to segment the ventral cavity in any patient's computed tomography (CT) using the CNN model, then further segment the body tissue into multiple compartments using hysteresis thresholding followed by morphological operations. It is important to segment ventral cavity firstly to allow accurate separation of compartments with similar Hounsfield unit (HU) inside and outside the ventral cavity. RESULTS: The ventral cavity segmentation CNN model was trained and tested with manually labeled ventral cavities in 60 CTs. Dice scores (mean ± standard deviation) for ventral cavity segmentation were 0.966 ± 0.012. Tested on CT datasets with intravenous (IV) and oral contrast, the Dice scores were 0.96 ± 0.02, 0.94 ± 0.06, 0.96 ± 0.04, 0.95 ± 0.04, and 0.99 ± 0.01 for bone, VAT, SAT, muscle, and lung, respectively. The respective Dice scores were 0.97 ± 0.02, 0.94 ± 0.07, 0.93 ± 0.06, 0.91 ± 0.04, and 0.99 ± 0.01 for non-contrast CT datasets. CONCLUSION: A body tissue decomposition procedure was developed to automatically segment multiple compartments of the ventral body. The proposed method enables fully automated quantification of three-dimensional (3D) ventral body composition metrics from CT images.

Quantitative proton radiation therapy dosimetry using the storage phosphor europium-doped potassium chloride.

PURPOSE: To (a) characterize the fundamental optical and dosimetric properties of the storage phosphor europium-doped potassium chloride for quantitative proton dosimetry, and (b) investigate if its dose radiation response can be described by an analytic radiation transport model. METHODS: Cylindrical KCl:Eu2+ dosimeters with dimensions of 6 mm diameter and 1 mm thickness were fabricated in-house. The dosimeters were irradiated using both a Mevion S250 passive scattering proton therapy system and a Varian Clinac iX linear accelerator. Photostimulated luminescence (PSL) emission spectra, excitation spectra, and luminescence lifetimes were measured for both proton and photon irradiations. Dosimetric properties including radiation hardness, dose linearity, signal stabilization, dose rate sensitivity, and energy dependence were studied using a laboratory optical reader after irradiations. The dosimeters were modeled using physical quantities including mass stopping powers in the storage phosphor and water for a given proton beam, and mass energy absorption coefficients and massing stopping powers in detector and water for a given photon beam. RESULTS: KCl:Eu2+ exhibited optical emission and stimulation peaks at 421 and 560 nm, respectively, for both proton and photon irradiations, enabling postirradiation readouts using a visible light source while detecting the PSL using a photomultiplier tube. KCl:Eu2+ showed a linear response from 0 to 8 Gy absorbed dose-to-water, a large dynamic range up to 60 Gy, dose-rate independence measured from 83 to 500 MU/min, and a PSL lifetime of <5 ms that is sufficiently short for supporting rapid scanning in a two-dimensional geometry. KCl:Eu2+ was highly reusable with only a slight signal decrease of ~3% at accumulated doses over 100 Gy, which could be managed by a periodic recalibration. The detected PSL signal strength of the dosimeter in the proton field had been calculated accurately to a maximum discrepancy of 2% using known physical quantities along with its prior signal strength as measured in a photon field at the same dose-to-water. This discrepancy might be attributed to an under-response due to linear energy transfer (LET) effect. However, comparisons of depth-dose measurements in a spread-out Bragg peak (SOBP) field with a parallel-plate ionization chamber showed no clear evidence of LET effects. Furthermore, range measurements agreed with ionization chamber measurements to within 1 mm. CONCLUSIONS: KCl:Eu2+ showed linear response over a large dynamic range for proton irradiations and reliably reproduced SOBP measurements as measured by ionization chambers. Its relatively low atomic number of 18 and near LET independence make it suited for quantitative proton dosimetry. In addition, its high radiation hardness means that it can be reused numerous times. Any potential measurement artifacts encountered in complex irradiation conditions should be able to be corrected for using known physical quantities.

Development and evaluation of machine learning models for voxel dose predictions in online adaptive magnetic resonance guided radiation therapy.

PURPOSE: Daily online adaptive plan quality in magnetic resonance imaging guided radiation therapy (MRgRT) is difficult to assess in relation to the fully optimized, high quality plans traditionally established offline. Machine learning prediction models developed in this work are capable of predicting 3D dose distributions, enabling the evaluation of online adaptive plan quality to better inform adaptive decision-making in MRgRT. METHODS: Artificial neural networks predicted 3D dose distributions from input variables related to patient anatomy, geometry, and target/organ-at-risk relationships in over 300 treatment plans from 53 patients receiving adaptive, linac-based MRgRT for abdominal cancers. The models do not include any beam related variables such as beam angles or fluence and were optimized to balance errors related to raw dose and specific plan quality metrics used to guide daily online adaptive decisions. RESULTS: Averaged over all plans, the dose prediction error and the absolute error were 0.1 ± 3.4 Gy (0.1 ± 6.2%) and 3.5 ± 2.4 Gy (6.4 ± 4.3%) respectively. Plan metric prediction errors were -0.1 ± 1.5%, -0.5 ± 2.1%, -0.9 ± 2.2 Gy, and 0.1 ± 2.7 Gy for V95, V100, D95, and Dmean respectively. Plan metric prediction absolute errors were 1.1 ± 1.1%, 1.5 ± 1.5%, 1.9 ± 1.4 Gy, and 2.2 ± 1.6 Gy. Approximately 10% (25) of the plans studied were clearly identified by the prediction models as inferior quality plans needing further optimization and refinement. CONCLUSION: Machine learning prediction models for treatment plan 3D dose distributions in online adaptive MRgRT were developed and tested. Clinical integration of the models requires minimal effort, producing 3D dose predictions for a new patient's plan using only target and OAR structures as inputs. These models can enable improved workflows for MRgRT through more informed plan optimization and plan quality assessment in real time.

Dose uncertainty and resolution of polymer gel dosimetry using an MRI guided radiation therapy system's onboard 0.35 T scanner.

Magnetic Resonance Imaging (MRI) scanners are widely used for 3D gel dosimeters readout. However, limited access to MRI scanners is a challenge in MRI-based gel dosimetry. Recent clinical implementation of MRI-guided radiation therapy machines provides potential opportunities for onboard gel dosimetry using its MRI subsystem. The objective of this study was to investigate the feasibility of gel dosimetry using ViewRay's onboard 0.35 T MRI scanner. A BANG® polymer gel dosimeter was irradiated by three beams of 3 × 3 cm2 field size. The T2 relaxation rate (R2) of the irradiated gel was measured using a Philips 1.5 T Ingenia MRI and a ViewRay 0.35 T onboard MRI and spin-echo pulse sequences. The number of signal averages (NSA) was set to 16 for the ViewRay acquisitions and one for the Philips 1.5 T MRI to achieve similar signal-to-noise ratios. The in-plane spatial resolution was 1.5 × 1.5 mm2 and the slice thickness was 5 mm. The relative dose uncertainty was obtained using R2 versus dose curves to compare the performance of dosimetry using the two different MRIs and field strengths. The dose uncertainty decreased from 12% at 2 Gy to 3.5% at 7.5 Gy at 1.5 T. The dose uncertainty decreased from 13% at 2 Gy to 4% at 7.5 Gy with NSA = 16 and 3 × 3 mm2 pixel size, and from 10.5% at 2 Gy to 3.2% at 7.5 Gy with NSA = 16 and denoised R2 maps (1.5 × 1.5 mm2 pixel size) at 0.35 T. The mean of dose resolution was 0.4 Gy at 1.5 T while the mean of dose resolution was 0.8 Gy and 0.64 Gy at 0.35 T by downsampling and denoising the R2 map, respectively. Therefore, comparable dose uncertainty was achievable using the ViewRay's onboard 0.35 T and Philips 1.5 T MRI scanners. 3D gel dosimetry using onboard low-field MRI scanner provides ViewRay users a 3D high resolution dosimetry option besides film and ionization chamber.

Using prediction models to evaluate magnetic resonance image guided radiation therapy plans.

Comprehensive analysis of daily, online adaptive plan quality and safety in magnetic resonance imaging (MRI) guided radiation therapy is critical to its widespread use. Artificial neural network models developed with offline plans created after simulation were used to analyze and compare online plans that were adapted and reoptimized in real time prior to treatment. Roughly one third of 60Co adapted plans were of inferior quality relative to fully optimized, offline plans, but MRI-linac adapted plans were essentially equivalent to offline plans. The models also enabled clear justification that MRI-linac plans are superior to 60Co in an overwhelming majority of cases.