Non-rigid Image Registration Algorithm Research Articles

Hybrid Eulerian–Lagrangian framework for structural full-field vibration quantification and modal shape visualization

Vibration measurement and modal analysis under a high spatial and temporal resolution have high significance for revealing inner structural characteristics. This work develops a hybrid Eulerian–Lagrangian framework for structural full-field dynamic vibration quantification and modal shape visualization and validates its performance through vision-based vibration experiments. By assuming a linear vibration system, modal responses that are estimated from the vibration signals on a structural surface provide a reliable basis for manipulating temporal and spatial vibrations in Eulerian motion processing. Then, dynamic spatial motion can be quantified by using the improved non-rigid image registration algorithm. Meanwhile, from the Lagrangian perspective, high-quality visualization of modal shapes is implemented by compensating spatial motion according to magnified dense displacement fields. The experiment results showed that the proposed hybrid framework has superiority in controlled modal tests and has potential for more practical applications.

Ultrasound strain mapping of the mouse Achilles tendon during passive dorsiflexion

Immediately prior to inserting into bone, many healthy tendons experience impingement from nearby bony structures. However, super-physiological levels of impingement are implicated in insertional tendinopathies. Unfortunately, the mechanisms underlying the connection between impingement and tendon pathology remain poorly understood, in part due to the shortage of well-characterized animal models of impingement at clinically relevant sites. As a first step towards developing a model of excessive tendon impingement, the objective of this study was to characterize the mechanical strain environment in the mouse Achilles tendon insertion under passive dorsiflexion and confirm that – like humans – mice experience impingement of the tendon insertion from the calcaneus (heel bone) in dorsiflexed ankle positions. Based on previous work in humans, we hypothesized that during dorsiflexion, the mouse Achilles tendon insertion would experience high levels of transverse compressive strain due to calcaneal impingement. A custom-built loading platform was used to apply passive dorsiflexion, while an ultrasound transducer positioned over the Achilles tendon captured radiofrequency images. A non-rigid image registration algorithm was then used to map the transverse compressive strain based on the acquired ultrasound image sequences. Our results demonstrate that during passive dorsiflexion, transverse compressive strains were produced throughout the Achilles tendon, with significantly larger strain magnitudes at the tendon insertion than at the midsubstance. Furthermore, there was increasing transverse compressive strain observed within the Achilles tendon as a function of increasing dorsiflexion angle. This study enhances our understanding of the unique mechanical loading environment of the Achilles tendon under physiologically relevant conditions.

Accelerating B-spline interpolation on GPUs: Application to medical image registration

Background and ObjectiveB-spline interpolation (BSI) is a popular technique in the context of medical imaging due to its adaptability and robustness in 3D object modeling. A field that utilizes BSI is Image Guided Surgery (IGS). IGS provides navigation using medical images, which can be segmented and reconstructed into 3D models, often through BSI. Image registration tasks also use BSI to transform medical imaging data collected before the surgery and intra-operative data collected during the surgery into a common coordinate space. However, such IGS tasks are computationally demanding, especially when applied to 3D medical images, due to the complexity and amount of data involved. Therefore, optimization of IGS algorithms is greatly desirable, for example, to perform image registration tasks intra-operatively and to enable real-time applications. A traditional CPU does not have sufficient computing power to achieve these goals and, thus, it is preferable to rely on GPUs. In this paper, we introduce a novel GPU implementation of BSI to accelerate the calculation of the deformation field in non-rigid image registration algorithms. MethodsOur BSI implementation on GPUs minimizes the data that needs to be moved between memory and processing cores during loading of the input grid, and leverages the large on-chip GPU register file for reuse of input values. Moreover, we re-formulate our method as trilinear interpolations to reduce computational complexity and increase accuracy. To provide pre-clinical validation of our method and demonstrate its benefits in medical applications, we integrate our improved BSI into a registration workflow for compensation of liver deformation (caused by pneumoperitoneum, i.e., inflation of the abdomen) and evaluate its performance. ResultsOur approach improves the performance of BSI by an average of 6.5× and interpolation accuracy by 2× compared to three state-of-the-art GPU implementations. Through pre-clinical validation, we demonstrate that our optimized interpolation accelerates a non-rigid image registration algorithm, which is based on the Free Form Deformation (FFD) method, by up to 34%. ConclusionOur study shows that we can achieve significant performance and accuracy gains with our novel parallelization scheme that makes effective use of the GPU resources. We show that our method improves the performance of real medical imaging registration applications used in practice today.

基于互信息的Active Demons非刚性图像配准算法

基于互信息的Active Demons非刚性图像配准算法

Organ Dynamics and Hemodynamic of the Whole HH25 Avian Embryonic Heart, Revealed by Ultrasound Biomicroscopy, Boundary Tracking, and Flow Simulations

Congenital heart malformations occur to substantial number of pregnancies. Studies showed that abnormal flow biomechanical environments could lead to malformations, making it important to understand the biomechanical environment of the developing heart. We performed 4D high-frequency ultrasound scans of chick embryonic hearts at HH25 to study the biomechanics of the whole heart (atria and ventricle). A novel and high-fidelity motion estimation technique, based on temporal motion model and non-rigid image registration algorithm, allowed automatic tracking of fluid-structure boundaries from scan images, and supported flow simulations. Results demonstrated that atrial appendages were the most contractile portion of the atria, having disproportionately high contribution to atrial blood pumping for its volume in the atria. However, the atria played a small role in blood pumping compared to the ventricle, as it had much lower ejection energy expenditure, and as the ventricle appeared to be able to draw inflow from the veins directly during late diastole. Spatially and temporally averaged wall shear stresses (WSS) for various cardiac structures were 0.062–0.068 Pa, but spatial-averaged WSS could be as high as 0.54 Pa in the RV. WSS was especially elevated at the atrial inlet, atrioventricular junction, regions near to the outflow tract, and at dividing lines between the left and right atrium and left and right side of the ventricle, where septation had begun and the lumen had narrowed. Elevated WSS could serve as biomechanics stimulation for proper growth and development.

Fast Adaptive Bases Algorithm for Non-rigid Image Registration

Fast Adaptive Bases Algorithm for Non-rigid Image Registration

NON-RIGID IMAGE REGISTRATION USING SELF-SUPERVISED FULLY CONVOLUTIONAL NETWORKS WITHOUT TRAINING DATA.

A novel non-rigid image registration algorithm is built upon fully convolutional networks (FCNs) to optimize and learn spatial transformations between pairs of images to be registered in a self-supervised learning framework. Different from most existing deep learning based image registration methods that learn spatial transformations from training data with known corresponding spatial transformations, our method directly estimates spatial transformations between pairs of images by maximizing an image-wise similarity metric between fixed and deformed moving images, similar to conventional image registration algorithms. The image registration is implemented in a multi-resolution image registration framework to jointly optimize and learn spatial transformations and FCNs at different spatial resolutions with deep self-supervision through typical feedforward and backpropagation computation. The proposed method has been evaluated for registering 3D structural brain magnetic resonance (MR) images and obtained better performance than state-of-the-art image registration algorithms.

Temporal Subtraction of Serial CT Images with Large Deformation Diffeomorphic Metric Mapping in the Identification of Bone Metastases

Purpose To determine the improvement of radiologist efficiency and performance in the detection of bone metastases at serial follow-up computed tomography (CT) by using a temporal subtraction (TS) technique based on an advanced nonrigid image registration algorithm. Materials and Methods This retrospective study was approved by the institutional review board, and informed consent was waived. CT image pairs (previous and current scans of the torso) in 60 patients with cancer (primary lesion location: prostate, n = 14; breast, n = 16; lung, n = 20; liver, n = 10) were included. These consisted of 30 positive cases with a total of 65 bone metastases depicted only on current images and confirmed by two radiologists who had access to additional imaging examinations and clinical courses and 30 matched negative control cases (no bone metastases). Previous CT images were semiautomatically registered to current CT images by the algorithm, and TS images were created. Seven radiologists independently interpreted CT image pairs to identify newly developed bone metastases without and with TS images with an interval of at least 30 days. Jackknife free-response receiver operating characteristics (JAFROC) analysis was conducted to assess observer performance. Reading time was recorded, and usefulness was evaluated with subjective scores of 1-5, with 5 being extremely useful and 1 being useless. Significance of these values was tested with the Wilcoxon signed-rank test. Results The subtraction images depicted various types of bone metastases (osteolytic, n = 28; osteoblastic, n = 26; mixed osteolytic and blastic, n = 11) as temporal changes. The average reading time was significantly reduced (384.3 vs 286.8 seconds; Wilcoxon signed rank test, P = .028). The average figure-of-merit value increased from 0.758 to 0.835; however, this difference was not significant (JAFROC analysis, P = .092). The subjective usefulness survey response showed a median score of 5 for use of the technique (range, 3-5). Conclusion TS images obtained from serial CT scans using nonrigid registration successfully depicted newly developed bone metastases and showed promise for their efficient detection. © RSNA, 2017 Online supplemental material is available for this article.

Parallel implementation of a nonrigid image registration algorithm for lung tumor boundary tracking in quasi real-time MRI.

This study presents an accelerated implementation of a two-dimensional moving mesh point correspondence algorithm using a GPU for tracking mobile tumor boundaries during radiation therapy. Normal CPU implementation of this algorithm is computationally intensive and time-consuming which limits its clinical utility, hence the need for a faster GPU implementation. One of the computationally intensive parts of the registration algorithm involves numerically solving a partial differential equation. In this paper we demonstrate that the computational performance of the algorithms can be improved by utilizing a shared memory implementation on the GPU. Evaluations in comparison to 600 manually drawn contours showed that the proposed GPU-based tracking of the tumor boundaries yielded similar level of accuracy as the CPU based approach with improved computational efficiency.

A GPU-Accelerated Deformable Image Registration Algorithm With Applications to Right Ventricular Segmentation

Delineation of the cardiac right ventricle is essential in generating clinical measurements such as ejection fraction and stroke volume. Given manual segmentation on the first frame, one approach to segment right ventricle from all of the magnetic resonance images is to find point correspondence between the sequence of images. Finding the point correspondence with non-rigid transformation requires a deformable image registration algorithm, which often involves computationally expensive optimization. The central processing unit (CPU)-based implementation of point correspondence algorithm has been shown to be accurate in delineating organs from a sequence of images in recent studies. The purpose of this study is to develop computationally efficient approaches for deformable image registration. We propose a graphics processing unit (GPU) accelerated approach to improve the efficiency. The proposed approach consists of two parallelization components: Parallel compute unified device architecture (CUDA) version of the deformable registration algorithm; and the application of an image concatenation approach to further parallelize the algorithm. Three versions of the algorithm were implemented: 1) CPU; 2) GPU with only intra-image parallelization (sequential image registration); and 3) GPU with inter and intra-image parallelization (concatenated image registration). The proposed methods were evaluated over a data set of 16 subjects. CPU, GPU sequential image, and GPU concatenated image methods took an average of 113.13, 16.50, and 5.96 s to segment a sequence of 20 images, respectively. The proposed parallelization approach offered a computational performance improvement of around $19\times $ in comparison to the CPU implementation while retaining the same level of segmentation accuracy. This paper demonstrated that the GPU computing could be utilized for improving the computational performance of a non-rigid image registration algorithm without compromising the accuracy.

Circumferential optical coherence tomography angiography imaging of the swine esophagus using a micromotor balloon catheter.

We demonstrate a micromotor balloon imaging catheter for ultrahigh speed endoscopic optical coherence tomography (OCT) which provides wide area, circumferential structural and angiographic imaging of the esophagus without contrast agents. Using a 1310 nm MEMS tunable wavelength swept VCSEL light source, the system has a 1.2 MHz A-scan rate and ~8.5 µm axial resolution in tissue. The micromotor balloon catheter enables circumferential imaging of the esophagus at 240 frames per second (fps) with a ~30 µm (FWHM) spot size. Volumetric imaging is achieved by proximal pullback of the micromotor assembly within the balloon at 1.5 mm/sec. Volumetric data consisting of 4200 circumferential images of 5,000 A-scans each over a 2.6 cm length, covering a ~13 cm(2) area is acquired in <18 seconds. A non-rigid image registration algorithm is used to suppress motion artifacts from non-uniform rotational distortion (NURD), cardiac motion or respiration. En face OCT images at various depths can be generated. OCT angiography (OCTA) is computed using intensity decorrelation between sequential pairs of circumferential scans and enables three-dimensional visualization of vasculature. Wide area volumetric OCT and OCTA imaging of the swine esophagus in vivo is demonstrated.

3D Non-rigid Image Registration Algorithm for MR-Guided Microwave Thermocoagulation of Liver Tumors

3D Non-rigid Image Registration Algorithm for MR-Guided Microwave Thermocoagulation of Liver Tumors

Evaluation of Non-Rigid Image-Registration Algorithms Using Discrepancy Distance Between Organ Contours

Purpose: Non-rigid image registration (NIR) is useful for adaptive radiotherapy. However, no method has been established for evaluating the quality of the algorithms used in NIR. To remedy this situation, we demonstrate herein a novel method to evaluate NIR algorithms. Methods: We define the NIR error as the discrepancy distance between (i) the organ contours obtained from computed tomography (CT) images acquired during the treatment period (reference contours) and (ii) the contours obtained from the treatment-planning CT images that are constructed by automated propagation during the treatment period (evaluation contours). However, the continuous positional relationship between the points where the reference contour intersects the evaluation contour is assumed to be maintained. In addition, we adapt the proposed method so that it can be applied to the contours of complex organs such as spherical and tubular organs. To demonstrate this method, we measure the contours of the prostate, right seminal vesicle, left seminal vesicle, urinary bladder, and rectum. The obtained NIR error presented in two-dimensional (2D) discrepancy maps. Results: The 2D discrepancy maps show the difference between the reference and evaluation contours from CT images. The proposed method measures the difference between the contours of spherical and tubular organs and evaluates the NIR error based on the positional relationship between the points constituting the contours. Conclusions: This study accounts for and measures the continuous positional relationship between corresponding points in the contours of complex-shaped spherical and tubular organs with irregularities and evaluates NIR algorithms based on these organ contours.

VLSI-Assisted Nonrigid Registration Using Modified Demons Algorithm

Increasing demand of high-speed portable modules for multimedia applications has motivated the development of hardware-based solutions for image processing applications. Most of the nonrigid image registration algorithms are found to be unsuitable for hardware implementation because of their nonlinearity and computationally intensive nature. In this paper, an algorithm for nonrigid image registration based on Demons approximation is proposed. The algorithm has been simulated in MATLAB and results show a 15% improvement in peak-signal-to-noise-ratio with a 17% reduction in registration time for $256 \times 256$ image over the original Demons algorithm. The proposed algorithm is synthesized in Virtex6-xc6vlx760-2-ff1760 and maximum synthesized frequency is found to be 174 MHz. The proposed architecture provides the low cost, high-speed solution for the registration process, which is also helpful for making a portable system.

Physical Constraint Finite Element Model for Medical Image Registration.

Due to being derived from linear assumption, most elastic body based non-rigid image registration algorithms are facing challenges for soft tissues with complex nonlinear behavior and with large deformations. To take into account the geometric nonlinearity of soft tissues, we propose a registration algorithm on the basis of Newtonian differential equation. The material behavior of soft tissues is modeled as St. Venant-Kirchhoff elasticity, and the nonlinearity of the continuum represents the quadratic term of the deformation gradient under the Green- St.Venant strain. In our algorithm, the elastic force is formulated as the derivative of the deformation energy with respect to the nodal displacement vectors of the finite element; the external force is determined by the registration similarity gradient flow which drives the floating image deforming to the equilibrium condition. We compared our approach to three other models: 1) the conventional linear elastic finite element model (FEM); 2) the dynamic elastic FEM; 3) the robust block matching (RBM) method. The registration accuracy was measured using three similarities: MSD (Mean Square Difference), NC (Normalized Correlation) and NMI (Normalized Mutual Information), and was also measured using the mean and max distance between the ground seeds and corresponding ones after registration. We validated our method on 60 image pairs including 30 medical image pairs with artificial deformation and 30 clinical image pairs for both the chest chemotherapy treatment in different periods and brain MRI normalization. Our method achieved a distance error of 0.320±0.138 mm in x direction and 0.326±0.111 mm in y direction, MSD of 41.96±13.74, NC of 0.9958±0.0019, NMI of 1.2962±0.0114 for images with large artificial deformations; and average NC of 0.9622±0.008 and NMI of 1.2764±0.0089 for the real clinical cases. Student’s t-test demonstrated that our model statistically outperformed the other methods in comparison (p-values <0.05).

Optimization of heart block in the left-sided whole breast radiation treatments.

Purpose: Blocks have been used to protect heart from potential radiation damage in left-sided breast treatments. Since cardiac motion pattern may not be fully captured on conventional 3DCT or 4DCT simulation scans, this study was intended to investigate the optimization of the heart block design taking the cardiac motion into consideration.Materials and Methods: Whole breast treatment plans using two opposed tangential fields were designed based on 4DCT simulation images for 10 left-sided breast cancer patients. Using an OBI system equipped to a Varian Linac, beam-eye viewed fluoroscopy images were acquired for each of the treatment beams after patient treatment setup, and the MLC heart blocks were overlaid onto the fluoroscopy images with an in-house software package. A non-rigid image registration and tracking algorithm was utilized to track the cardiac motion on the fluoroscopy images with minimal manual delineation for initialization, and the tracked cardiac motion information was used to optimize the heart block design to minimize the radiation damage to heart while avoiding the over-shielding that may lead to underdosing certain breast tissues.Results: Twenty-three sets of fluoroscopy images were acquired on 23 different days of treatment for the 10 patients. As expected, heart moved under the influences of both respiratory and cardiac motion. It was observed that for 16 out of the 23 treatments, heart moved beyond the planed heart block into treatment fields and MLC had to be adjusted to fully block heart. The adjustment was made for all but one patient. The number of the adjusted MLC leaves ranged from 1 to 16 (mean = 10), and the MLC leaf position adjustment ranged from 2 to 10 mm (mean = 6 mm). The added heart block areas ranged from 3 to 1230 mm2 (mean = 331 mm2).Conclusion: In left-sided whole breast radiation treatments, simulation CT (and 4DCT) based heart block design may not provide adequate heart protection for all the treatments. A fluoroscopy-based method has been developed to adaptively optimize the heart MLC block to achieve optimal heart protection.

Image-based motion correction for optical mapping of cardiac electrical activity.

Optical mapping, with membrane-bound, voltage-sensitive dyes, is widely used for in vitro recording of cardiac electrical activity. The spatial registration of such maps is lost when the heart moves with respect to a fixed photodetector array and contraction can generate substantial artifact if background fluorescence is not uniformly distributed. While motion artifact is commonly suppressed with electromechanical uncoupling agents, there are circumstances where these are undesirable. This study outlines a novel image-based approach for retrospective motion artifact correction. Isolated Langendorff-supported rat hearts (n=8), stained with di-4-ANEPPS, were illuminated at 516±14nm and fluorescent emission (>565±10nm) was acquired with a charge multiplying CCD camera. Background fluorescence was segmented in successive frames and stabilized using a non-rigid image registration algorithm. The resultant image deformation was used to estimate material point movement on the heart surface, so that total fluorescence could be mapped frame-by-frame to appropriate reference pixels. Finally, residual motion artifact was identified and removed. The effectiveness of this correction method was evaluated over 18 experimental datasets. Signal-to-noise ratio was increased more than fourfold, and activation time and action potential duration (APD) could be estimated at 24% more pixels than in the raw data. The variability of all APD measures was substantially reduced (i.e. APD50 estimated as 83.8±45.8ms before correction was 52.1±4.7ms afterward). This approach provides a robust means of recovering optical action potentials in the presence of substantial motion artifact.

Exact surface registration of retinal surfaces from 3-D optical coherence tomography images.

Nonrigid registration of optical coherence tomography (OCT) images is an important problem in studying eye diseases, evaluating the effect of pharmaceuticals in treating vision loss, and performing group-wise cross-sectional analysis. High dimensional nonrigid registration algorithms required for cross-sectional and longitudinal analysis are still being developed for accurate registration of OCT image volumes, with the speckle noise in images presenting a challenge for registration. Development of algorithms for segmentation of OCT images to generate surface models of retinal layers has advanced considerably and several algorithms are now available that can segment retinal OCT images into constituent retinal surfaces. Important morphometric measurements can be extracted if accurate surface registration algorithm for registering retinal surfaces onto corresponding template surfaces were available. In this paper, we present a novel method to perform multiple and simultaneous retinal surface registration, targeted to registering surfaces extracted from ocular volumetric OCT images. This enables a point-to-point correspondence (homology) between template and subject surfaces, allowing for a direct, vertex-wise comparison of morphometric measurements across subject groups. We demonstrate that this approach can be used to localize and analyze regional changes in choroidal and nerve fiber layer thickness among healthy and glaucomatous subjects, allowing for cross-sectional population wise analysis. We also demonstrate the method's ability to track longitudinal changes in optic nerve head morphometry, allowing for within-individual tracking of morphometric changes. This method can also, in the future, be used as a precursor to 3-D OCT image registration to better initialize nonrigid image registration algorithms closer to the desired solution.

Nonrigid medical image registration with locally linear reconstruction

Nonrigid image registration plays an important role in medical imaging systems. In complex circumstances, it is still a challenging task, especially when input images are corrupted by the spatially-varying intensity distortion. To address this difficulty, we propose a novel locally linear reconstruction based dissimilarity measurement (LLR). The core idea behind the LLR is that in each local region, the aligned floating image is linearly reconstructed by the reference image. Due to LLR, we present a multi-resolution nonrigid image registration algorithm. Extensive experiments on various simulated and real medical images have demonstrated that our algorithm outperforms the state-of-the-art approaches in registration accuracy. (C) 2014 Elsevier B.V. All rights reserved.

Medical image registration by combining global and local information: a chain-type diffeomorphic demons algorithm

The demons algorithm is a popular algorithm for non-rigid image registration because of its computational efficiency and simple implementation. The deformation forces of the classic demons algorithm were derived from image gradients by considering the deformation to decrease the intensity dissimilarity between images. However, the methods using the difference of image intensity for medical image registration are easily affected by image artifacts, such as image noise, non-uniform imaging and partial volume effects. The gradient magnitude image is constructed from the local information of an image, so the difference in a gradient magnitude image can be regarded as more reliable and robust for these artifacts. Then, registering medical images by considering the differences in both image intensity and gradient magnitude is a straightforward selection. In this paper, based on a diffeomorphic demons algorithm, we propose a chain-type diffeomorphic demons algorithm by combining the differences in both image intensity and gradient magnitude for medical image registration. Previous work had shown that the classic demons algorithm can be considered as an approximation of a second order gradient descent on the sum of the squared intensity differences. By optimizing the new dissimilarity criteria, we also present a set of new demons forces which were derived from the gradients of the image and gradient magnitude image. We show that, in controlled experiments, this advantage is confirmed, and yields a fast convergence.