John Neylon

A GPU based high‐resolution multilevel biomechanical head and neck model for validating deformable image registration

PURPOSE Validating the usage of deformable image registration (dir) for daily patient positioning is critical for adaptive radiotherapy (RT) applications pertaining to head and neck (HN) radiotherapy. The authors present a methodology for generating biomechanically realistic ground-truth data for validating dir algorithms for HN anatomy by (a) developing a high-resolution deformable biomechanical HN model from a planning CT, (b) simulating deformations for a range of interfraction posture changes and physiological regression, and (c) generating subsequent CT images representing the deformed anatomy. METHODS The biomechanical model was developed using HN kVCT datasets and the corresponding structure contours. The voxels inside a given 3D contour boundary were clustered using a graphics processing unit (GPU) based algorithm that accounted for inconsistencies and gaps in the boundary to form a volumetric structure. While the bony anatomy was modeled as rigid body, the muscle and soft tissue structures were modeled as mass-spring-damper models with elastic material properties that corresponded to the underlying contoured anatomies. Within a given muscle structure, the voxels were classified using a uniform grid and a normalized mass was assigned to each voxel based on its Hounsfield number. The soft tissue deformation for a given skeletal actuation was performed using an implicit Euler integration with each iteration split into two substeps: one for the muscle structures and the other for the remaining soft tissues. Posture changes were simulated by articulating the skeletal structure and enabling the soft structures to deform accordingly. Physiological changes representing tumor regression were simulated by reducing the target volume and enabling the surrounding soft structures to deform accordingly. Finally, the authors also discuss a new approach to generate kVCT images representing the deformed anatomy that accounts for gaps and antialiasing artifacts that may be caused by the biomechanical deformation process. Accuracy and stability of the model response were validated using ground-truth simulations representing soft tissue behavior under local and global deformations. Numerical accuracy of the HN deformations was analyzed by applying nonrigid skeletal transformations acquired from interfraction kVCT images to the models skeletal structures and comparing the subsequent soft tissue deformations of the model with the clinical anatomy. RESULTS The GPU based framework enabled the model deformation to be performed at 60 frames/s, facilitating simulations of posture changes and physiological regressions at interactive speeds. The soft tissue response was accurate with a R(2) value of >0.98 when compared to ground-truth global and local force deformation analysis. The deformation of the HN anatomy by the model agreed with the clinically observed deformations with an average correlation coefficient of 0.956. For a clinically relevant range of posture and physiological changes, the model deformations stabilized with an uncertainty of less than 0.01 mm. CONCLUSIONS Documenting dose delivery for HN radiotherapy is essential accounting for posture and physiological changes. The biomechanical model discussed in this paper was able to deform in real-time, allowing interactive simulations and visualization of such changes. The model would allow patient specific validations of the dir method and has the potential to be a significant aid in adaptive radiotherapy techniques.

Near Real-Time Assessment of Anatomic and Dosimetric Variations for Head and Neck Radiation Therapy via Graphics Processing Unit-based Dose Deformation Framework.

PURPOSE The purpose of this study was to systematically monitor anatomic variations and their dosimetric consequences during intensity modulated radiation therapy (IMRT) for head and neck (H&N) cancer by using a graphics processing unit (GPU)-based deformable image registration (DIR) framework. METHODS AND MATERIALS Eleven IMRT H&N patients undergoing IMRT with daily megavoltage computed tomography (CT) and weekly kilovoltage CT (kVCT) scans were included in this analysis. Pretreatment kVCTs were automatically registered with their corresponding planning CTs through a GPU-based DIR framework. The deformation of each contoured structure in the H&N region was computed to account for nonrigid change in the patient setup. The Jacobian determinant of the planning target volumes and the surrounding critical structures were used to quantify anatomical volume changes. The actual delivered dose was calculated accounting for the organ deformation. The dose distribution uncertainties due to registration errors were estimated using a landmark-based gamma evaluation. RESULTS Dramatic interfractional anatomic changes were observed. During the treatment course of 6 to 7 weeks, the parotid gland volumes changed up to 34.7%, and the center-of-mass displacement of the 2 parotid glands varied in the range of 0.9 to 8.8 mm. For the primary treatment volume, the cumulative minimum and mean and equivalent uniform doses assessed by the weekly kVCTs were lower than the planned doses by up to 14.9% (P=.14), 2% (P=.39), and 7.3% (P=.05), respectively. The cumulative mean doses were significantly higher than the planned dose for the left parotid (P=.03) and right parotid glands (P=.006). The computation including DIR and dose accumulation was ultrafast (∼45 seconds) with registration accuracy at the subvoxel level. CONCLUSIONS A systematic analysis of anatomic variations in the H&N region and their dosimetric consequences is critical in improving treatment efficacy. Nearly real-time assessment of anatomic and dosimetric variations is feasible using the GPU-based DIR framework. Clinical implementation of this technology may enable timely plan adaptation and improved outcome.

A nonvoxel-based dose convolution/superposition algorithm optimized for scalable GPU architectures

PURPOSE Real-time adaptive planning and treatment has been infeasible due in part to its high computational complexity. There have been many recent efforts to utilize graphics processing units (GPUs) to accelerate the computational performance and dose accuracy in radiation therapy. Data structure and memory access patterns are the key GPU factors that determine the computational performance and accuracy. In this paper, the authors present a nonvoxel-based (NVB) approach to maximize computational and memory access efficiency and throughput on the GPU. METHODS The proposed algorithm employs a ray-tracing mechanism to restructure the 3D data sets computed from the CT anatomy into a nonvoxel-based framework. In a process that takes only a few milliseconds of computing time, the algorithm restructured the data sets by ray-tracing through precalculated CT volumes to realign the coordinate system along the convolution direction, as defined by zenithal and azimuthal angles. During the ray-tracing step, the data were resampled according to radial sampling and parallel ray-spacing parameters making the algorithm independent of the original CT resolution. The nonvoxel-based algorithm presented in this paper also demonstrated a trade-off in computational performance and dose accuracy for different coordinate system configurations. In order to find the best balance between the computed speedup and the accuracy, the authors employed an exhaustive parameter search on all sampling parameters that defined the coordinate system configuration: zenithal, azimuthal, and radial sampling of the convolution algorithm, as well as the parallel ray spacing during ray tracing. The angular sampling parameters were varied between 4 and 48 discrete angles, while both radial sampling and parallel ray spacing were varied from 0.5 to 10 mm. The gamma distribution analysis method (γ) was used to compare the dose distributions using 2% and 2 mm dose difference and distance-to-agreement criteria, respectively. Accuracy was investigated using three distinct phantoms with varied geometries and heterogeneities and on a series of 14 segmented lung CT data sets. Performance gains were calculated using three 256 mm cube homogenous water phantoms, with isotropic voxel dimensions of 1, 2, and 4 mm. RESULTS The nonvoxel-based GPU algorithm was independent of the data size and provided significant computational gains over the CPU algorithm for large CT data sizes. The parameter search analysis also showed that the ray combination of 8 zenithal and 8 azimuthal angles along with 1 mm radial sampling and 2 mm parallel ray spacing maintained dose accuracy with greater than 99% of voxels passing the γ test. Combining the acceleration obtained from GPU parallelization with the sampling optimization, the authors achieved a total performance improvement factor of >175 000 when compared to our voxel-based ground truth CPU benchmark and a factor of 20 compared with a voxel-based GPU dose convolution method. CONCLUSIONS The nonvoxel-based convolution method yielded substantial performance improvements over a generic GPU implementation, while maintaining accuracy as compared to a CPU computed ground truth dose distribution. Such an algorithm can be a key contribution toward developing tools for adaptive radiation therapy systems.

Distortion-free diffusion MRI using an MRI-guided Tri-Cobalt 60 radiotherapy system: Sequence verification and preliminary clinical experience

Purpose: Monitoring tumor response during the course of treatment and adaptively modifying treatment plan based on tumor biological feedback may represent a new paradigm for radiotherapy. Diffusion MRI has shown great promises in assessing and predicting tumor response to radiotherapy. However, the conventional diffusion‐weighted single‐shot echo‐planar‐imaging (DW‐ssEPI) technique suffers from limited resolution, severe distortion, and possibly inaccurate ADC at low field strength. The purpose of this work was to develop a reliable, accurate and distortion‐free diffusion MRI technique that is practicable for longitudinal tumor response evaluation and adaptive radiotherapy on a 0.35 T MRI‐guided radiotherapy system. Methods: A diffusion‐prepared turbo spin echo readout (DP‐TSE) sequence was developed and compared with the conventional diffusion‐weighted single‐shot echo‐planar‐imaging sequence on a 0.35 T MRI‐guided radiotherapy system (ViewRay). A spatial integrity phantom was used to quantitate and compare the geometric accuracy of the two diffusion sequences for three orthogonal orientations. The apparent diffusion coefficient (ADC) accuracy was evaluated on a diffusion phantom under both 0 °C and room temperature to cover a diffusivity range between 0.40 × 10−3 and 2.10 × 10−3 mm2/s. Ten room temperature measurements repeated on five different days were conducted to assess the ADC reproducibility of DP‐TSE. Two glioblastoma (GBM) and six sarcoma patients were included to examine the in vivo feasibility. The target registration error (TRE) was calculated to quantitate the geometric accuracy where structural CT or MR images were co‐registered to the diffusion images as references. ADC maps from DP‐TSE and DW‐ssEPI were calculated and compared. A tube phantom was placed next to patients not treated on ViewRay, and ADCs of this reference tube were also compared. Results: The proposed DP‐TSE passed the spatial integrity test (< 1 mm within 100 mm radius and < 2 mm within 175 mm radius) under the three orthogonal orientations. The detected errors were 0.474 ± 0.355 mm, 0.475 ± 0.287 mm, and 0.546 ± 0.336 mm in the axial, coronal, and sagittal plane. DW‐ssEPI, however, failed the tests due to severe distortion and low signal intensity. Noise correction must be performed for the DW‐ssEPI to avoid ADC quantitation errors, whereas it is optional for DP‐TSE. At 0 °C, the two sequences provided accurate quantitation with < 3% variation with the reference. In the room temperature study, discrepancies between ADCs from DP‐TSE and the reference were within 4%, but could be as high as 8% for DW‐ssEPI after the noise correction. Excellent ADC reproducibility with a coefficient of variation < 5% was observed among the 10 measurements of DP‐TSE, indicating desirable robustness for ADC‐based tumor response assessment. In vivo TRE in DP‐TSE was less than 1.6 mm overall, whereas it could be greater than 12 mm in DW‐ssEPI. For GBM patients, the CSF and brain tissue ADCs from DP‐TSE were within the ranges found in literature. ADC differences between the two techniques were within 8% among the six sarcoma patients. For the reference tube that had a relatively low diffusivity, the two diffusion sequences provided matched measurements. Conclusion: A diffusion technique with excellent geometric fidelity, accurate, and reproducible ADC measurement was demonstrated for longitudinal tumor response assessment using a low‐field MRI‐guided radiotherapy system.

Analytic Intermodel Consistent Modeling of Volumetric Human Lung Dynamics

Human lung undergoes breathing-induced deformation in the form of inhalation and exhalation. Modeling the dynamics is numerically complicated by the lack of information on lung elastic behavior and fluid-structure interactions between air and the tissue. A mathematical method is developed to integrate deformation results from a deformable image registration (DIR) and physics-based modeling approaches in order to represent consistent volumetric lung dynamics. The computational fluid dynamics (CFD) simulation assumes the lung is a poro-elastic medium with spatially distributed elastic property. Simulation is performed on a 3D lung geometry reconstructed from four-dimensional computed tomography (4DCT) dataset of a human subject. The heterogeneous Youngs modulus (YM) is estimated from a linear elastic deformation model with the same lung geometry and 4D lung DIR. The deformation obtained from the CFD is then coupled with the displacement obtained from the 4D lung DIR by means of the Tikhonov regularization (TR) algorithm. The numerical results include 4DCT registration, CFD, and optimal displacement data which collectively provide consistent estimate of the volumetric lung dynamics. The fusion method is validated by comparing the optimal displacement with the results obtained from the 4DCT registration.

A neural network approach for fast, automated quantification of DIR performance

Purpose: A critical step in adaptive radiotherapy (ART) workflow is deformably registering the simulation CT with the daily or weekly volumetric imaging. Quantifying the deformable image registration accuracy under these circumstances is a complex task due to the lack of known ground‐truth landmark correspondences between the source data and target data. Generating landmarks manually (using experts) is time‐consuming, and limited by image quality and observer variability. While image similarity metrics (ISM) may be used as an alternative approach to quantify the registration error, there is a need to characterize the ISM values by developing a nonlinear cost function and translate them to physical distance measures in order to enable fast, quantitative comparison of registration performance. Methods: In this paper, we present a proof‐of‐concept methodology for automated quantification of DIR performance. A nonlinear cost function was developed as a combination of ISM values and governed by the following two expectations for an accurate registration: (a) the deformed data obtained from transforming the simulation CT data with the deformation vector field (DVF) should match the target image data with near perfect similarity, and (b) the similarity between the simulation CT and deformed data should match the similarity between the simulation CT and the target image data. A deep neural network (DNN) was developed that translated the cost function values to actual physical distance measure. To train the neural network, patient‐specific biomechanical models of the head‐and‐neck anatomy were employed. The biomechanical model anatomy was systematically deformed to represent changes in patient posture and physiological regression. Volumetric source and target images with known ground‐truth deformations vector fields were then generated, representing the daily or weekly imaging data. Annotated data was then fed through a supervised machine learning process, iteratively optimizing a nonlinear model able to predict the target registration error (TRE) for given ISM values. The cost function for sub‐volumes enclosing critical radiotherapy structures in the head‐and‐neck region were computed and compared with the ground truth TRE values. Results: When examining different combinations of registration parameters for a single DIR, the neural network was able to quantify DIR error to within a single voxel for 95% of the sub‐volumes examined. In addition, correlations between the neural network predicted error and the ground‐truth TRE for the Planning Target Volume and the parotid contours were consistently observed to be > 0.9. For variations in posture and tumor regression for 10 different patients, patient‐specific neural networks predicted the TRE to within a single voxel > 90% on average. Conclusions: The formulation presented in this paper demonstrates the ability for fast, accurate quantification of registration performance. DNN provided the necessary level of abstraction to estimate a quantified TRE from the ISM expectations described above, when sufficiently trained on annotated data. In addition, biomechanical models facilitated the DNN with the required variations in the patient posture and physiological regression. With further development and validation on clinical patient data, such networks have potential impact in patient and site‐specific optimization, and stream‐lining clinical registration validation.

Feasibility of deriving a novel imaging biomarker based on patient-specific lung elasticity for characterizing the degree of COPD in lung SBRT patients

OBJECTIVE: Lung tissue elasticity is an effective spatial representation for Chronic Obstructive Pulmonary Disease phenotypes and pathophysiology. We investigated a novel imaging biomarker based on the voxel-by-voxel distribution of lung tissue elasticity. Our approach combines imaging and biomechanical modeling to characterize tissue elasticity. METHODS: We acquired 4DCT images for 13 lung cancer patients with known COPD diagnoses based on GOLD 2017 criteria. Deformation vector fields (DVFs) from the deformable registration of end-inhalation and end-exhalation breathing phases were taken to be the ground-truth. A linear elastic biomechanical model was assembled from end-exhalation datasets with a density-guided initial elasticity distribution. The elasticity estimation was formulated as an iterative process, where the elasticity was optimized based on its ability to reconstruct the ground-truth. An imaging biomarker (denoted YM1-3) derived from the optimized elasticity distribution, was compared with the current gold standard, RA950 using confusion matrix and area under the receiver operating characteristic (AUROC) curve analysis. RESULTS: The estimated elasticity had 90 % accuracy when representing the ground-truth DVFs. The YM1-3 biomarker had higher diagnostic accuracy (86% vs 71 %), higher sensitivity (0.875 vs 0.5), and a higher AUROC curve (0.917 vs 0.875) as compared to RA950. Along with acting as an effective spatial indicator of lung pathophysiology, the YM1-3 biomarker also proved to be a better indicator for diagnostic purposes than RA950. CONCLUSIONS: Overall, the results suggest that, as a biomarker, lung tissue elasticity will lead to new end points for clinical trials and new targeted treatment for COPD subgroups. ADVANCES IN KNOWLEDGE: The derivation of elasticity information directly from 4DCT imaging data is a novel method for performing lung elastography. The work demonstrates the need for a mechanics-based biomarker for representing lung pathophysiology.

Fast simulated annealing and adaptive Monte Carlo sampling based parameter optimization for dense optical-flow deformable image registration of 4DCT lung anatomy

Deformable image registration (DIR) is an important step in radiotherapy treatment planning. An optimal input registration parameter set is critical to achieve the best registration performance with the specific algorithm. Methods In this paper, we investigated a parameter optimization strategy for Optical-flow based DIR of the 4DCT lung anatomy. A novel fast simulated annealing with adaptive Monte Carlo sampling algorithm (FSA-AMC) was investigated for solving the complex non-convex parameter optimization problem. The metric for registration error for a given parameter set was computed using landmark-based mean target registration error (mTRE) between a given volumetric image pair. To reduce the computational time in the parameter optimization process, a GPU based 3D dense optical-flow algorithm was employed for registering the lung volumes. Numerical analyses on the parameter optimization for the DIR were performed using 4DCT datasets generated with breathing motion models and open-source 4DCT datasets. Results showed that the proposed method efficiently estimated the optimum parameters for optical-flow and closely matched the best registration parameters obtained using an exhaustive parameter search method.

WE-AB-202-09: Feasibility and Quantitative Analysis of 4DCT-Based High Precision Lung Elastography

PURPOSE The purpose of this project is to derive high precision elastography measurements from 4DCT lung scans to facilitate the implementation of elastography in a radiotherapy context. METHODS 4DCT scans of the lungs were acquired, and breathing stages were subsequently registered to each other using an optical flow DIR algorithm. The displacement of each voxel gleaned from the registration was taken to be the ground-truth deformation. These vectors, along with the 4DCT source datasets, were used to generate a GPU-based biomechanical simulation that acted as a forward model to solve the inverse elasticity problem. The lung surface displacements were applied as boundary constraints for the model-guided lung tissue elastography, while the inner voxels were allowed to deform according to the linear elastic forces within the model. A biomechanically-based anisotropic convergence magnification technique was applied to the inner voxels in order to amplify the subtleties of the interior deformation. Solving the inverse elasticity problem was accomplished by modifying the tissue elasticity and iteratively deforming the biomechanical model. Convergence occurred when each voxel was within 0.5 mm of the ground-truth deformation and 1 kPa of the ground-truth elasticity distribution. To analyze the feasibility of the model-guided approach, we present the results for regions of low ventilation, specifically, the apex. RESULTS The maximum apical boundary expansion was observed to be between 2 and 6 mm. Simulating this expansion within an apical lung model, it was observed that 100% of voxels converged within 0.5 mm of ground-truth deformation, while 91.8% converged within 1 kPa of the ground-truth elasticity distribution. A mean elasticity error of 0.6 kPa illustrates the high precision of our technique. CONCLUSION By utilizing 4DCT lung data coupled with a biomechanical model, high precision lung elastography can be accurately performed, even in low ventilation regions of the lungs. This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE-1144087.

TU-H-CAMPUS-JeP1-01: Characterization of a Parameterized Image Similarity Cost Function for Automated Patient and Site Specific DIR Accuracy Optimizations

PURPOSE Clinical deformable image registration (DIR) accuracy is difficult to quantify, making validation and verification manual and time consuming. It has also been seen that DIR accuracy can be improved by patient and site specific optimization. We previously investigated a cost function to parameterize image similarity metrics (ISMs) and provide a quantifiable comparison of DIR performance. In this abstract, we characterize and optimize the cost function response (CFR) and correlation to target registration error (TRE) in a dense registration parameter space. METHODS A validated head-and-neck biomechanical model was used to induce clinical deformations in patient kVCT data (source), outputting known ground truth deformation vector fields (DVFs) and simulated CTs of deformed anatomy (target). The source and target were registered repeatedly using an in-house optical flow DIR, systematically sampling the registration parameter space, and producing TREs and data volumes deformed according to the DIR DVF (warp). Normalized mutual information (NMI) was chosen as the test ISM. The cost function variables (CFV) were also systematically sampled to characterize their effect on the CFR. Sub-volume analysis was performed on the data, defined by structures of interest in the head-and-neck region. RESULTS Strong inverse correlation was observed between the TRE and CFR when the data was subdivided around specific structures like the PTV, parotids, and mandible, reaching values of -0.95. The CFR slope changed by an order of magnitude by manipulating the CFV. Response and correlation were improved by adjusting on a structure specific basis. The CFR achieved a predominantly convex response, which should facilitate its use as an energy term in an optimization scheme. CONCLUSIONS Replacing the time consuming, manually placed landmark assessment methodology is necessary for patient and site specific registration optimization to be clinically feasible. The proposed cost function is a promising alternative for fast, fully automated DIR performance quantification.