Accurate Deformable Image Registration Research Articles

Diffeomorphic transformer-based abdomen MRI-CT deformable image registration.

Stereotactic body radiotherapy (SBRT) is a well-established treatment modality for liver metastases in patients unsuitable for surgery. Both CT and MRI are useful during treatment planning for accurate target delineation and to reduce potential organs-at-risk (OAR) toxicity from radiation. MRI-CT deformable image registration (DIR) is required to propagate the contours defined on high-contrast MRI to CT images. An accurate DIR method could lead to more precisely defined treatment volumes and superior OAR sparing on the treatment plan. Therefore, it is beneficial to develop an accurate MRI-CT DIR for liver SBRT. To create a new deep learning model that can estimate the deformation vector field (DVF) for directly registering abdominal MRI-CT images. The proposed method assumed a diffeomorphic deformation. By using topology-preserved deformation features extracted from the probabilistic diffeomorphic registration model, abdominal motion can be accurately obtained and utilized for DVF estimation. The model integrated Swin transformers, which have demonstrated superior performance in motion tracking, into the convolutional neural network (CNN) for deformation feature extraction. The model was optimized using a cross-modality image similarity loss and a surface matching loss. To compute the image loss, a modality-independent neighborhood descriptor (MIND) was used between the deformed MRI and CT images. The surface matching loss was determined by measuring the distance between the warped coordinates of the surfaces of contoured structures on the MRI and CT images. To evaluate the performance of the model, a retrospective study was carried out on a group of 50 liver cases that underwent rigid registration of MRI and CT scans. The deformed MRI image was assessed against the CT image using the target registration error (TRE), Dice similarity coefficient (DSC), and mean surface distance (MSD) between the deformed contours of the MRI image and manual contours of the CT image. When compared to only rigid registration, DIR with the proposed method resulted in an increase of the mean DSC values of the liver and portal vein from 0.850±0.102 and 0.628±0.129 to 0.903±0.044 and 0.763±0.073, a decrease of the mean MSD of the liver from 7.216±4.513mm to 3.232±1.483mm, and a decrease of the TRE from 26.238±2.769mm to 8.492±1.058mm. The proposed DIR method based on a diffeomorphic transformer provides an effective and efficient way to generate an accurate DVF from an MRI-CT image pair of the abdomen. It could be utilized in the current treatment planning workflow for liver SBRT.

Uncertainty estimation and evaluation of deformation image registration based convolutional neural networks

Objective. Fast and accurate deformable image registration (DIR), including DIR uncertainty estimation, is essential for safe and reliable clinical deployment. While recent deep learning models have shown promise in predicting DIR with its uncertainty, challenges persist in proper uncertainty evaluation and hyperparameter optimization for these methods. This work aims to develop and evaluate a model that can perform fast DIR and predict its uncertainty in seconds. Approach. This study introduces a novel probabilistic multi-resolution image registration model utilizing convolutional neural networks to estimate a multivariate normal distributed dense displacement field (DDF) in a multimodal image registration problem. To assess the quality of the DDF distribution predicted by the model, we propose a new metric based on the Kullback–Leibler divergence. The performance of our approach was evaluated against three other DIR algorithms (VoxelMorph, Monte Carlo dropout, and Monte Carlo B-spline) capable of predicting uncertainty. The evaluation of the models included not only the quality of the deformation but also the reliability of the estimated uncertainty. Our application investigated the registration of a treatment planning computed tomography (CT) to follow-up cone beam CT for daily adaptive radiotherapy. Main results. The hyperparameter tuning of the models showed a trade-off between the estimated uncertainty’s reliability and the deformation’s accuracy. In the optimal trade-off, our model excelled in contour propagation and uncertainty estimation (p <0.05) compared to existing uncertainty estimation models. We obtained an average dice similarity coefficient of 0.89 and a KL-divergence of 0.15. Significance. By addressing challenges in DIR uncertainty estimation and evaluation, our work showed that both the DIR and its uncertainty can be reliably predicted, paving the way for safe deployment in a clinical environment.

Deformable image registration for composite planned doses during adaptive radiation therapy

IntroductionSome patients have significant anatomic changes during radiotherapy, necessitating an adaptive repeat CT-simulation and re-planning. This yields two unique planning datasets that introduce uncertainty into total dose records. This study explored the impact of using deformable image registration (DIR) to spatially align repeat CT-simulation images and calculate total planned dose distributions. Materials & methodsData from 5 head-and-neck, 5 lung, and 5 sarcoma patients who had unanticipated re-planning during radiotherapy were analyzed in a treatment planning system (RayStation v6.1 RaySearch Laboratories). Total planned doses to normal tissues were calculated using two methods and the previously generated manual contours defined on each CT. The first method, termed ‘parameter addition’, simply sums the relevant DVH metrics from the initial and re-planned distributions without spatially registering the CTs. The second, termed ‘dose accumulation’, uses a validated hybrid contour/intensity-based DIR algorithm to deform initial CT and dose distribution onto the repeat CT and re-planning dose distribution. DVH metrics from the summed distribution on the repeat CT are then calculated. Dose differences for organs-at-risk between parameter addition and dose accumulation ≥100 cGy were assumed to be clinically relevant. To elucidate whether relevant differences were due to registration accuracy or contouring variability between CTs, the analysis was repeated using contours on the first CT and the same contours deformed to the repeat CT with DIR. ResultsFor all patients, high overall DIR accuracy was verified visually (qualitatively) and numerically (quantitatively) using image similarity and contour-based metrics. All head-and-neck and lung patients, and one sarcoma patient (11 of 15 total) had dose differences between parameter addition and dose accumulation ≥100 cGy, with absolute mean differences of 160 cGy (range 101–436 cGy) seen in 41 of 205 total DVH criteria. In 22 of these 41 criteria, these differences were attributed to contouring variability between CTs. After correcting for contouring variations using DIR, the mean absolute differences in 7 of these 22 criteria with a relevant result (across 6 patients) was 146 cGy (range 100–502 cGy). In only 4 DVH criteria, the DIR mapped contours had higher variations than the original contours. One lung patient had a DVH criteria exceeding the clinical dose constraint by 125 cGy with parameter addition, and with accurate DIR and dose accumulation, the criteria was actually 97 cGy lower than the constraint. ConclusionsThe use of DIR to generate total planned dose records revealed substantial dose differences in most cases compared to commonly used clinical methods (i.e. parameter addition), and altered the planned acceptance criteria in a minority. DIR is recommended to be used for future adaptive re-plans to generate total planned dose records and facilitate accurate re-contouring. More accurate dose records may also improve our understanding of clinical outcomes.

A hybrid method of correcting CBCT for proton range estimation with deep learning and deformable image registration

Objective. This study aimed to develop a novel method for generating synthetic CT (sCT) from cone-beam CT (CBCT) of the abdomen/pelvis with bowel gas pockets to facilitate estimation of proton ranges. Approach. CBCT, the same-day repeat CT, and the planning CT (pCT) of 81 pediatric patients were used for training (n = 60), validation (n = 6), and testing (n = 15) of the method. The proposed method hybridizes unsupervised deep learning (CycleGAN) and deformable image registration (DIR) of the pCT to CBCT. The CycleGAN and DIR are respectively applied to generate the geometry-weighted (high spatial-frequency) and intensity-weighted (low spatial-frequency) components of the sCT, thereby each process deals with only the component weighted toward its strength. The resultant sCT is further improved in bowel gas regions and other tissues by iteratively feeding back the sCT to adjust incorrect DIR and by increasing the contribution of the deformed pCT in regions of accurate DIR. Main results. The hybrid sCT was more accurate than deformed pCT and CycleGAN-only sCT as indicated by the smaller mean absolute error in CT numbers (28.7 ± 7.1 HU versus 38.8 ± 19.9 HU/53.2 ± 5.5 HU; P ≤ 0.012) and higher Dice similarity of the internal gas regions (0.722 ± 0.088 versus 0.180 ± 0.098/0.659 ± 0.129; P ≤ 0.002). Accordingly, the hybrid method resulted in more accurate proton range for the beams intersecting gas pockets (11 fields in 6 patients) than the individual methods (the 90th percentile error in 80% distal fall-off, 1.8 ± 0.6 mm versus 6.5 ± 7.8 mm/3.7 ± 1.5 mm; P ≤ 0.013). The gamma passing rates also showed a significant dosimetric advantage by the hybrid method (99.7 ± 0.8% versus 98.4 ± 3.1%/98.3 ± 1.8%; P ≤ 0.007). Significance. The hybrid method significantly improved the accuracy of sCT and showed promises in CBCT-based proton range verification and adaptive replanning of abdominal/pelvic proton therapy even when gas pockets are present in the beam path.

A multi-scale framework with unsupervised joint training of convolutional neural networks for pulmonary deformable image registration

To achieve accurate and fast deformable image registration (DIR) for pulmonary CT, we proposed a Multi-scale DIR framework with unsupervised Joint training of Convolutional Neural Network (MJ-CNN). MJ-CNN contains three models at multi-scale levels for a coarse-to-fine DIR to avoid being trapped in a local minimum. It is trained based on image similarity and deformation vector field (DVF) smoothness, requiring no supervision of ground-truth DVF. The three models are first trained sequentially and separately for their own registration tasks, and then are trained jointly for an end-to-end optimization under the multi-scale framework. In this study, MJ-CNN was trained using public SPARE 4D-CT data. The trained MJ-CNN was then evaluated on public DIR-LAB 4D-CT dataset as well as clinical CT-to-CBCT and CBCT-to-CBCT registration. For 4D-CT inter-phase registration, MJ-CNN achieved comparable accuracy to conventional iteration optimization-based methods, and showed the smallest registration errors compared to recently published deep learning-based DIR methods, demonstrating the efficacy of the proposed multi-scale joint training scheme. Besides, MJ-CNN trained using one dataset (SPARE) could generalize to a different dataset (DIR-LAB) acquired by different scanners and imaging protocols. Furthermore, MJ-CNN trained on 4D-CTs also performed well on CT-to-CBCT and CBCT-to-CBCT registration without any re-training or fine-tuning, demonstrating MJ-CNN’s robustness against applications and imaging techniques. MJ-CNN took about 1.4 s for DVF estimation and required no manual-tuning of parameters during the evaluation. MJ-CNN is able to perform accurate DIR for pulmonary CT with nearly real-time speed, making it very applicable for clinical tasks.

Evaluating the Accuracy of Commercial Deformable Image Registration Software for Real Patient Images Using Anthropomorphic Modeling

Deformable image registration (DIR) is widely used nowadays to register images acquired throughout the radiation oncology treatment course for target delineation, dose accumulation, and adaptive planning. However, the accuracy of commercial DIR software remains questionable and there lacks a “ground truth” to benchmark DIR software for real patient images. We introduced a new method based on anthropomorphic modeling to provide accurate DIR for head and neck patients and evaluated the accuracy of commercial DIR software. Four head and neck patients from the Cancer Imaging Archive, Collection Head-Neck Cetuximab were selected to demonstrate the concept. Each patient had two PETCTs and a planning CT taken with the patient at different positions. Two patients had arms up in PETCT and down in planning CT. The bony structures in each CT were automatically segmented into individual bone groups (e.g. each vertebra) using image processing, clustering, and pattern matching. Model-based DIR rigidly registers each bone group using Iterative Closest Point algorithm then deforms soft tissue with the bones using finite element method. Model-based DIR can be considered as the “truth” of DIR at the bones. The accuracy of rigid registrations and four commercial DIR software were evaluated. The mean absolute errors of DIR for multiple rigid registration (the best of three rigid registrations aligning to base of skull, mid-neck, and lower neck/shoulder respectively), DIR software A (constrained, intensity based, free-form algorithm), B (hybrid algorithm designed for head and neck), C (multi-resolution B-Spline algorithm) and D (hybrid algorithm using image intensity and region of interest constraints) were listed in Table 1.Abstract TU_38_3272; Table 1Mean absolute errors of multiple rigid registration and four DIR software for head and neck patient imagesMean absolute error (cm)HeadJawSpineClavicleScapulaeHumerus (Arms up/down)Humerus (Arm down)SternumRibsMultiple rigid registrations0.260.510.721.892.147.922.430.610.77Software A0.570.250.611.591.506.551.580.350.65Software B0.640.300.631.161.376.830.920.300.78Software C0.430.350.851.471.515.221.170.561.07Software D0.660.320.741.311.885.841.680.450.85 Open table in a new tab Anthropomorphic modeling provides the truth of DIR for bones in head and neck patient images which can be used to measure the accuracy of commercial DIR software. For the patients studied, the mean DIR errors were on the order of millimeters except for large posture change (arms up/down). The performance of DIR software was comparable with each other and with multiple rigid registrations despite the difference in the algorithm used. There is not an algorithm that outperforms others in all regions.

44. Deformable image registration for dose accumulation: Principle and evaluation

Most Treatment Planning Systems for adaptive radiotherapy today propose Deformable Image Registration (DIR) tools which enable contour propagation or dose accumulation between images acquired over the course of treatment. While the results of DIR for contour propagation can be easily assessed and eventually manually corrected, the evaluation of DIR for dose accumulation relies on a knowledge of anatomical point correspondences and is therefore much more challenging. This talk will review the methodological principle of DIR for dose accumulation and current evaluation methods. Accumulating dose in a tissue element contained in a voxel of the reference image requires to determine the location of that element in the considered longitudinal images. The role of DIR is to automatically establish those spatial correspondences usually in the form of displacement vector fields (DVF). The first natural way to evaluate DIR results is to deform the longitudinal image onto the reference image based on the DVF and to measure the resulting similarity between the images. However, as demonstrated in previous studies, popular DIR algorithms for contour propagation can lead to high similarity between the deformed and reference images and critically wrong anatomical point correspondences. For this reason, image similarity measures or visual assessment with image fusion tools should be considered in clinical practice only after the local accuracy (in millimeters) has been quantified for a significant number of cases. Among DIR validation methods for dose accumulation, the most reliable one consists of manually identifying pairs of landmarks in the images and to relate the misalignment of those landmark pairs after DIR to the spatial resolution of the dose distributions. This approach can be easily implemented for some anatomical localizations but identifying correspondences on the surface of smooth organs or in large homogeneous tissues is not always feasible. In that case, DIR validation will have to verify that the estimated DVF respect physical considerations such as a reasonable range of local volume variations measured with the Jacobian determinant of the DVF. DIR algorithms can also be tested with images derived from numerical or physical phantoms which provide a ground truth deformation. Phantoms have demonstrated that traditional DIR algorithms could fail to estimate complex deformations such as a sliding motion between organs and justified the need to develop anatomical site-specific models. Finally, a limited number of studies have addressed DIR of longitudinal images presenting inconsistent content (i.e. in case of tumor growth or erosion) and a special attention should be paid in those situations. In parallel of efforts to develop more accurate DIR methods, it is important today to define a commissioning strategy for available DIR solutions to be used in clinical trials involving dose accumulation. In that sense, a Task Group of the American Association of Medical Physicists recently reported its recommendations.

Learning-based deformable registration for infant MRI by integrating random forest with auto-context model.

Accurately analyzing the rapid structural evolution of human brain in the first year of life is a key step in early brain development studies, which requires accurate deformable image registration. However, due to (a) dynamic appearance and (b) large anatomical changes, very few methods in the literature can work well for the registration of two infant brain MR images acquired at two arbitrary development phases, such as birth and one-year-old. To address these challenging issues, we propose a learning-based registration method, which can handle the anatomical structures and the appearance changes between the two infant brain MR images with possible time gap. Specifically, in the training stage, we employ a multioutput random forest regression and auto-context model to learn the evolution of anatomical shape and appearance from a training set of longitudinal infant images. To make the learning procedure more robust, we further harness the multimodal MR imaging information. Then, in the testing stage, for registering the two new infant images scanned at two different development phases, the learned model will be used to predict both the deformation field and appearance changes between the images under registration. After that, it becomes much easier to deploy any conventional image registration method to complete the remaining registration since the above-mentioned challenges for state-of-the-art registration methods have been well addressed. We have applied our proposed registration method to intersubject registration of infant brain MR images acquired at 2-week-old, 3-month-old, 6-month-old, and 9-month-old with the images acquired at 12-month-old. Promising registration results have been achieved in terms of registration accuracy, compared with the counterpart nonlearning based registration methods. The proposed new learning-based registration method have tackled the challenging issues in registering infant brain images acquired from the first year of life, by leveraging the multioutput random forest regression with auto-context model, which can learn the evolution of shape and appearance from a training set of longitudinal infant images. Thus, for the new infant image, its deformation field to the template and also its template-like appearances can be predicted by the learned models. We have extensively compared our method with state-of-the-art deformable registration methods, as well as multiple variants of our method, which show that our method can achieve higher accuracy even for the difficult cases with large appearance and shape changes between subject and template images.

Accuracy of deformable image registration on magnetic resonance images in digital and physical phantoms.

Accurate deformable image registration is necessary for longitudinal studies. The error associated with commercial systems has been evaluated using computed tomography (CT). Several in-house algorithms have been evaluated for use with magnetic resonance imaging (MRI), but there is still relatively little information about MRI deformable image registration. This work presents an evaluation of two deformable image registration systems, one commercial (Velocity) and one in-house (demons-based algorithm), with MRI using two different metrics to quantify the registration error. The registration error was analyzed with synthetic MR images. These images were generated from interpatient and intrapatient variation models trained on 28 patients. Four synthetic post-treatment images were generated for each of four synthetic pretreatment images, resulting in 16 image registrations for both the T1- and T2-weighted images. The synthetic post-treatment images were registered to their corresponding synthetic pretreatment image. The registration error was calculated between the known deformation vector field and the generated deformation vector field from the image registration system. The registration error was also analyzed using a porcine phantom with ten implanted 0.35-mm diameter gold markers. The markers were visible on CT but not MRI. CT, T1-weighted MR, and T2-weighted MR images were taken in four different positions. The markers were contoured on the CT images and rigidly registered to their corresponding MR images. The MR images were deformably registered and the distance between the projected marker location and true marker location was measured as the registration error. The synthetic images were evaluated only on Velocity. Root mean square errors (RMSEs) of 0.76 mm in the left-right (LR) direction, 0.76 mm in the anteroposterior (AP) direction, and 0.69 mm in the superior-inferior (SI) direction were observed for the T1-weighted MR images. RMSEs of 1.1 mm in the LR direction, 0.75 mm in the AP direction, and 0.81 mm in the SI direction were observed for the T2-weighted MR images. The porcine phantom MR images, when evaluated with Velocity, had RMSEs of 1.8, 1.5, and 2.7 mm in the LR, AP, and SI directions for the T1-weighted images and 1.3, 1.2, and 1.6 mm in the LR, AP, and SI directions for the T2-weighted images. When the porcine phantom images were evaluated with the in-house demons-based algorithm, RMSEs were 1.2, 1.5, and 2.1 mm in the LR, AP, and SI directions for the T1-weighted images and 0.81, 1.1, and 1.1 mm in the LR, AP, and SI directions for the T2-weighted images. The MRI registration error was low for both Velocity and the in-house demons-based algorithm according to both image evaluation methods, with all RMSEs below 3 mm. This implies that both image registration systems can be used for longitudinal studies using MRI.

TU‐AB‐202‐02: Deformable Image Registration Accuracy Between External Beam Radiotherapy and HDR Brachytherapy CT Images for Cervical Cancer Using a 3D‐Printed Deformable Pelvis Phantom

Purpose:Accurate deformable image registration (DIR) between external beam radiotherapy (EBRT) and HDR brachytherapy (BT) CT images in cervical cancer is challenging. DSC has been evaluated only on the basis of the consistency of the structure, and its use does not guarantee an anatomically reasonable deformation. We evaluate the DIR accuracy for cervical cancer with DSC and anatomical landmarks using a 3D‐printed pelvis phantom.Methods:A 3D‐printed, deformable female pelvis phantom was created on the basis of the patient's CT image. Urethane and silicon were used as materials for creating the uterus and bladder, respectively, in the phantom. We performed DIR in two cases: case‐A with a full bladder (170 ml) in both the EBRT and BT images and case‐B with a full bladder in the BT image and a half bladder (100 ml) in the EBRT image. DIR was evaluated using DSCs and 70 uterus and bladder landmarks. A Hybrid intensity and structure DIR algorithm with two settings (RayStation) was used.Results:In the case‐A, DSCs of the intensity‐based DIR were 0.93 and 0.85 for the bladder and uterus, respectively, whereas those of hybrid‐DIR were 0.98 and 0.96, respectively. The mean landmark error values of intensity‐based DIR were 0.73±0.29 and 1.70±0.19 cm for the bladder and uterus, respectively, whereas those of Hybrid‐DIR were 0.43±0.33 and 1.23±0.25 cm, respectively. In both cases, the Hybrid‐DIR accuracy was better than the intensity‐based DIR accuracy for both evaluation methods. However, for several bladder landmarks, the Hybrid‐DIR landmark errors were larger than the corresponding intensity‐based DIR errors (e.g., 2.26 vs 1.25 cm).Conclusion:Our results demonstrate that Hybrid‐DIR can perform with a better accuracy than the intensity‐based DIR for both DSC and landmark errors; however, Hybrid‐DIR shows a larger landmark error for some landmarks because the technique focuses on both the structure and intensity.

TH‐EF‐BRA‐11: Feasibility of Super‐Resolution Time‐Resolved 4DMRI for Multi‐Breath Volumetric Motion Simulation in Radiotherapy Planning

Purpose:To develop a novel super‐resolution time‐resolved 4DMRI technique to evaluate multi‐breath, irregular and complex organ motion without respiratory surrogate for radiotherapy planning.Methods:The super‐resolution time‐resolved (TR) 4DMRI approach combines a series of low‐resolution 3D cine MRI images acquired during free breathing (FB) with a high‐resolution breath‐hold (BH) 3DMRI via deformable image registration (DIR). Five volunteers participated in the study under an IRB‐approved protocol. The 3D cine images with voxel size of 5×5×5 mm3 at two volumes per second (2Hz) were acquired coronally using a T1 fast field echo sequence, half‐scan (0.8) acceleration, and SENSE (3) parallel imaging. Phase‐encoding was set in the lateral direction to minimize motion artifacts. The BH image with voxel size of 2×2×2 mm3 was acquired using the same sequence within 10 seconds. A demons‐based DIR program was employed to produce super‐resolution 2Hz 4DMRI. Registration quality was visually assessed using difference images between TR 4DMRI and 3D cine and quantitatively assessed using average voxel correlation. The fidelity of the 3D cine images was assessed using a gel phantom and a 1D motion platform by comparing mobile and static images.Results:Owing to voxel intensity similarity using the same MRI scanning sequence, accurate DIR between FB and BH images is achieved. The voxel correlations between 3D cine and TR 4DMRI are greater than 0.92 in all cases and the difference images illustrate minimal residual error with little systematic patterns. The 3D cine images of the mobile gel phantom preserve object geometry with minimal scanning artifacts.Conclusion:The super‐resolution time‐resolved 4DMRI technique has been achieved via DIR, providing a potential solution for multi‐breath motion assessment. Accurate DIR mapping has been achieved to map high‐resolution BH images to low‐resolution FB images, producing 2Hz volumetric high‐resolution 4DMRI. Further validation and improvement are still required prior to clinical applications.This study is in part supported by the NIH (U54CA137788/U54CA132378).

SU‐E‐J‐114: A Practical Hybrid Method for Improving the Quality of CT‐CBCT Deformable Image Registration for Head and Neck Radiotherapy

Purpose:Accurate deformable image registration (DIR) between CT and CBCT in H&amp;N is challenging. In this study, we propose a practical hybrid method that uses not only the pixel intensities but also organ physical properties, structure volume of interest (VOI), and interactive local registrations.Methods:Five oropharyngeal cancer patients were selected retrospectively. For each patient, the planning CT was registered to the last fraction CBCT, where the anatomy difference was largest. A three step registration strategy was tested; Step1) DIR using pixel intensity only, Step2) DIR with additional use of structure VOI and rigidity penalty, and Step3) interactive local correction. For Step1, a public‐domain open‐source DIR algorithm was used (cubic B‐spline, mutual information, steepest gradient optimization, and 4‐level multi‐resolution). For Step2, rigidity penalty was applied on bony anatomies and brain, and a structure VOI was used to handle the body truncation such as the shoulder cut‐off on CBCT. Finally, in Step3, the registrations were reviewed on our in‐house developed software and the erroneous areas were corrected via a local registration using level‐set motion algorithm.Results:After Step1, there were considerable amount of registration errors in soft tissues and unrealistic stretching in the posterior to the neck and near the shoulder due to body truncation. The brain was also found deformed to a measurable extent near the superior border of CBCT. Such errors could be effectively removed by using a structure VOI and rigidity penalty. The rest of the local soft tissue error could be corrected using the interactive software tool. The estimated interactive correction time was approximately 5 minutes.Conclusion:The DIR using only the image pixel intensity was vulnerable to noise and body truncation. A corrective action was inevitable to achieve good quality of registrations. We found the proposed three‐step hybrid method efficient and practical for CT/CBCT registrations in H&amp;N.My department receives grant support from Industrial partners: (a) Varian Medical Systems, Palo Alto, CA, and (b) Philips HealthCare, Best, Netherlands

TU‐CD‐BRA‐08: Single‐Energy Computed Tomography‐Based Pulmonary Perfusion Imaging: Proof‐Of‐Principle in a Canine Model

Purpose:Pulmonary perfusion imaging has provided significant insights into pulmonary diseases, and can be useful in radiotherapy. The purpose of this study was to prospectively establish proof‐of‐principle in a canine model for single‐energy CT‐based perfusion imaging, which has the potential for widespread clinical implementation.Methods:Single‐energy CT perfusion imaging is based on: (1) acquisition of inspiratory breath‐hold CT scans before and after intravenous injection of iodinated contrast medium, (2) deformable image registration (DIR) of the two CT image data sets, and (3) subtraction of the pre‐contrast image from post‐contrast image, yielding a map of Hounsfield unit (HU) enhancement. These subtraction image data sets hypothetically represent perfused blood volume, a surrogate for perfusion. In an IACUC‐approved clinical trial, we acquired pre‐ and post‐contrast CT scans in the prone posture for six anesthetized, mechanically‐ventilated dogs. The elastix algorithm was used for DIR. The registration accuracy was quantified using the target registration errors (TREs) for 50 pulmonary landmarks in each dog. The gradient of HU enhancement between gravity‐dependent (ventral) and non‐dependent (dorsal) regions was evaluated to quantify the known effect of gravity, i.e., greater perfusion in ventral regions.Results:The lung volume difference between the two scans was 4.3±3.5% on average (range 0.3%–10.1%). DIR demonstrated an average TRE of 0.7±1.0 mm. HU enhancement in lung parenchyma was 34±10 HU on average and varied considerably between individual dogs, indicating the need for improvement of the contrast injection protocol. HU enhancement in ventral (gravity‐dependent) regions was found to be greater than in dorsal regions. A population average ventral‐to‐dorsal gradient of HU enhancement was strong (R2=0.94) and statistically significant (p&lt;0.01).Conclusion:This canine study demonstrated relatively accurate DIR and a strong ventral‐to‐dorsal gradient of HU enhancement, providing proof‐of‐principle for single‐energy CT pulmonary perfusion imaging. This ongoing study will enroll more dogs and investigate the physiological significance.This study was supported by a Philips Healthcare/Radiological Society of North America (RSNA) Research Seed Grant (RSD1458).

SU‐E‐J‐117: Comparison of Different QA Methods for Deformable Image Registration to the Known Errors for Prostate and Head‐And‐Neck Virtual Phantoms

Purpose:Several methods can evaluate the accuracy of deformable image registration (DIR), but they are not necessarily a true measurement of DIR accuracy. The purpose here is to evaluate how these methods compare to known errors found with prostate and head‐and‐neck virtual phantoms.Methods:Quality assurance of DIR falls in three basic categories: contour comparison, landmark tracking, and image similarity. These different methods were utilized to evaluate the performance of four DIR algorithms, MIM and three from Velocity: Deformable (DEF), Deformable Multi‐Pass (DMP), and Extended Deformable Multi‐Pass (XMP). For contour comparison, organs were contoured (total of 15) on both the undeformed and deformed images. Then, the DIR algorithms were used to transfer contours from the undeformed to the deformed image and compared to that drawn directly on the deformed image. For landmark tracking, we found visible landmarks and measured their locations on the deformed and the undeformed images. The resulting deformation measurement was then compared to that predicted by the algorithm. For image similarity, we calculated the root‐mean‐square and the mean‐absolute differences between the deformed and warped undeformed images. Finally, we calculated the actual spatial registration error for each DIR algorithm, from the known deformations.Results:The MIM algorithm produced the lowest average errors for the landmark analysis, the closest image similarity, and overall the most accurate contour transfers. When compared to the known deformations, MIM also produced the lowest average error, but the largest spatial registration errors. The lowest maximum spatial errors were made by DEF and DMP for the prostate and head‐and‐neck phantoms, respectively.Conclusion:Contour comparison, landmark tracking, and image similarity all indicated that MIM was the most accurate DIR algorithm. These metrics were aligned with MIM producing the lowest known mean errors, but did not catch that MIM was also by far producing the largest known errors.

SU‐E‐J‐116: Uncertainties Associated with Dose Summation of High‐Dose Rate Brachytherapy and Intensity Modulated Radiotherapy for Gynecological Cases

Purpose:Determining the cumulative dose distribution (CDD) for gynecological patients treated with both high‐dose rate (HDR) brachytherapy and intensity‐modulated radiotherapy (IMRT) is challenging. The purpose of this work is to study the uncertainty of performing this with a structure‐guided deformable (SGD) approach in Velocity.Methods:For SGD, the Hounsfield units inside specified contours are overridden to set uniform values. Deformable image registration (DIR) is the run on these process images, which forces the DIR to focus on these contour boundaries. 18 gynecological cancer patients were used in this study. The original bladder and rectum planning contours for these patients were used to drive the SGD. A second set of contours were made of the bladder by the same person with the intent of carefully making them completely consistent with each other. This second set was utilized to evaluate the spatial accuracy of the SGD. The determined spatial accuracy was then multiplied by the local dose gradient to determine a dose uncertainty associated with the SGD dose warping. The normal tissue complication probability (NTCP) was then calculated for each dose volume histogram (DVH) that included four different probabilistic uncertainties associated with the spatial errors (e.g., 68.3% and 95.4%).Results:The NTCPs for each DVH (e.g., NTCP_−95.4%, NTCP_−68.3%, NTCP_68.3%, NTCP_95.4%) differed amongst patients. All patients had an NTCP_−95.4% close to 0%, while NTCP_95.4% ranged from 0.67% to 100%. Nine patients had an NTCP_−95.4% less than 50% while the remaining nine patients had NTCP_95.4% greater than 50%.Conclusion:The uncertainty associated with this CDD technique renders a large NTCP uncertainty. Thus, it is currently not practical for clinical use. The two ways to improve this would be to use more precise contours to drive the SGD and to use a more accurate DIR algorithm.

MO‐C‐17A‐11: A Segmentation and Point Matching Enhanced Deformable Image Registration Method for Dose Accumulation Between HDR CT Images

Purpose:To propose and validate a novel and accurate deformable image registration (DIR) scheme to facilitate dose accumulation among treatment fractions of high‐dose‐rate (HDR) gynecological brachytherapy.Method:We have developed a method to adapt DIR algorithms to gynecologic anatomies with HDR applicators by incorporating a segmentation step and a point‐matching step into an existing DIR framework. In the segmentation step, random walks algorithm is used to accurately segment and remove the applicator region (AR) in the HDR CT image. A semi‐automatic seed point generation approach is developed to obtain the incremented foreground and background point sets to feed the random walks algorithm. In the subsequent point‐matching step, a feature‐based thin‐plate spline‐robust point matching (TPS‐RPM) algorithm is employed for AR surface point matching. With the resulting mapping, a DVF characteristic of the deformation between the two AR surfaces is generated by B‐spline approximation, which serves as the initial DVF for the following Demons DIR between the two AR‐free HDR CT images. Finally, the calculated DVF via Demons combined with the initial one serve as the final DVF to map doses between HDR fractions.Results:The segmentation and registration accuracy are quantitatively assessed by nine clinical HDR cases from three gynecological cancer patients. The quantitative results as well as the visual inspection of the DIR indicate that our proposed method can suppress the interference of the applicator with the DIR algorithm, and accurately register HDR CT images as well as deform and add interfractional HDR doses.Conclusions:We have developed a novel and robust DIR scheme that can perform registration between HDR gynecological CT images and yield accurate registration results. This new DIR scheme has potential for accurate interfractional HDR dose accumulation.This work is supported in part by the National Natural ScienceFoundation of China (no 30970866 and no 81301940).

Effects of quantum noise in 4D-CT on deformable image registration and derived ventilation data

Quantum noise is common in CT images and is a persistent problem in accurate ventilation imaging using 4D-CT and deformable image registration (DIR). This study focuses on the effects of noise in 4D-CT on DIR and thereby derived ventilation data. A total of six sets of 4D-CT data with landmarks delineated in different phases, called point-validated pixel-based breathing thorax models (POPI), were used in this study. The DIR algorithms, including diffeomorphic morphons (DM), diffeomorphic demons (DD), optical flow and B-spline, were used to register the inspiration phase to the expiration phase. The DIR deformation matrices (DIRDM) were used to map the landmarks. Target registration errors (TRE) were calculated as the distance errors between the delineated and the mapped landmarks. Noise of Gaussian distribution with different standard deviations (SD), from 0 to 200 Hounsfield Units (HU) in amplitude, was added to the POPI models to simulate different levels of quantum noise. Ventilation data were calculated using the ΔV algorithm which calculates the volume change geometrically based on the DIRDM. The ventilation images with different added noise levels were compared using Dice similarity coefficient (DSC). The root mean square (RMS) values of the landmark TRE over the six POPI models for the four DIR algorithms were stable when the noise level was low (SD <150 HU) and increased with added noise when the level is higher. The most accurate DIR was DD with a mean RMS of 1.5 ± 0.5 mm with no added noise and 1.8 ± 0.5 mm with noise (SD = 200 HU). The DSC values between the ventilation images with and without added noise decreased with the noise level, even when the noise level was relatively low. The DIR algorithm most robust with respect to noise was DM, with mean DSC = 0.89 ± 0.01 and 0.66 ± 0.02 for the top 50% ventilation volumes, as compared between 0 added noise and SD = 30 and 200 HU, respectively. Although the landmark TRE were stable with low noise, the differences between ventilation images increased with noise level, even when the noise was low, indicating ventilation imaging from 4D-CT was sensitive to image noise. Therefore, high quality 4D-CT is essential for accurate ventilation images.

SU‐E‐J‐66: Effects of Noise in 4D‐CT On Deformable Image Registration and Derived Ventilation Data

Purpose: Deformable image registration (DIR) and 4D‐CT have been proposed to generate ventilation images. Quantum noise is common in CT images. This study focuses on the effects of noise in 4D‐CT on DIR and the derived ventilation data. Methods: Diffeomorphic morphons (DM), diffeomorphic demons (DD), optical flow and B‐Spline were used to register the end‐inspiration phase to the end‐expiration phase of 6 4D‐CT sets with landmarks delineated on different phases, called point‐validated pixel‐based breathing thorax models (POPI). Landmarks at expiration were mapped to inspiration using DIR deformation matrices (DIRDM) for each POPI model. Target registration errors (TRE) were calculated as the distances between the delineated and the mapped landmarks. Gaussian noise with different standard deviations (SD) of the amplitude was added to the POPI models to simulate different levels of quantum noise. Ventilation estimations were performed by calculating the volume change geometrically, based on the DIRDM. Ventilation images with different CT noise levels were compared using Dice similarity coefficient (DSC). Results: The root mean square (RMS) values of the landmark TRE over the 6 POPI models for the 4 DIR algorithms were stable when the noise level was below SD=150 Hounsfield Units (HU), and increased with the noise level. The most accurate DIR was DD with mean RMS of 1.5±0.5 and 1.8+−0.5 mm at the added noise SD=0 and 200 HU respectively. The DSC values between the ventilation images with and without added noise decreased with the noise level. The most consistent DIR was DM with mean DSC=0.89+− 0.01 and 0.66+−0.02 for the top 50% ventilation volumes with the added noise of 0 and 30, 0 and 200 HU respectively. Conclusion: While the landmark TRE was stable with noise level for low noise, the differences between ventilation images increased, indicating that 4D‐CT based ventilation imaging is sensitive to image noise. This work was partially supported by a grant from the Varian Medical Systems, Inc.

TU‐C‐141‐02: Development of a Hyrbid Biomechanical Model Based Deformable Image Registration; Application in Lung

Purpose: To investigate the reduction in maximum errors in 4D CT lung Deformable Image Registration (DIR) using a hybrid biomechanical model based algorithm with an intensity based registration refinement. Methods: Morfeus, a biomechanical modeling algorithm for deformable image registration was used to register 3D inhale to exhale images of the lungs acquired via 4DCT. The predicted exhale image was obtained by interpolation. An intensity based registration algorithm based on Sum of Absolute Differences (SAD) and BSpline free form deformations (drop3D, Munich, Germany) was then employed to refine the results by a second registration from the Morfeus predicted exhale to the actual exhale image. The BSpline grid point spacing for the refinement was set to 9mm. To measure the accuracy, Target Registration Error (TRE) was calculated based on the Euclidean distance of corresponding bifurcation points on both images. Improvements in the overall TRE as well as reduction in maximum errors were evaluated. Results: The hybrid method has been evaluated on 5 patients to date with 45 to 76 identified bifurcation points per patient. The Morfeus TRE ranged from 2.2±1.0 mm (mean±SD) to 3.6±1.9 mm. This was reduced to 1.4±0.8 mm up to 1.6±2.0 mm with the hybrid method. More importantly, the number of bifurcation points (% of total points) with an error greater than 5 mm was reduced from 54 (22%) points (across all patients, range 3 to 24 points per patient) with Morfeus to 10 (4%) points (range 0 to 4 points per patient) with the hybrid method. Conclusion: The proposed hybrid method combines the flexibility of the biomechanical model based algorithm, including the ability to model the sliding interface between the lungs and the chest wall, with the advantages of intensity based registration for the contrast rich lung images to provide a highly accurate DIR algorithm. This research is supported by the NIH 5RO1CA124714‐02. KK Brock has financial interest in the deformable registration technology through a licensing agreement with RaySearch Laboratories.

Control over structure-specific flexibility improves anatomical accuracy for point-based deformable registration in bladder cancer radiotherapy

Future developments in image guided adaptive radiotherapy (IGART) for bladder cancer require accurate deformable image registration techniques for the precise assessment of tumor and bladder motion and deformation that occur as a result of large bladder volume changes during the course of radiotherapy treatment. The aim was to employ an extended version of a point-based deformable registration algorithm that allows control over tissue-specific flexibility in combination with the authors' unique patient dataset, in order to overcome two major challenges of bladder cancer registration, i.e., the difficulty in accounting for the difference in flexibility between the bladder wall and tumor and the lack of visible anatomical landmarks for validation. The registration algorithm used in the current study is an extension of the symmetric-thin plate splines-robust point matching (S-TPS-RPM) algorithm, a symmetric feature-based registration method. The S-TPS-RPM algorithm has been previously extended to allow control over the degree of flexibility of different structures via a weight parameter. The extended weighted S-TPS-RPM algorithm was tested and validated on CT data (planning- and four to five repeat-CTs) of five urinary bladder cancer patients who received lipiodol injections before radiotherapy. The performance of the weighted S-TPS-RPM method, applied to bladder and tumor structures simultaneously, was compared with a previous version of the S-TPS-RPM algorithm applied to bladder wall structure alone and with a simultaneous nonweighted S-TPS-RPM registration of the bladder and tumor structures. Performance was assessed in terms of anatomical and geometric accuracy. The anatomical accuracy was calculated as the residual distance error (RDE) of the lipiodol markers and the geometric accuracy was determined by the surface distance, surface coverage, and inverse consistency errors. Optimal parameter values for the flexibility and bladder weight parameters were determined for the weighted S-TPS-RPM. The weighted S-TPS-RPM registration algorithm with optimal parameters significantly improved the anatomical accuracy as compared to S-TPS-RPM registration of the bladder alone and reduced the range of the anatomical errors by half as compared with the simultaneous nonweighted S-TPS-RPM registration of the bladder and tumor structures. The weighted algorithm reduced the RDE range of lipiodol markers from 0.9-14 mm after rigid bone match to 0.9-4.0 mm, compared to a range of 1.1-9.1 mm with S-TPS-RPM of bladder alone and 0.9-9.4 mm for simultaneous nonweighted registration. All registration methods resulted in good geometric accuracy on the bladder; average error values were all below 1.2 mm. The weighted S-TPS-RPM registration algorithm with additional weight parameter allowed indirect control over structure-specific flexibility in multistructure registrations of bladder and bladder tumor, enabling anatomically coherent registrations. The availability of an anatomically validated deformable registration method opens up the horizon for improvements in IGART for bladder cancer.