Jinkoo Kim

Analysis of deformable image registration accuracy using computational modeling

Computer aided modeling of anatomic deformation, allowing various techniques and protocols in radiation therapy to be systematically verified and studied, has become increasingly attractive. In this study the potential issues in deformable image registration (DIR) were analyzed based on two numerical phantoms: One, a synthesized, low intensity gradient prostate image, and the other a lung patients CT image data set. Each phantom was modeled with region-specific material parameters with its deformation solved using a finite element method. The resultant displacements were used to construct a benchmark to quantify the displacement errors of the Demons and B-Spline-based registrations. The results show that the accuracy of these registration algorithms depends on the chosen parameters, the selection of which is closely associated with the intensity gradients of the underlying images. For the Demons algorithm, both single resolution (SR) and multiresolution (MR) registrations required approximately 300 iterations to reach an accuracy of 1.4 mm mean error in the lung patients CT image (and 0.7 mm mean error averaged in the lung only). For the low gradient prostate phantom, these algorithms (both SR and MR) required at least 1600 iterations to reduce their mean errors to 2 mm. For the B-Spline algorithms, best performance (mean errors of 1.9 mm for SR and 1.6 mm for MR, respectively) on the low gradient prostate was achieved using five grid nodes in each direction. Adding more grid nodes resulted in larger errors. For the lung patients CT data set, the B-Spline registrations required ten grid nodes in each direction for highest accuracy (1.4 mm for SR and 1.5 mm for MR). The numbers of iterations or grid nodes required for optimal registrations depended on the intensity gradients of the underlying images. In summary, the performance of the Demons and B-Spline registrations have been quantitatively evaluated using numerical phantoms. The results show that parameter selection for optimal accuracy is closely related to the intensity gradients of the underlying images. Also, the result that the DIR algorithms produce much lower errors in heterogeneous lung regions relative to homogeneous (low intensity gradient) regions, suggests that feature-based evaluation of deformable image registration accuracy must be viewed cautiously.

Combining scatter reduction and correction to improve image quality in cone‐beam computed tomography (CBCT)

PURPOSE The authors propose a combined scatter reduction and correction method to improve image quality in cone-beam computed tomography (CBCT). Although using a beam-block approach similar to previous techniques to measure the scatter, this method differs in that the authors utilize partially blocked projection data obtained during scatter measurement for CBCT image reconstruction. This study aims to evaluate the feasibility of the proposed approach. METHODS A 1D grid, composed of lead septa, was placed between the radiation source and the imaging object for scatter measurement. Image data were collected from the grid interspace regions while the scatter distribution was measured in the blocked regions under the grid. Scatter correction was performed by subtracting the measured scatter from the imaging data. Image information in the penumbral regions of the grid was derived. Three imaging modes were developed to reconstruct full CBCT images from partial projection data. The single-rotation half-fan mode uses interpolation to fill the missing data. The dual-rotation half-fan mode uses two rotations, with the grid offset by half a grid cycle, to acquire two complementary sets of projections, which are then merged to form complete projections for reconstruction. The single-rotation full-fan mode was designed for imaging a small object or a region of interest. Full-fan projection images were acquired over a 360 degrees scan angle with the grid shifting a distance during the scan. An enlarged Catphan phantom was used to evaluate potential improvement in image quality with the proposed technique. An anthropomorphic pelvis phantom was used to validate the feasibility of reconstructing a complete set of CBCT images from the partially blocked projections using three imaging modes. Rigid-body image registration was performed between the CBCT images from the single-rotation half-fan mode and the simulation CT and the results were compared to that for the CBCT images from dual-rotation mode and conventional CBCT images. RESULTS The proposed technique reduced the streak artifact index from 58% to 1% in comparison with the conventional CBCT. It also improved CT number linearity from 0.880 to 0.998 and the contrast-to-noise ratio (CNR) from 4.29 to 6.42. Complete sets of CBCT images with overall improved image quality were achieved for all three image modes. The longitudinal resolution was slightly compromised for the single-rotation half-fan mode. High resolution was retained for the dual-rotation half-fan and single-rotation full-fan modes in the longitudinal direction. The registration error for the CBCT images from the single-rotation half-fan mode was 0.8 +/- 0.3 mm in the longitudinal direction and negligible in the other directions. CONCLUSIONS The proposed method provides combined scatter correction and direct scatter reduction. Scatter correction may eliminate scatter artifacts, while direct scatter reduction may improve the CNR to compensate the CNR degradation due to scatter correction. Complete sets of CBCT images are reconstructed in all three imaging modes. The single-rotation mode can be used for rigid-body patient alignment despite degradation in longitudinal resolution. The dual-rotation mode may be used to improve CBCT image quality for soft tissue delineation in adaptive radiation therapy.

Image-Guided Localization Accuracy of Stereoscopic Planar and Volumetric Imaging Methods for Stereotactic Radiation Surgery and Stereotactic Body Radiation Therapy: A Phantom Study

PURPOSE To evaluate the positioning accuracies of two image-guided localization systems, ExacTrac and On-Board Imager (OBI), in a stereotactic treatment unit. METHODS AND MATERIALS An anthropomorphic pelvis phantom with eight internal metal markers (BBs) was used. The center of one BB was set as plan isocenter. The phantom was set up on a treatment table with various initial setup errors. Then, the errors were corrected using each of the investigated systems. The residual errors were measured with respect to the radiation isocenter using orthogonal portal images with field size 3 × 3 cm(2). The angular localization discrepancies of the two systems and the correction accuracy of the robotic couch were also studied. A pair of pre- and post-cone beam computed tomography (CBCT) images was acquired for each angular correction. Then, the correction errors were estimated by using the internal BBs through fiducial marker-based registrations. RESULTS The isocenter localization errors (μ ±σ) in the left/right, posterior/anterior, and superior/inferior directions were, respectively, -0.2 ± 0.2 mm, -0.8 ± 0.2 mm, and -0.8 ± 0.4 mm for ExacTrac, and 0.5 ± 0.7 mm, 0.6 ± 0.5 mm, and 0.0 ± 0.5 mm for OBI CBCT. The registration angular discrepancy was 0.1 ± 0.2° between the two systems, and the maximum angle correction error of the robotic couch was 0.2° about all axes. CONCLUSION Both the ExacTrac and the OBI CBCT systems showed approximately 1 mm isocenter localization accuracies. The angular discrepancy of two systems was minimal, and the robotic couch angle correction was accurate. These positioning uncertainties should be taken as a lower bound because the results were based on a rigid dosimetry phantom.

Examining Margin Reduction and Its Impact on Dose Distribution for Prostate Cancer Patients Undergoing Daily Cone-Beam Computed Tomography

PURPOSE To examine the dosimetric impact of margin reduction and quantify residual error after three-dimensional (3D) image registration using daily cone-beam computed tomography (CBCT) for prostate cancer patients. METHODS AND MATERIALS One hundred forty CBCTs from 5 prostate cancer patients were examined. Two intensity-modulated radiotherapy plans were generated on CT simulation on the basis of two planning target volume (PTV) margins: 10 mm all around the prostate and seminal vesicles except 6 mm posteriorly (10/6) and 5 mm all around except 3 mm posteriorly (5/3). Daily CBCT using the Varian On-Board Imaging System was acquired. The 10/6 and 5/3 simulation plans were overlaid onto each CBCT, and each CBCT plan was calculated. To examine residual error, PlanCT/CBCT intensity-based 3D image registration was performed for prostate localization using center of mass and maximal border displacement. RESULTS Prostate coverage was within 2% between the 10/6 and 5/3 plans. Seminal vesicle coverage was reduced with the 5/3 plan compared with the 10/6 plan, with coverage difference within 7%. The 5/3 plan allowed 30-50% sparing of bladder and rectal high-dose regions. For residual error quantification, center of mass data show that 99%, 93%, and 96% of observations fall within 3 mm in the left-right, anterior-posterior, and superior-inferior directions, respectively. Maximal border displacement observations range from 79% to 99%, within 5 mm for all directions. CONCLUSION Cone-beam CT dosimetrically validated a 10/6 margin when soft-tissue localization is not used. Intensity-based 3D image registration has the potential to improve target localization and to provide guidelines for margin definition.

Using patient-specific phantoms to evaluate deformable image registration algorithms for adaptive radiation therapy

The quality of adaptive treatment planning depends on the accuracy of its underlying deformable image registration (DIR). The purpose of this study is to evaluate the performance of two DIR algorithms, B‐spline‐based deformable multipass (DMP) and deformable demons (Demons), implemented in a commercial software package. Evaluations were conducted using both computational and physical deformable phantoms. Based on a finite element method (FEM), a total of 11 computational models were developed from a set of CT images acquired from four lung and one prostate cancer patients. FEM generated displacement vector fields (DVF) were used to construct the lung and prostate image phantoms. Based on a fast‐Fourier transform technique, image noise power spectrum was incorporated into the prostate image phantoms to create simulated CBCT images. The FEM‐DVF served as a gold standard for verification of the two registration algorithms performed on these phantoms. The registration algorithms were also evaluated at the homologous points quantified in the CT images of a physical lung phantom. The results indicated that the mean errors of the DMP algorithm were in the range of 1.0~3.1mm for the computational phantoms and 1.9 mm for the physical lung phantom. For the computational prostate phantoms, the corresponding mean error was 1.0–1.9 mm in the prostate, 1.9–2.4 mm in the rectum, and 1.8–2.1 mm over the entire patient body. Sinusoidal errors induced by B‐spline interpolations were observed in all the displacement profiles of the DMP registrations. Regions of large displacements were observed to have more registration errors. Patient‐specific FEM models have been developed to evaluate the DIR algorithms implemented in the commercial software package. It has been found that the accuracy of these algorithms is patient‐dependent and related to various factors including tissue deformation magnitudes and image intensity gradients across the regions of interest. This may suggest that DIR algorithms need to be verified for each registration instance when implementing adaptive radiation therapy. PACS numbers: 87.10.Kn, 87.55.km, 87.55.Qr, 87.57.nj

Radiosurgery of multiple brain metastases with single-isocenter dynamic conformal arcs (SIDCA)

PURPOSE To propose single-isocenter dynamic conformal arcs (SIDCA), a novel technique for radiosurgery of multiple brain metastases, and to compare SIDCA with volumetric modulated arc therapy (VMAT) and multiple-isocenter dynamic conformal arcs (MIDCA) for plan quality. METHODS AND MATERIALS SIDCA, MIDCA, and VMAT plans were created on 6 patients with 3-5 metastases. Plans were evaluated using Radiation Therapy Oncology Group conformity index (RCI), Paddick conformity index (PCI), gradient index (GI), volumes that received more than 100% (V(100%)), 50% (V(50%)), 25% (V(25%)) and 10% (V(10%)) of prescription dose, total monitor units (MUs), and delivery time (DT). RESULTS SIDCA achieved conformal plans (RCI = 1.38 ± 0.12, PCI = 0.72 ± 0.06) with steep dose fall-off (GI = 3.97 ± 0.51). MIDCA plans had comparable plan quality and MUs as SIDCA, but 52% longer DT. The VMAT plans had better conformity (RCI = 1.15 ± 0.09, p < 0.01 and PCI = 0.86 ± 0.06, p < 0.01) than SIDCA, worse GI (4.34 ± 0.46, p < 0.01), higher V(25%) (p = 0.05) and V(10%) (p = 0.02), 49% less MUs and 46% shorter DT. CONCLUSIONS All three techniques achieved conformal plans with steep dose fall-off from targets. SIDCA plans had similar plan quality as MIDCA but more efficient to delivery. SIDCA plans had lower peripheral dose spread than VMAT; VMAT plans had better conformity and faster delivery time than SIDCA.

Clinical commissioning and use of the Novalis Tx linear accelerator for SRS and SBRT

The purpose of this study was to perform comprehensive measurements and testing of a Novalis Tx linear accelerator, and to develop technical guidelines for commissioning from the time of acceptance testing to the first clinical treatment. The Novalis Tx (NTX) linear accelerator is equipped with, among other features, a high‐definition MLC (HD120 MLC) with 2.5 mm central leaves, a 6D robotic couch, an optical guidance positioning system, as well as X‐ray‐based image guidance tools to provide high accuracy radiation delivery for stereotactic radiosurgery and stereotactic body radiation therapy procedures. We have performed extensive tests for each of the components, and analyzed the clinical data collected in our clinic. We present technical guidelines in this report focusing on methods for: (1) efficient and accurate beam data collection for commissioning treatment planning systems, including small field output measurements conducted using a wide range of detectors; (2) commissioning tests for the HD120 MLC; (3) data collection for the baseline characteristics of the on‐board imager (OBI) and ExacTrac X‐ray (ETX) image guidance systems in conjunction with the 6D robotic couch; and (4) end‐to‐end testing of the entire clinical process. Established from our clinical experience thus far, recommendations are provided for accurate and efficient use of the OBI and ETX localization systems for intra‐ and extracranial treatment sites. Four results are presented. (1) Basic beam data measurements: Our measurements confirmed the necessity of using small detectors for small fields. Total scatter factors varied significantly (30% to approximately 62%) for small field measurements among detectors. Unshielded stereotactic field diode (SFD) overestimated dose by ~ 2% for large field sizes. Ion chambers with active diameters of 6 mm suffered from significant volume averaging. The sharpest profile penumbra was observed for the SFD because of its small active diameter (0.6 mm). (2) MLC commissioning: Winston Lutz test, light/radiation field congruence, and Picket Fence tests were performed and were within criteria established by the relevant task group reports. The measured mean MLC transmission and dynamic leaf gap of 6 MV SRS beam were 1.17% and 0.36 mm, respectively. (3) Baseline characteristics of OBI and ETX: The isocenter localization errors in the left/right, posterior/anterior, and superior/inferior directions were, respectively, −0.2±0.2 mm, −0.8±0.2 mm, and −0.8±0.4 mm for ETX, and 0.5±0.7 mm, 0.6±0.5 mm, and 0.0±0.5 mm for OBI cone‐beam computed tomography. The registration angular discrepancy was 0.1±0.2°, and the maximum robotic couch error was 0.2°. (4) End‐to‐end tests: The measured isocenter dose differences from the planned values were 0.8% and 0.4%, measured respectively by an ion chamber and film. The gamma pass rate, measured by EBT2 film, was 95% (3% DD and 1 mm DTA). Through a systematic series of quantitative commissioning experiments and end‐to‐end tests and our initial clinical experience, described in this report, we demonstrate that the NTX is a robust system, with the image guidance and MLC requirements to treat a wide variety of sites — in particular for highly accurate delivery of SRS and SBRT‐based treatments. PACS numbers: 87.55.Qr, 87.53.Ly, 87.59.‐e

Effects of x-ray and CT image enhancements on the robustness and accuracy of a rigid 3D/2D image registration.

A rigid body three-dimensional/two-dimensional (3D/2D) registration method has been implemented using mutual information, gradient ascent, and 3D texturemap-based digitally reconstructed radiographs. Nine combinations of commonly used x-ray and computed tomography (CT) image enhancement methods, including window leveling, histogram equalization, and adaptive histogram equalization, were examined to assess their effects on accuracy and robustness of the registration method. From a set of experiments using an anthropomorphic chest phantom, we were able to draw several conclusions. First, the CT and x-ray preprocessing combination with the widest attraction range was the one that linearly stretched the histograms onto the entire display range on both CT and x-ray images. The average attraction ranges of this combination were 71.3 mm and 61.3 deg in the translation and rotation dimensions, respectively, and the average errors were 0.12 deg and 0.47 mm. Second, the combination of the CT image with tissue and bone information and the x-ray images with adaptive histogram equalization also showed subvoxel accuracy, especially the best in the translation dimensions. However, its attraction ranges were the smallest among the examined combinations (on average 36 mm and 19 deg). Last the bone-only information on the CT image did not show convergency property to the correct registration.

Direct dose mapping versus energy/mass transfer mapping for 4D dose accumulation: fundamental differences and dosimetric consequences

The direct dose mapping (DDM) and energy/mass transfer (EMT) mapping are two essential algorithms for accumulating the dose from different anatomic phases to the reference phase when there is organ motion or tumor/tissue deformation during the delivery of radiation therapy. DDM is based on interpolation of the dose values from one dose grid to another and thus lacks rigor in defining the dose when there are multiple dose values mapped to one dose voxel in the reference phase due to tissue/tumor deformation. On the other hand, EMT counts the total energy and mass transferred to each voxel in the reference phase and calculates the dose by dividing the energy by mass. Therefore it is based on fundamentally sound physics principles. In this study, we implemented the two algorithms and integrated them within the Eclipse treatment planning system. We then compared the clinical dosimetric difference between the two algorithms for ten lung cancer patients receiving stereotactic radiosurgery treatment, by accumulating the delivered dose to the end-of-exhale (EE) phase. Specifically, the respiratory period was divided into ten phases and the dose to each phase was calculated and mapped to the EE phase and then accumulated. The displacement vector field generated by Demons-based registration of the source and reference images was used to transfer the dose and energy. The DDM and EMT algorithms produced noticeably different cumulative dose in the regions with sharp mass density variations and/or high dose gradients. For the planning target volume (PTV) and internal target volume (ITV) minimum dose, the difference was up to 11% and 4% respectively. This suggests that DDM might not be adequate for obtaining an accurate dose distribution of the cumulative plan, instead, EMT should be considered.

Comparison of similarity measures for rigid-body CT/dual X-ray image registrations

A set of experiments were conducted to evaluate six similarity measures for intensity-based rigid-body 3D/2D image registration. Similarity measure is an index that measures the similarity between a digitally reconstructed radiograph (DRR) and an x-ray planar image. The registration is accomplished by maximizing the sum of the similarity measures between biplane x-ray images and the corresponding DRRs in an iterative fashion. We have evaluated the accuracy and attraction ranges of the registrations using six different similarity measures on phantom experiments for head, thorax, and pelvis. The images were acquired using Varian Medial System On-Board Imager. Our results indicated that normalized cross correlation and entropy of difference showed a wide attraction range (62 deg and 83 mm mean attraction range, ωmean), but the worst accuracy (4.2 mm maximum error, emax). The gradient-based similarity measures, gradient correlation and gradient difference, and the pattern intensity showed sub-millimeter accuracy, but narrow attraction ranges (ωmean=29 deg, 31 mm). Mutual information was in-between of these two groups (emax=2.5 mm, ωmean= 48 deg, 52 mm). On the data of 120 x-ray pairs from eight IRB approved prostate patients, the gradient difference showed the best accuracy. In the clinical applications, registrations starting with the mutual information followed by the gradient difference may provide the best accuracy and the most robustness.