Voxel-based statistical analysis of uncertainties associated with deformable image registration

Estimating medical image registration error and confidence: A taxonomy and scoping review

Given that image registration is a fundamental and ubiquitous task in both clinical and research domains of the medical field, errors in registration can have serious consequences. Since such errors can mislead clinicians during image-guided therapies or bias the results of a downstream analysis, methods to estimate registration error are becoming more popular. To give structure to this new heterogenous field we developed a taxonomy and performed a scoping review of methods that quantitatively and automatically provide a dense estimation of registration error. The taxonomy breaks down error estimation methods into Approach (Image- or Transformation-based), Framework (Machine Learning or Direct) and Measurement (error or confidence) components. Following the PRISMA guidelines for scoping reviews, the 570 records found were reduced to twenty studies that met inclusion criteria, which were then reviewed according to the proposed taxonomy. Trends in the field, advantages and disadvantages of the methods, and potential sources of bias are also discussed. We provide suggestions for best practices and identify areas of future research.

Validation of accuracy deformable image registration contour propagation using a benchmark virtual HN phantom dataset

Virtual anatomic phantoms offer precise voxel mapping of the variation of anatomy with ground truth deformation vector fields (DVFs). Dice similarity coefficient (DSC) and mean distance to agreement (MDA) are the standard metrics for evaluating geometric contour congruence when testing deformable registration (DIR) algorithms. A HN virtual patient phantom data set was used for a kVCT‐kVCT automatic propagation contour validation study employing the Accuray DIR algorithm. Furthermore, since TomoTherapy uses MVCT images of the relevant anatomy for adaptive monitoring, the kVCT image data set quality was transformed to an MVCT image data set quality to study intermodal kVCT‐MVCT DIR accuracy. The results of the study indicate that the Accuray DIR algorithm can be expected to autopropagate HN contours adequately, on average, within tolerances recommended by TG‐132 (DSC 0.8‐0.9, MDA within voxel width). However, contours critical to dosimetric planning should always be visually proofed for accuracy. Using standard reconstruction MVCT image quality causes slightly less, but acceptable, agreement with ground truth contours.

Adaptive Radiation Therapy (ART) Strategies and Technical Considerations: A State of the ART Review From NRG Oncology

The integration of adaptive radiation therapy (ART), or modifying the treatment plan during the treatment course, is becoming more widely available in clinical practice. ART offers strong potential for minimizing treatment-related toxicity while escalating or de-escalating target doses based on the dose to organs at risk. Yet, ART workflows add complexity into the radiation therapy planning and delivery process that may introduce additional uncertainties. This work sought to review presently available ART workflows and technological considerations such as image quality, deformable image registration, and dose accumulation. Quality assurance considerations for ART components and minimum recommendations are described. Personnel and workflow efficiency recommendations are provided, as is a summary of currently available clinical evidence supporting the implementation of ART. Finally, to guide future clinical trial protocols, an example ART physician directive and a physics template following standard NRG Oncology protocol is provided.

Technical Note: Solving the "Chinese postman problem" for effective contour deformation

PURPOSE

To develop a practical approach for accurate contour deformation when deformable image registration (DIR) is used for atlas-based segmentation or contour propagation in image-guided radiotherapy.

METHODS

We developed a contour deformation approach based on 3D mesh operations. The 2D contours represented by a series of points in each slice were first converted to a 3D triangular mesh, which was deformed by the deformation vectors resulting from DIR. A set of parallel 2D planes then cut through the deformed 3D mesh, generating unordered points and line segments, to be reorganized into a set of 2D contour points. The reorganization problem was equivalent to solving the "Chinese postman problem" (CPP) by traversing a graph built from the unordered points with the least cost. Alternatively, deformation could be applied to a binary image converted from the original contours. The deformed binary image was then converted back into contours at the CT slice locations. We validated the mesh-based contour deformation approach using lung and heart contours from 10 patients with thoracic cancer.

RESULTS

DIR could change the 3D mesh considerably, complicating 2D contour representations after deformation. CPP could effectively reorganize the points in 2D planes regardless of how complicated the 2D contours were. Among the 10 patients, the Dice similarity coefficient between the mesh-based contour and binary image-based contour was 97.6% ± 0.3% for lung and 97.5% ± 0.7% for heart, and the Hausdoroff distance between them was 19.8 ± 5.1 mm for lung and 6.1 ± 2.2 mm for heart. Subjective evaluation showed that the mesh-based approach could keep fine details, especially for the lung. The image-based approach seemed to overprocess contours and suffered from image resolution limits.

CONCLUSION

We developed a practical approach for accurate contour deformation and demonstrated its effectiveness for both clinical and research applications.

Deformable image registration applied to lung SBRT: Usefulness and limitations

Radiation therapy (RT) of the lung requires deformation analysis. Deformable image registration (DIR) is the fundamental method to quantify deformations for various applications: motion compensation, contour propagation, dose accumulation, etc. DIR is therefore unavoidable in lung RT. DIR algorithms have been studied for decades and are now available both within commercial and academic packages. However, they are complex and have limitations that every user must be aware of before clinical implementation. In this paper, the main applications of DIR for lung RT with their associated uncertainties and their limitations are reviewed.

A method to detect landmark pairs accurately between intra-patient volumetric medical images

PURPOSES

An image processing procedure was developed in this study to detect large quantity of landmark pairs accurately in pairs of volumetric medical images. The detected landmark pairs can be used to evaluate of deformable image registration (DIR) methods quantitatively.

METHODS

Landmark detection and pair matching were implemented in a Gaussian pyramid multi-resolution scheme. A 3D scale-invariant feature transform (SIFT) feature detection method and a 3D Harris-Laplacian corner detection method were employed to detect feature points, i.e., landmarks. A novel feature matching algorithm, Multi-Resolution Inverse-Consistent Guided Matching or MRICGM, was developed to allow accurate feature pairs matching. MRICGM performs feature matching using guidance by the feature pairs detected at the lower resolution stage and the higher confidence feature pairs already detected at the same resolution stage, while enforces inverse consistency.

RESULTS

The proposed feature detection and feature pair matching algorithms were optimized to process 3D CT and MRI images. They were successfully applied between the inter-phase abdomen 4DCT images of three patients, between the original and the re-scanned radiation therapy simulation CT images of two head-neck patients, and between inter-fractional treatment MRIs of two patients. The proposed procedure was able to successfully detect and match over 6300 feature pairs on average. The automatically detected landmark pairs were manually verified and the mismatched pairs were rejected. The automatic feature matching accuracy before manual error rejection was 99.4%. Performance of MRICGM was also evaluated using seven digital phantom datasets with known ground truth of tissue deformation. On average, 11855 feature pairs were detected per digital phantom dataset with TRE = 0.77 ± 0.72 mm.

CONCLUSION

A procedure was developed in this study to detect large number of landmark pairs accurately between two volumetric medical images. It allows a semi-automatic way to generate the ground truth landmark datasets that allow quantitatively evaluation of DIR algorithms for radiation therapy applications.

Effect of deformable registration uncertainty on lung SBRT dose accumulation

PURPOSE

Deformable image registration (DIR) plays an important role in dose accumulation, such as incorporating breathing motion into the accumulation of the delivered dose based on daily 4DCBCT images. However, it is not yet well understood how the uncertainties associated with DIR methods affect the dose calculations and resulting clinical metrics. The purpose of this study is to evaluate the impact of DIR uncertainty on the clinical metrics derived from its use in dose accumulation.

METHODS

A biomechanical model based DIR method and a biomechanical-intensity-based hybrid method, which reduced the average registration error by 1.6 mm, were applied to ten lung cancer patients. A clinically relevant dose parameter [minimum dose to 0.5 cm(3) (Dmin)] was calculated for three dose scenarios using both algorithms. Dose scenarios included static (no breathing motion), predicted (breathing motion at the time of planning), and total accumulated (interfraction breathing motion). The relationship between the dose parameter and a combination of DIR uncertainty metrics, tumor volume, and dose heterogeneity of the plan was investigated.

RESULTS

Depending on the dose heterogeneity, tumor volume, and DIR uncertainty, in over 50% of the patients, differences greater than 1.0 Gy were observed in the Dmin of the tumor in the static dose calculation on exhale phase of the 4DCT. Such differences were due to the errors in propagating the tumor contours from the reference planning 4DCT phase onto a subsequent 4DCT phase using each DIR algorithm and calculating the dose on that phase. The differences in predicted dose were more subtle when breathing motion was modeled explicitly at the time of planning with only one patient exhibiting a greater than 1.0 Gy difference in Dmin. Dmin differences of up to 2.5 Gy were found in the total accumulated delivered dose due to difference in quantifying the interfraction variations. Such dose uncertainties could potentially be clinically significant.

CONCLUSIONS

Reductions in average uncertainty in DIR algorithms by 1.6 mm may have a clinically significant impact on the decision-making metrics used in dose planning and dose accumulation assessment.

A Novel Respiratory Motion Perturbation Model Adaptable to Patient Breathing Irregularities

PURPOSE

To develop a physical, adaptive motion perturbation model to predict tumor motion using feedback from dynamic measurement of breathing conditions to compensate for breathing irregularities.

METHODS AND MATERIALS

A novel respiratory motion perturbation (RMP) model was developed to predict tumor motion variations caused by breathing irregularities. This model contained 2 terms: the initial tumor motion trajectory, measured from 4-dimensional computed tomography (4DCT) images, and motion perturbation, calculated from breathing variations in tidal volume (TV) and breathing pattern (BP). The motion perturbation was derived from the patient-specific anatomy, tumor-specific location, and time-dependent breathing variations. Ten patients were studied, and 2 amplitude-binned 4DCT images for each patient were acquired within 2 weeks. The motion trajectories of 40 corresponding bifurcation points in both 4DCT images of each patient were obtained using deformable image registration. An in-house 4D data processing toolbox was developed to calculate the TV and BP as functions of the breathing phase. The motion was predicted from the simulation 4DCT scan to the treatment 4DCT scan, and vice versa, resulting in 800 predictions. For comparison, noncorrected motion differences and the predictions from a published 5-dimensional model were used.

RESULTS

The average motion range in the superoinferior direction was 9.4 ± 4.4 mm, the average ΔTV ranged from 10 to 248 mm³ (-26% to 61%), and the ΔBP ranged from 0 to 0.2 (-71% to 333%) between the 2 4DCT scans. The mean noncorrected motion difference was 2.0 ± 2.8 mm between 2 4DCT motion trajectories. After applying the RMP model, the mean motion difference was reduced significantly to 1.2 ± 1.8 mm (P=.0018), a 40% improvement, similar to the 1.2 ± 1.8 mm (P=.72) predicted with the 5-dimensional model.

CONCLUSIONS

A novel physical RMP model was developed with an average accuracy of 1.2 ± 1.8 mm for interfraction motion prediction, similar to that of a published lung motion model. This physical RMP was analytically derived and is able to adapt to breathing irregularities. Further improvement of this RMP model is under investigation.

A multiple-image-based method to evaluate the performance of deformable image registration in the pelvis

Deformable image registration (DIR) is essential for adaptive radiotherapy (RT) for tumor sites subject to motion, changes in tumor volume, as well as changes in patient normal anatomy due to weight loss. Several methods have been published to evaluate DIR-related uncertainties but they are not widely adopted. The aim of this study was, therefore, to evaluate intra-patient DIR for two highly deformable organs-the bladder and the rectum-in prostate cancer RT using a quantitative metric based on multiple image registration, the distance discordance metric (DDM). Voxel-by-voxel DIR uncertainties of the bladder and rectum were evaluated using DDM on weekly CT scans of 38 subjects previously treated with RT for prostate cancer (six scans/subject). The DDM was obtained from group-wise B-spline registration of each patient's collection of repeat CT scans. For each structure, registration uncertainties were derived from DDM-related metrics. In addition, five other quantitative measures, including inverse consistency error (ICE), transitivity error (TE), Dice similarity (DSC) and volume ratios between corresponding structures from pre- and post- registered images were computed and compared with the DDM. The DDM varied across subjects and structures; DDMmean of the bladder ranged from 2 to 13 mm and from 1 to 11 mm for the rectum. There was a high correlation between DDMmean of the bladder and the rectum (Pearson's correlation coefficient, R p  =  0.62). The correlation between DDMmean and the volume ratios post-DIR was stronger (R p  =  0.51; 0.68) than the correlation with the TE (bladder: R p  =  0.46; rectum: R p  =  0.47), or the ICE (bladder: R p  =  0.34; rectum: R p  =  0.37). There was a negative correlation between DSC and DDMmean of both the bladder (R p  =  -0.23) and the rectum (R p  =  -0.63). The DDM uncertainty metric indicated considerable DIR variability across subjects and structures. Our results show a stronger correlation with volume ratios and with the DSC using DDM compared to using ICE and TE. The DDM has the potential to quantitatively identify regions of large DIR uncertainties and consequently identify anatomical/scan outliers. The DDM can, thus, be applied to improve the adaptive RT process for tumor sites subject to motion.

Performance variations among clinically available deformable image registration tools in adaptive radiotherapy - how should we evaluate and interpret the result?

The purpose of this study is to evaluate the performance variations in commercial deformable image registration (DIR) tools for adaptive radiation therapy and further to interpret the differences using clinically available terms. Three clinical examples (prostate, head and neck (HN), and cranial spinal irradiation (CSI) with L‐spine boost) were evaluated in this study. Firstly, computerized deformed CT images were generated using simulation QA software with virtual deformations of bladder filling (prostate), neck flexion/bite‐block repositioning/tumor shrinkage (HN), and vertebral body rotation (CSI). The corresponding transformation matrices served as a “reference” for the following comparisons. Three commercialized DIR algorithms: the free‐form deformation from MIMVista 5.5 and the RegRefine from MIMMaestro 6.0, the multipass B‐spline from VelocityAI v3.0.1, and the adaptive demons from OnQ rts 2.1.15, were applied between the initial images and the deformed CT sets. The generated adaptive contours and dose distributions were compared with the “reference” and among each other. The performance in transferring contours was comparable among all three tools with an average Dice similarity coefficient of 0.81 for all the organs. However, the dose warping accuracy appeared to rely on the evaluation end points and methodologies. Point‐dose differences could show a difference of up to 23.3 Gy inside the PTVs and to overestimate up to 13.2 Gy for OARs, which was substantial for a 72 Gy prescription dose. Dosevolume histogram‐based evaluation might not be sensitive enough to illustrate all the detailed variations, while isodose assessment on a slice‐by‐slice basis could be tedious. We further explored the possibility of using 3D gamma index analysis for warping dose variation assessment, and observed differences in dose warping using different DIR tools. Overall, our results demonstrated that evaluation based only on the performance of contour transformation could not guarantee the accuracy in dose warping, while dose‐transferring validation strongly relied on the evaluation endpoint. As dose‐transferring errors could cause misinterpretations when attempting to accumulate dose for adaptive radiation therapy and more DIR tools are available for clinical use, a standard and clinically meaningful quality assurance criterion should be established for DIR QA in the near future.

PACS number(s): 87.57

Learning anatomy changes from patient populations to create artificial CT images for voxel-level validation of deformable image registration

The purpose of this study was to develop an approach to generate artificial computed tomography (CT) images with known deformation by learning the anatomy changes in a patient population for voxel‐level validation of deformable image registration. Using a dataset of CT images representing anatomy changes during the course of radiation therapy, we selected a reference image and registered the remaining images to it, either directly or indirectly, using deformable registration. The resulting deformation vector fields (DVFs) represented the anatomy variations in that patient population. The mean deformation, computed from the DVFs, and the most prominent variations, which were captured using principal component analysis (PCA), composed an active shape model that could generate random known deformations with realistic anatomy changes based on those learned from the patient population. This approach was applied to a set of 12 head and neck patients who received intensity‐modulated radiation therapy for validation. Artificial planning CT and daily CT images were generated to simulate a patient with known anatomy changes over the course of treatment and used to validate the deformable image registration between them. These artificial CT images potentially simulated the actual patients' anatomies and also showed realistic anatomy changes between different daily CT images. They were used to successfully validate deformable image registration applied to intrapatient deformation.

PACS number: 87.57.nj

A hybrid biomechanical intensity based deformable image registration of lung 4DCT

Deformable image registration (DIR) has been extensively studied over the past two decades due to its essential role in many image-guided interventions (IGI). IGI demands a highly accurate registration that maintains its accuracy across the entire region of interest. This work evaluates the improvement in accuracy and consistency by refining the results of Morfeus, a biomechanical model-based DIR algorithm. A hybrid DIR algorithm is proposed based on, a biomechanical model-based DIR algorithm and a refinement step based on a B-spline intensity-based algorithm. Inhale and exhale reconstructions of four-dimensional computed tomography (4DCT) lung images from 31 patients were initially registered using the biomechanical DIR by modeling contact surface between the lungs and the chest cavity. The resulting deformations were then refined using the intensity-based algorithm to reduce any residual uncertainties. Important parameters in the intensity-based algorithm, including grid spacing, number of pyramids, and regularization coefficient, were optimized on 10 randomly-chosen patients (out of 31). Target registration error (TRE) was calculated by measuring the Euclidean distance of common anatomical points on both images after registration. For each patient a minimum of 30 points/lung were used. Grid spacing of 8 mm, 5 levels of grid pyramids, and regularization coefficient of 3.0 were found to provide optimal results on 10 randomly chosen patients. Overall the entire patient population (n = 31), the hybrid method resulted in mean ± SD (90th%) TRE of 1.5 ± 1.4 (2.9) mm compared to 3.1 ± 1.9 (5.6) using biomechanical DIR and 2.6 ± 2.5 (6.1) using intensity-based DIR alone. The proposed hybrid biomechanical modeling intensity based algorithm is a promising DIR technique which could be used in various IGI procedures. The current investigation shows the efficacy of this approach for the registration of 4DCT images of the lungs with average accuracy of 1.5 mm.

An adaptive MR-CT registration method for MRI-guided prostate cancer radiotherapy

Magnetic Resonance images (MRI) have superior soft tissue contrast compared with CT images. Therefore, MRI might be a better imaging modality to differentiate the prostate from surrounding normal organs. Methods to accurately register MRI to simulation CT images are essential, as we transition the use of MRI into the routine clinic setting. In this study, we present a finite element method (FEM) to improve the performance of a commercially available, B-spline-based registration algorithm in the prostate region. Specifically, prostate contours were delineated independently on ten MRI and CT images using the Eclipse treatment planning system. Each pair of MRI and CT images was registered with the B-spline-based algorithm implemented in the VelocityAI system. A bounding box that contains the prostate volume in the CT image was selected and partitioned into a tetrahedral mesh. An adaptive finite element method was then developed to adjust the displacement vector fields (DVFs) of the B-spline-based registrations within the box. The B-spline and FEM-based registrations were evaluated based on the variations of prostate volume and tumor centroid, the unbalanced energy of the generated DVFs, and the clarity of the reconstructed anatomical structures. The results showed that the volumes of the prostate contours warped with the B-spline-based DVFs changed 10.2% on average, relative to the volumes of the prostate contours on the original MR images. This discrepancy was reduced to 1.5% for the FEM-based DVFs. The average unbalanced energy was 2.65 and 0.38 mJ cm(-3), and the prostate centroid deviation was 0.37 and 0.28 cm, for the B-spline and FEM-based registrations, respectively. Different from the B-spline-warped MR images, the FEM-warped MR images have clear boundaries between prostates and bladders, and their internal prostatic structures are consistent with those of the original MR images. In summary, the developed adaptive FEM method preserves the prostate volume during the transformation between the MR and CT images and improves the accuracy of the B-spline registrations in the prostate region. The approach will be valuable for the development of high-quality MRI-guided radiation therapy.

Frequency filtering based analysis on the cardiac induced lung tumor motion and its impact on the radiotherapy management

PURPOSE/OBJECTIVES

Lung tumor motion may be impacted by heartbeat in addition to respiration. This study seeks to quantitatively analyze heart-motion-induced tumor motion and to evaluate its impact on lung cancer radiotherapy.

METHODS/MATERIALS

Fluoroscopy images were acquired for 30 lung cancer patients. Tumor, diaphragm, and heart were delineated on selected fluoroscopy frames, and their motion was tracked and converted into temporal signals based on deformable registration propagation. The clinical relevance of heart impact was evaluated using the dose volumetric histogram of the redefined target volumes.

RESULTS

Correlation was found between tumor and cardiac motion for 23 patients. The heart-induced motion amplitude ranged from 0.2 to 2.6 mm. The ratio between heart-induced tumor motion and the tumor motion was inversely proportional to the amplitude of overall tumor motion. When the heart motion impact was integrated, there was an average 9% increase in internal target volumes for 17 patients. Dose coverage decrease was observed on redefined planning target volume in simulated SBRT plans.

CONCLUSIONS

The tumor motion of thoracic cancer patients is influenced by both heart and respiratory motion. The cardiac impact is relatively more significant for tumor with less motion, which may lead to clinically significant uncertainty in radiotherapy for some patients.