A comparative study of Powell's and Downhill Simplex algorithms for a fast multimodal surface matching in brain imaging.

Multimodal images registration can be very helpful for diagnostic applications. However, even if a lot of registration algorithms exist, only a few really work in clinical routines. We developed a method based on surface matching and compared two minimization algorithms: Powell's and Downhill Simplex. We studied the influence of some factors (chamfer map computation, number and order of parameters to determine, minimization criteria) on the final accuracy of the algorithm. Using this comparison, we improved some processing steps to allow a clinical use, and selected the simplex algorithm which presented the best results.     

Robust rigid registration of retinal angiograms through optimization

Retinal fundus photographs are employed as standard diagnostic tools in ophthalmology. Serial photographs of the flow of fluorescein and indocyanine green (ICG) dye are used to determine the areas of the retinal lesions. For objective measurements of features, the registration of the images is a necessity. In this paper, we employ optimization techniques for registration with the help of 2-parameter translational motion model of retinal angiograms, based on non-linear pre-processing (Wiener filtering and morphological gradient) and computation of the similarity criteria for the alignment of the two gradient images for any given rigid transformation. The optimization methods are effectively employed to minimize the similarity criterion.

The presence of noise, the variations in the background and the temporal variation of the fluorescence level pose serious problems in obtaining a robust registration of the retinal images. Moreover, local search strategies are not robust in the case of ICG angiograms, even if one uses a multiresolution approach.

The present work makes a systematic comparison of different optimization techniques, namely the minimization method derived from the optical flow formulation, the Nelder-Mead local search and the HCIAC ant colony metaheuristic, each optimizing a similarity criterion for the gradient images. The impact of the resolution and median filtering of gradient image is studied and the robustness of the approaches is tested through experimental studies, performed on macular fluorescein and ICG angiographies.

Our proposed optimization techniques have shown interesting results especially for high resolution difficult registration problems. Moreover, this approach seems promising for affine (6-parameter motion model) or elastical registrations.     

Digital subtraction CT angiography based on efficient 3D registration and refinement.

A novel method for fast, automatic 3D digital subtraction CT angiography (DS-CTA) is presented to generate artifact-free angiograms. The proposed method consists of two steps: 3D registration to align a CT image to the CT angiography (CTA) image and subtraction-and-refinement to extract blood vessels only. For efficient and accurate 3D registration in the first step, an normalized mutual information (NMI) based algorithm is adopted, and its fast version is developed by introducing a new measure. To further improve the subtracted image quality in the second step, a novel 3D refinement algorithm is suggested to effectively remove unwanted residuals. Experimental results of seven clinical CT/CTA head datasets demonstrate that cerebral vessels are well extracted from CTA images with almost no loss. The typical processing time is 3-9 min depending on the image size in a PC with a 2.4 GHz CPU.     

Automatic point correspondence using an artificial immune system optimization technique for medical image registration

In this paper, an automatic method for determining pairs of corresponding points between medical images is proposed. The method is based on the implementation of an artificial immune system (AIS). AIS is a relatively novel, population based category of algorithms, inspired by theoretical immunologic models. When used as function optimizers, AIS have the attractive property of locating the global optimum of a function as well as a large number of strong local optimum points. In this work, AIS has been applied both for the extraction of an optimal set of candidate points on the reference image and the definition of their corresponding ones on the second image. The performance of the proposed AIS algorithm is evaluated against the widely used Iterative Closest Point (ICP) algorithm in terms of the accuracy of the obtained correspondences and in terms of the accuracy of the point-based registration by the two correspondence algorithms and the Mutual Information criterion, as an intensity-based registration method. Qualitative and quantitative results involving 92 X-ray dental and 10 retinal image pairs subject to known and unknown transformations are presented. The results indicate a superior performance of the proposed AIS algorithm with respect to the ICP algorithm and the Mutual Information, in terms of both correct correspondence and registration accuracy.     

Comprehensive approach for correction of motion and distortion in diffusion-weighted MRI.

Patient motion and image distortion induced by eddy currents cause artifacts in maps of diffusion parameters computed from diffusion-weighted (DW) images. A novel and comprehensive approach to correct for spatial misalignment of DW imaging (DWI) volumes acquired with different strengths and orientations of the diffusion sensitizing gradients is presented. This approach uses a mutual information-based registration technique and a spatial transformation model containing parameters that correct for eddy current-induced image distortion and rigid body motion in three dimensions. All parameters are optimized simultaneously for an accurate and fast solution to the registration problem. The images can also be registered to a normalized template with a single interpolation step without additional computational cost. Following registration, the signal amplitude of each DWI volume is corrected to account for size variations of the object produced by the distortion correction, and the b-matrices are properly recalculated to account for any rotation applied during registration. Both qualitative and quantitative results show that this approach produces a significant improvement of diffusion tensor imaging (DTI) data acquired in the human brain.     

Dynamic scintigraphy: Calculation and imaging of regional distribution of quantitative parameters

The theory and general method of calculation of the parametric (functional) images at dynamic studies is described briefly. This method is illustrated by a general program DYNAM. PARAM. PRESENT. created for Clincom apparatus.Moreover, the selection of parameters and algorithms for calculation, the problems associated with the filtration of data, the correction for dead time and for non-homogeneity are discussed. In addition to the general method, there are presented here the possible ways of application of the simplified methods of construction of the parametric images by means of algebraic operations between images in special cases.Some aspects of that method are completed by examples of practical parametric images of some organs.     

Global optimization of mutual information: application to three-dimensional retrospective registration of magnetic resonance images.

A global optimization technique for image registration, based on mutual information, that can be used in conjunction with a multi-resolution paradigm is described. This technique combines genetic algorithm in continuous space, which is a stochastic method and is very efficient in large search space, with dividing rectangle, which is a deterministic method that theoretically guarantees global optimization and is efficient in small search space. Calculations were performed for determining the optimum parameters for implementing this method. This technique was applied to register magnetic resonance images of brain. For comparison, the registration results using AIR, a commonly employed software package, are presented.     

Quantitative and clinical analysis of SPECT image registration for epilepsy studies.

This study reports quantitative measurements of the accuracy of two popular voxel-based registration algorithms--Woods' automated image registration algorithm and mutual information correlation--and compares these with conventional surface matching (SM) registration. METHODS: The registration algorithms were compared (15 different matches each) for (a) three-dimensional brain phantom images, (b) an ictal SPECT image from a patient with partial epilepsy matched to itself after modification to simulate changes in the cerebral blood flow pattern and (c) ictal/interictal SPECT images from 15 patients with partial epilepsy. Blinded visual ranking and localization of the subtraction images derived from the patient images were also performed. RESULTS: Both voxel-based registration methods were more accurate than SM registration (P < 0.0005). Automated image registration algorithm was more accurate than mutual information correlation for the computer-simulated ictal/interictal images and the patient ictal/interictal studies (P < 0.05). The subtraction SPECTs from SM were poorer in visual ranking more often than the voxel-based methods (P < 0.05). CONCLUSION: Voxel intensity-based registration algorithms provide significant improvement in ictal/interictal SPECT registration accuracy and result in a clinically detectable improvement in the subtraction SPECT images.     

Registration, segmentation, and visualization of multimodal brain images.

This paper gives an overview of the studies performed at our institute over the last decade on the processing and visualization of brain images, in the context of international developments in the field. The focus is on multimodal image registration and multimodal visualization, while segmentation is touched upon as a preprocessing step for visualization. The state-of-the-art in these areas is discussed and suggestions for future research are given.     

Design and construction of a brain phantom to simulate neonatal MR images

This paper presents the design and construction of a 3D digital neonatal neurocranial phantom and its application for the simulation of brain magnetic resonance (MR) images. Commonly used digital brain phantoms (e.g. BrainWeb) are based on the adult brain. With the growing interest in computer-aided methods for neonatal MR image processing, there is a growing demand a digital phantom and brain MR image simulator especially for the neonatal brains. This is due to the pronounced differences between adult and neonatal brains not only in terms of size but also, more importantly, in terms of geometrical proportions and the need to subdivide white matter into two different tissue types in neonates. Therefore the neonatal brain phantom created in the here presented work consists of 9 different tissue types: skin, fat, muscle, skull, dura mater, gray matter, myelinated white matter, nonmyelinated white matter and cerebrospinal fluid. Each voxel has a vector consisting of 9 components, one for each of these nine tissue types. This digital phantom can be used to map simulated magnetic resonance signal intensities resulting in simulated MR images of the newborns head. These images with controlled degradation of the image data present a representative, reproducible data set ideal for development and evaluation of neonatal MRI analysis methods, e.g. segmentation and registration algorithms.     

Real-time imaging of two-dimensional cardiac strain using a harmonic phase magnetic resonance imaging (HARP-MRI) pulse sequence.

The harmonic phase (HARP) method provides automatic and rapid analysis of tagged magnetic resonance (MR) images for quantification and visualization of myocardial strain. In this article, the development and implementation of a pulse sequence that acquires HARP images in real time are described. In this pulse sequence, a CINE sequence of images with 1-1 spatial modulation of magnetization (SPAMM) tags are acquired during each cardiac cycle, alternating between vertical and horizontal tags in successive heartbeats. An incrementing train of imaging RF flip angles is used to compensate for the decay of the harmonic peaks due to both T(1) relaxation and the applied imaging pulses. The magnitude images displaying coarse anatomy are automatically reconstructed and displayed in real time after each heartbeat. HARP strain images are generated offline at a rate of four images per second; real-time processing should be possible with faster algorithms or computers. A comparison of myocardial contractility in non-breath-hold and breath-hold experiments in normal humans is presented.     

Improved image registration by using fourier interpolation

This paper presents a technique for performing two-dimensional rigid-body image registration for functional magnetic resonance images (fMRI). The method provides accurate motion correction without local distortion. The approach is to perform the translation and rotation in the Fourier domain. For images sampled on a grid, such as in echo-planar imaging (EPI), one potential stumbling block to this approach is the computational burden of reconstruction, since the rotated image will no longer be on the Cartesian grid. A method of approximating rotations via local translations (shearing) is presented, which keeps the data on the Cartesian grid. This can provide quite accurate approximations with only a moderate amount of computation. A mean squared error (MSE) criterion is used for determining the registration parameters. This method is tested on several sets of simulated images and shown to have an accuracy ranging from 0.02 to 0.3 pixels for images with SNRs ranging from 100 to 10, respectively. They techniques have been tested on several sets of images. They are shown to work well on real subjects, for both echo-planar and spiral data acquisition schemes. The techniques are used in an activation study in which the subject moved his head during image collection. After use of this registration technique, the activation is easily detected.     

Implementation of a digital image superposition algorithm for radionuclide images: an assessment of its accuracy and reproducibility

An operator-interactive algorithm to achieve superposition of organ images has been used with a dedicated nuclear medicine computer system. Its purpose is to achieve organ registration in 128 X 128 digitized images before a direct numerical comparison of the regional distribution of a deposited radiotracer is performed. The accuracy and reproducibility of the algorithm for myocardial images has been tested by four operators, using a set of 28 image pairs in which the relative position of the heart differed by more than 10 mm for each pair. Comparing their results with the known displacements on two occasions provided an assessment of these two important parameters. The accuracy and reproducibility for superposing myocardial images by this digital technique are found to be well within the spatial resolution (FWHM) of the imaging system of the Tl-201 tracer studied.     

Evaluation of Four Volume-Based Image Registration Algorithms

We evaluated 4 volume-based automatic image registration algorithms from 2 commercially available treatment planning systems (Philips Syntegra and BrainScan). The algorithms based on cross correlation (CC), local correlation (LC), normalized mutual information (NMI), and BrainScan mutual information (BSMI) were evaluated with: (1) the synthetic computed tomography (CT) images, (2) the CT and magnetic resonance (MR) phantom images, and (3) the CT and MR head image pairs from 12 patients with brain tumors. For the synthetic images, the registration results were compared with known transformation parameters, and all algorithms achieved accuracy of submillimeter in translation and subdegree in rotation. For the phantom images, the registration results were compared with those provided by frame and marker-based manual registration. For the patient images, the results were compared with anatomical landmark–based manual registration to qualitatively determine how the results were close to a clinically acceptable registration. NMI and LC outperformed CC and BSMI, with the sense of being closer to a clinically acceptable result. As for the robustness, NMI and BSMI outperformed CC and LC. A guideline of image registration in our institution was given, and final visual assessment is necessary to guarantee reasonable results.     

Evaluation of Four Volume-Based Image Registration Algorithms

We evaluated 4 volume-based automatic image registration algorithms from 2 commercially available treatment planning systems (Philips Syntegra and BrainScan). The algorithms based on cross correlation (CC), local correlation (LC), normalized mutual information (NMI), and BrainScan mutual information (BSMI) were evaluated with: (1) the synthetic computed tomography (CT) images, (2) the CT and magnetic resonance (MR) phantom images, and (3) the CT and MR head image pairs from 12 patients with brain tumors. For the synthetic images, the registration results were compared with known transformation parameters, and all algorithms achieved accuracy of submillimeter in translation and subdegree in rotation. For the phantom images, the registration results were compared with those provided by frame and marker-based manual registration. For the patient images, the results were compared with anatomical landmark–based manual registration to qualitatively determine how the results were close to a clinically acceptable registration. NMI and LC outperformed CC and BSMI, with the sense of being closer to a clinically acceptable result. As for the robustness, NMI and BSMI outperformed CC and LC. A guideline of image registration in our institution was given, and final visual assessment is necessary to guarantee reasonable results.     

Image content extraction: application to MR images of the brain.

A system for automatically extracting image content features was developed that combines registration to a labeled atlas with natural language processing of free-text radiology reports. The system was then tested with T1-weighted, spoiled gradient-echo magnetic resonance (MR) imaging studies of the brain performed in nine patients. The locations of 599 structures were visually assessed by an experienced radiologist and compared with the locations indicated by automated output. The in-plane accuracy of the contours was subjectively evaluated as either good, moderate, or poor. The criterion for classifying a structure as correctly located was that 90% or more of all the images containing the structure had to be correctly identified. For 98% of the structures, the images identified by the automated algorithm agreed with those identified by the radiologist, and in 83% of cases, image contours showed a good in-plane overlap. The results of this validation study demonstrate that this combination of registration and natural language processing is accurate in identifying relevant images from brain MR imaging studies. However, the range of applicability of this technique has yet to be determined by applying the technique to a large number of studies.     

Registration of emission and transmission whole-body scintillation-camera images.

In this work, a method for registration of whole-body (WB) scintillation-camera images is presented. The primary motive for the development is to perform activity quantification using the conjugate view method on an image basis. Accurate image registration is required for sequential anterior and posterior scans, for serial emission images for analysis of the biokinetics, and for transmission and emission images for a pixel-based attenuation correction. METHODS: Registration is performed by maximization of the mutual information. The spatial transformation has been tailored for the registration of WB images and is composed of global and local transformations, including rigid, projective, and curved transformations. A coarse registration is first performed using cross-correlation and direct pixel scaling. Optimization is then performed in a sequence, beginning with the 2 legs independently, followed by the upper body and head. Evaluation is performed for clinical images of an (131)I-labeled monoclonal antibody and for Monte Carlo-simulated images. An anthropomorphic WB computer phantom, which has been especially modified to match the patient position during WB scanning, is used for the simulations. RESULTS: For simulated images, registration errors are within 1 pixel (<3.6 mm) for a sufficient image count level. Separate evaluation of the influence of noise shows that the errors increase below a total image count of approximately 10(5) (signal-to-noise ratio, approximately 4). For clinical evaluations, the deviations between point markers are 9 +/- 5 mm. CONCLUSION: An automatic registration method for WB images has been developed, which is applicable to emission-emission and transmission-emission registration. This method has been applied in more than 50 clinical studies and has shown to be robust and reliable.     

An image registration strategy for multi-echo fMRI.

Functional magnetic resonance imaging (fMRI) based on multiple-echo T(2)* mapping has attracted much attention recently. The contrasts in the parametric T(2)* maps are usually too low to allow direct image registration. In this study, an image registration strategy has been proposed for single-shot multi-echo data sets acquired for dynamic T(2)* mapping. We performed image registration of the T(2)*-weighted images before the calculation of the T(2)* parameter maps using two different strategies. One is to perform separate image registration on each echo and the other is to use the same motion correction parameters extracted from the second echo for all the data. Both strategies increase the number of activated voxels and reduce the effective noise level. The results also indicate that, for a single-shot dual-echo image data set, it is slightly preferable to use the second echo for direct image registration and then apply the same motion correction parameters to the first echo images. J. Magn. Reson Imaging 1999;10:154-158.     

Computer vision guided virtual craniofacial reconstruction.

The problem of virtual craniofacial reconstruction from a sequence of computed tomography (CT) images is addressed and is modeled as a rigid surface registration problem. Two different classes of surface matching algorithms, namely the data aligned rigidity constrained exhaustive search (DARCES) algorithm and the iterative closest point (ICP) algorithm are first used in isolation. Since the human bone can be reasonably approximated as a rigid body, 3D rigid surface registration techniques such as the DARCES and ICP algorithms are deemed to be well suited for the purpose of aligning the fractured bone fragments. A synergistic combination of these two algorithms, termed as the hybrid DARCES-ICP algorithm, is proposed. The hybrid algorithm is shown to result in a more accurate mandibular reconstruction when compared to the individual algorithms used in isolation. The proposed scheme for virtual reconstructive surgery would prove to be of tremendous benefit to the operating surgeons as it would allow them to pre-visualize the reconstructed mandible (i.e., the end-product of their work), before performing the actual surgical procedure. Experimental results on both phantom and real (human) patient datasets are presented.     

Automated image registration of gated cardiac single-photon emission computed tomography and magnetic resonance imaging

PURPOSE: To develop an automatic registration method for electrocardiogram-gated myocardial perfusion single-photon emission computed tomography (SPECT) and cardiac cine-magnetic resonance imaging (MRI). MATERIALS AND METHODS: Paired myocardial perfusion SPECT (MPS) and MRI from 20 patients were considered. MR images were presegmented by heart localization based on detection of cardiac motion and optimal thresholding. A registration algorithm based on mutual information was subsequently applied to all time frames or a selected subset from both modalities. RESULTS: A preprocessing step significantly improved the accuracy of the registration when compared to automatic registration performed without preprocessing. Errors in translation parameters (T(x), T(y), T(z)) averaged (1.0 +/- 1.5, 1.1 +/- 1.3, 0.9 +/- 0.9) pixels with MRI segmentation and (4.6 +/- 3.2, 3.4 +/- 2.6, 3.0 +/- 3.4) pixels without MRI segmentation. Errors in rotation parameters (R(x), R(y), R(z)) averaged (5.4 +/- 2.9, 3.4 +/- 2.7, 4.5 +/- 3.6) degrees with MRI segmentation and (9.3 +/- 6.1, 4.8 +/- 4.3, 14.6 +/- 12.6) degrees without MRI segmentation. Error was calculated as the absolute difference between the expert manual and the automatic registration transformation. CONCLUSION: Automatic registration of gated MPS and cine MRI is possible with the use of a mutual information-based technique when MR images are presegmented. Cardiac motion can be used to isolate the left ventricle (LV) on the MR images automatically, and subsequently the segmented MR images can be coregistered with gated MPS.