Joe P. Windham

Eigenimage filtering in MR imaging.

This article presents the technical aspects of a linear filter, referred to as eigenimage filtering, and its applications in magnetic resonance (MR) imaging. The technique is used to obtain a single composite image depicting a particular feature of interest while suppressing one or more interfering features. The appropriate weighting components to be used in the linear filter are determined on the criterion that the desired feature is enhanced while the interfering features are suppressed. The criterion is expressed mathematically as a ratio. By applying Rayleighs principle, the ratio is maximized by finding the eigenvector associated with the maximum eigenvalue of the corresponding generalized eigenvalue problem. The appropriate weighting factors for the linear filter are the elements of the eigenvector which maximize the ratio. The utilization of the technique is demonstrated in its application to a simulated MR image sequence as well as to acquired MR image sequences of a normal and an abnormal brain. Index Terms: Magnetic resonance imaging, physics and instrumentation—Magnetic resonance imaging, techniques—Eigenimage filtering.

Time course of ADCW changes in ischemic stroke : Beyond the human eye!

BACKGROUND AND PURPOSE Using newly developed computerized image analysis, we studied the heterogeneity of apparent diffusion coefficient of water (ADCw) values in human ischemic stroke within 10 hours of onset. METHODS Echo-planar trace diffusion-weighted images from 9 patients with focal cortical ischemic stroke were obtained within 10 hours of symptom onset. An Iterative Self-Organizing Data Analysis (ISODATA) clustering algorithm was implemented to segment different tissue types with a series of DW images. ADCw maps were calculated from 4 DW images on a pixel-by-pixel basis. The segmented zones within the lesion were characterized as low, pseudonormal, or high, expressed as a ratio of the mean+/-SD of ADCw of contralateral noninvolved tissue. RESULTS The average ADCW in the ischemic stroke region within 10 hours of onset was significantly depressed compared with homologous contralateral tissue (626.6+/-76.8 versus 842.9+/-60.4x10(-6) mm2/s; P<0.0001). Nevertheless, ISODATA segmentation yielded multiple zones within the stroke region that were characterized as low, pseudonormal, and high. The mean proportion of low:pseudonormal:high was 72%:20%:8%. CONCLUSIONS Despite low average ADCW, computer-assisted segmentation of DW MRI detected heterogeneous zones within ischemic lesions corresponding to low, pseudonormal, and high ADCw not visible to the human eye. This supports acute elevation of ADCw in human ischemic stroke and, accordingly, different temporal rates of tissue evolution toward infarction.

Registration and warping of magnetic resonance images to histological sections.

We present a method for coregistration and warping of magnetic resonance images (MRI) to histological sections for comparison purposes. This methodology consists of a modified head and hat surface-based registration algorithm followed by a new automated warping approach using nonlinear thin plate splines to compensate for distortions between the data sets. To test the methodology, 15 male Wistar rats were subjected to focal cerebral ischemia via permanent occlusion of the middle cerebral artery. The MRI images were acquired in separate groups of animals at 16-24 h (n = 9) and 48-168 h (n = 6) postocclusion. After imaging, animals were immediately sacrificed and hematoxylin- and eosin-stained brain sections were obtained for histological analysis. The MRI was coregistered and warped to histological sections. The MRI lesion areas were defined using the Eigenimage (EI) filter technique. The EI is a linear filter that maximizes the projection of a desired tissue (ischemic tissue) while it minimizes the projection of undesired tissues (nonischemic tissue) onto a composite image called an EI. When using coregistration without warping the MRI lesion area demonstrated poor correlation (r = 0.55, p > 0.01) with a percent difference between the two lesion areas of 22.5% +/- 10.8%. After warping, the MRI and histology had significant correlation (r = 0.97, p < 0.01) and a decreased percent difference of 5.56% +/- 4.31%. This methodology is simple and robust for coregistration and warping of MRI to histological sections and can be utilized in many applications for comparison of MRI to histological data.

A comparative analysis of several transformations for enhancement and segmentation of magnetic resonance image scene sequences

The performance of the eigenimage filter is compared with those of several other filters as applied to magnetic resonance image (MRI) scene sequences for image enhancement and segmentation. Comparisons are made with principal component analysis, matched, modified-matched, maximum contrast, target point, ratio, log-ratio, and angle image filters. Signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), segmentation of a desired feature (SDF), and correction for partial volume averaging effects (CPV) are used as performance measures. For comparison, analytical expressions for SNRs and CNRs of filtered images are derived, and CPV by a linear filter is studied. Properties of filters are illustrated through their applications to simulated and acquired MRI sequences of a phantom study and a clinical case; advantages and weaknesses are discussed. The conclusion is that the eigenimage filter is the optimal linear filter that achieves SDF and CPV simultaneously.

A multidimensional nonlinear edge-preserving filter for magnetic resonance image restoration

The paper presents a multidimensional nonlinear edge-preserving filter for restoration and enhancement of magnetic resonance images (MRI). The filter uses both interframe (parametric or temporal) and intraframe (spatial) information to filter the additive noise from an MRI scene sequence. It combines the approximate maximum likelihood (equivalently, least squares) estimate of the interframe pixels, using MRI signal models, with a trimmed spatial smoothing algorithm, using a Euclidean distance discriminator to preserve partial volume and edge information. (Partial volume information is generated from voxels containing a mixture of different tissues.) Since the filters structure is parallel, its implementation on a parallel processing computer is straightforward. Details of the filter implementation for a sequence of four multiple spin-echo images is explained, and the effects of filter parameters (neighborhood size and threshold value) on the computation time and performance of the filter is discussed. The filter is applied to MRI simulation and brain studies, serving as a preprocessing procedure for the eigenimage filter. (The eigenimage filter generates a composite image in which a feature of interest is segmented from the surrounding interfering features.) It outperforms conventional pre and post-processing filters, including spatial smoothing, low-pass filtering with a Gaussian kernel, median filtering, and combined vector median with average filtering.

Segmentation of the hippocampus from brain MRI using deformable contours

The application of a discrete dynamic contour model for segmentation of the hippocampus from brain MRI has been investigated. Solutions to several common problems of dynamic contours in this case and similar cases have been developed. A new method for extracting the discontinuous boundary of a structure with multiple edges near the structure has been developed. The method is based on detecting and following edges by external forces. The reliability of the final contour and the model stability have been improved by using a continuous mapping of the external energy and limiting movements of the contour. The problem of optimizing the internal force weight has been overcome by making it dependent on the amount of the external force. Finally, the results of applying the proposed algorithm, which implements the above modifications, to multiple applications have been evaluated.

Optimal linear transformation for MRI feature extraction

This paper presents development and application of a feature extraction method for magnetic resonance imaging (MRI), without explicit calculation of tissue parameters. A three-dimensional (3-D) feature space representation of the data is generated in which normal tissues are clustered around prespecified target positions and abnormalities are clustered elsewhere. This is accomplished by a linear minimum mean square error transformation of categorical data to target positions. From the 3-D histogram (cluster plot) of the transformed data, clusters are identified and regions of interest (ROIs) for normal and abnormal tissues are defined. These ROIs are used to estimate signature (prototype) vectors for each tissue type which in turn are used to segment the MRI scene. The proposed feature space is compared to those generated by tissue-parameter-weighted images, principal component images, and angle images, demonstrating its superiority for feature extraction and scene segmentation. Its relationship with discriminant analysis is discussed. The method and its performance are illustrated using a computer simulation and MRI images of an egg phantom and a human brain.

Optimal transformation for correcting partial volume averaging effects in magnetic resonance imaging

Segmentation of a feature of interest while correcting for partial volume averaging effects is a major tool for identification of hidden abnormalities, fast and accurate volume calculation, and three-dimensional visualization in the field of magnetic resonance imaging (MRI). The authors discuss the optimal transformation for simultaneous segmentation of a desired feature and correction of partial volume averaging effects while maximizing the signal-to-noise ratio (SNR) of the desired feature. It is proved that correction of partial volume averaging effects requires the removal of the interfering features from the scene. It is also proved that correction of partial volume averaging effects can be achieved merely by a linear transformation. It is shown that the optimal transformation matrix is easily obtained using the Gram-Schmidt orthogonalization procedure which is numerically stable. Applications of the technique to MRI simulation, phantom, and brain images are shown. It is shown that in all cases the desired feature is segmented from the interfering features and partial volume information is visualized in the resulting transformed images. >

A fast and accurate algorithm for volume determination in MRI.

The determination of volumes in clinical MRI studies are prohibitive because of the time required to compute an accurate volume. Techniques that speed up the calculation are prone to large errors which make most impractical for an accurate diagnosis. A linear filter, called the eigenimage filter, has been developed that separates a desired feature from other features which interfere with its observation in an image. Using the images produced by this technique (eigenimages), the amount of operator interaction required to calculate volumes are significantly reduced. The technique also has the ability to correct for partial volume averaging effects and as a result a more accurate volume can be determined. The technique was applied to a computer simulation and two phantom studies. The time required to calculate the volume was less than 1 min per slice and the errors in accuracy and reproducibility were less than 2% for all studies.

A 3D deformable surface model for segmentation of objects from volumetric data in medical images.

In this paper we present a new 3D discrete dynamic surface model. The model consists of vertices and edges, which connect adjacent vertices. Basic geometry of the model surface is generated by triangle patches. The model deforms by internal and external forces. Internal forces are obtained from local geometry of the model and are related to the local curvature of the surface. External forces, on the other hand, are based on the image data and are calculated from desired image features. We also present a method for generating an initial volume for the model from a stack of initial contours, drawn by the user on cross sections of the volumetric data.