The utility of principal component analysis for the image display of brain lesions. A preliminary, comparative study

High dynamic range processing for magnetic resonance imaging

Purpose

To minimize feature loss in T₁- and T₂-weighted MRI by merging multiple MR images acquired at different T_R and T_E to generate an image with increased dynamic range.

Materials and Methods

High Dynamic Range (HDR) processing techniques from the field of photography were applied to a series of acquired MR images. Specifically, a method to parameterize the algorithm for MRI data was developed and tested. T₁- and T₂-weighted images of a number of contrast agent phantoms and a live mouse were acquired with varying T_R and T_E parameters. The images were computationally merged to produce HDR-MR images. All acquisitions were performed on a 7.05 T Bruker PharmaScan with a multi-echo spin echo pulse sequence.

Results

HDR-MRI delineated bright and dark features that were either saturated or indistinguishable from background in standard T₁- and T₂-weighted MRI. The increased dynamic range preserved intensity gradation over a larger range of T₁ and T₂ in phantoms and revealed more anatomical features in vivo.

Conclusions

We have developed and tested a method to apply HDR processing to MR images. The increased dynamic range of HDR-MR images as compared to standard T₁- and T₂-weighted images minimizes feature loss caused by magnetization recovery or low SNR.

PCATMIP: enhancing signal intensity in diffusion-weighted magnetic resonance imaging

Diffusion-weighted MRI studies generally lose signal intensity to physiological motion, which can adversely affect quantification/diagnosis. Averaging over multiple repetitions, often used to improve image quality, does not eliminate the signal loss. In this article, PCATMIP, a combined principal component analysis and temporal maximum intensity projection approach, is developed to address this problem. Data are first acquired for a fixed number of repetitions. Assuming that physiological fluctuations of image intensities locally are likely temporally correlated unlike random noise, a local moving boxcar in the spatial domain is used to reconstruct low-noise images by considering the most relevant principal components in the temporal domain. Subsequently, a temporal maximum intensity projection yields a high signal-intensity image. Numerical and experimental studies were performed for validation and to determine optimal parameters for increasing signal intensity and minimizing noise. Subsequently, a combined principal component analysis and temporal maximum intensity projection approach was used to analyze diffusion-weighted porcine liver MRI scans. In these scans, the variability of apparent diffusion coefficient values among repeated measurements was reduced by 59% relative to averaging, and there was an increase in the signal intensity with higher intensity differences observed at higher b-values. In summary, a combined principal component analysis and temporal maximum intensity projection approach is a postprocessing approach that corrects for bulk motion-induced signal loss and improves apparent diffusion coefficient measurement reproducibility.

Quantitative segmentation of principal carotid atherosclerotic lesion components by feature space analysis based on multicontrast MRI at 1.5 T

The purpose of this paper is to evaluate the capability of feature space analysis (FSA) for quantifying the relative volumes of principal components (thrombus, calcification, fibrous, normal intima, and lipid) of atherosclerotic plaque tissue in multicontrast magnetic resonance images (mc-MRI) acquired in a setup resembling clinical conditions ex vivo. Utilizing endogenous contrast, proton density, T1-weighted, and T2-weighted images were acquired for 13 carotid endarterectomy (CEA) tissues under near-clinical conditions (human 1.5 T GE Excite scanner with sequence parameters comparable to an in vivo acquisition). An FSA algorithm was utilized to segment and quantify the principal components of atherosclerotic plaques. Pilot in vivo mc-MRI images were analyzed in the same way as the ex vivo images for exploring the possible adaptation of this technique to in vivo imaging. Relative abundance of principal plaque components in CEA tissues as determined by mc-MRI/FSA were compared to those measured by histology. Mean differences +/- standard deviations were 5.8 +/- 4.1% for thrombus, 1.5 +/-1.4 % for calcification, 4.0 +/-2.8% for fibrous, 8.2 +/- 10% for normal intima, and 2.4 +/- 2.2% for lipid. Reasonable quantitative agreement between the classification results obtained with FSA and histological data was obtained for near-clinical imaging conditions. Combination of mc-MRI and FSA may have an application for determining atherosclerotic lesion composition and monitoring treatment in vivo.

Multispectral image visualization through first-order fusion

We present a new formalism for the treatment and understanding of multispectral images and multisensor imagery based on first-order contrast information. Although little attention has been paid to the utility of multispectral contrast, we develop a theory for multispectral contrast that enables us to produce an optimal grayscale visualization of the first-order contrast of an image with an arbitrary number of bands. We demonstrate how our technique can reveal significantly more interpretive information to an image analyst, who can use it in a number of image understanding algorithms. Existing grayscale visualization strategies are reviewed. A variety of experimental results are presented to support the performance of the new method.

Feature space analysis: effects of MRI protocols

We present a method for exploring the relationship between the image segmentation results obtained by an optimal feature space method and the MRI protocols used. The steps of the work accomplished are as follows. (1) Patients with brain tumors were imaged on a 1.5 T General Electric Signa MRI System using multiple protocols (T1 and T2-weighted fast spin-echo and FLAIR). T1-weighted images were acquired before and after gadolinium injection. (2) Image volumes were co-registered, and images of a slice through the center of the tumor were selected for processing. (3) For each patient, several image sets were defined by selecting certain MR images (e.g., 4T2's+ IT1, 4T2's+FLAIR, 2T2's+ 1T1). (4) Using each image set, the optimal feature space was generated and images were segmented into normal tissues and different tumor zones. (5) Segmentation results obtained using the different MRI sets were compared. Based on the analysis results from 27 image sets, we found that the locations of the clusters for the tumor zones and their corresponding regions in the image domain changed as a function of the MR images (MRI protocols) used. However, the segmentation results for the total lesion and normal tissues remained relatively unchanged.

Feature space analysis of MRI

This paper presents an MRI feature-space image-analysis method and its application to brain tumor studies. The proposed method generates a transformed feature space in which the normal tissues (white matter, gray matter, and CSF) become orthonormal. As such, the method is expected to have site-to-site and patient-to-patient consistency, and is useful for identification of tissue types, segmentation of tissues, and quantitative measurements on tissues. The steps of the work accomplished are as follows: (1) Four T2-weighted and two T1-weighted images (before and after injection of gadolinium) were acquired for 10 tumor patients. (2) Images were analyzed by an image analyst according to the proposed algorithm. (3) Biopsy samples were extracted from each patient and were subsequently analyzed by the pathology laboratory. (4) Image-analysis results were compared with the biopsy results. Pre- and postsurgery feature spaces were also compared. The proposed method made it possible to visualize the MRI feature space and to segment the image. In all cases, the operators were able to find clusters for normal and abnormal tissues. Also, clusters for different zones of the tumor were found. The method successfully segmented the image into normal tissues (white matter, gray matter, and CSF) and different zones of the lesion (tumor, cyst, edema, radiation necrosis, necrotic core, and infiltrated tumor). The results agreed with those obtained from the biopsy samples. Comparison of pre- with postsurgery and radiation feature spaces illustrated that the original solid tumor was not present in the second study, but a new tissue component appeared in a different location of the feature space. This tissue could be radiation necrosis generated as a result of radiation.

Multivariate cluster analysis of dynamic iodine-123 iodobenzamide SPET dopamine D2 receptor images in schizophrenia

This paper describes the application of a multivariate statistical technique to investigate striatal dopamine D2 receptor concentrations measured by iodine-123 iodobenzamide (123I-IBZM) single-photon emission tomography (SPET). This technique enables the automatic segmentation of dynamic nuclear medicine images based on the underlying time-activity curves present in the data. Once the time-activity curves have been extracted, each pixel can be mapped back on to the underlying distribution, considerably reducing image noise. Cluster analysis has been verified using computer simulations and phantom studies. The technique has been applied to SPET images of dopamine D2 receptors in a total of 20 healthy and 20 schizophrenic volunteers (22 male, 18 female), using the ligand 123I-IBZM. Following automatic image segmentation, the concentration of striatal dopamine D2 receptors shows a significant left-sided asymmetry in male schizophrenics compared with male controls. The mean left-minus-right laterality index for controls is -1.52 (95% CI -3.72-0.66) and for patients 4.04 (95% CI 1.07-7.01). Analysis of variance shows a case-by-sex-by-side interaction, with F=10.01, P=0. 005. We can now demonstrate that the previously observed male sex-specific D2 receptor asymmetry in schizophrenia, which had failed to attain statistical significance, is valid. Cluster analysis of dynamic nuclear medicine studies provides a powerful tool for automatic segmentation and noise reduction of the images, removing much of the subjectivity inherent in region-of-interest analysis. The observed striatal D2 asymmetry could reflect long hypothesized disruptions in dopamine-rich cortico-striatal-limbic circuits in schizophrenic males.

Feature extraction techniques for exploratory visualization of vector-valued imagery

This paper addresses the exploratory visualization of multispectral image data. In such data, each component of the vector pixel corresponds to a different imaging modality or a different combination of imaging parameters, and may provide different levels of contrast sensitivity between different regions of the underlying image. We address the problem of presenting this multidimensional data to human observers by synthesizing a display matched to their visual capabilities. Specifically, we seek to determine a data-adaptive linear projection of the vector data to one dimension that produces a grayscale image providing maximum discrimination between the different regions of the underlying object. The approach is equivalent to the extraction of the best linear feature of the vector field. Several new feature-extraction criteria that take into account both the spatial and multivariate structures of the data are proposed and illustrated by simulations on test images.

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 (ROI's) for normal and abnormal tissues are defined. These ROI's 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.

Magnetic resonance image synthesis using a flexible model

Image synthesis methods are based on the hypothesis that a magnetic resonance (MR) image with optimized contrast can be reproduced by synthesis from three calculated basic images of T1, T2 and spin density. This method, however, is limited by noise due to uncertainties in the initial measurements. The principal component analysis (PCA) method is based on an information theory approach that decomposes MR images into a small set of characteristic feature images. PCA images, or eigenimages, show morphology by condensing the structural information from the source images. Eigenimages have also been shown to improve contrast-to-noise ratio (CNR) compared with source images. In this study we have developed a method of synthesizing MR images using a flexible model, comprising a set of eigenimages derived from PCA. A matching process has been carried out to find the best fit between the model and a synthetic image calculated from the Bloch equations. The method has been applied to MR images obtained from a group of patients with intracranial lesions. The images derived from the flexible model show increased lesion conspicuity, reduced artefact and comparable CNR to the directly acquired images while maintaining the MR characteristic information for diagnosis.

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.

Target-point combination of MR images

A method is described for combining multiple magnetic resonance images of the same anatomic slice to produce a single image which incorporates the favorable contrast features of each of the original images. The target-point method is a general method that includes linear combination as a subset and is designed to deal with the clinical need to maximize the contrast-to-noise ratio between several pairs of tissue simultaneously. Although it is intrinsically a nonlinear method, noise propagates approximately uniformly into the combined image. In examples of brain images the target-point method produces images with higher mutual contrast than the first principal component weighted sun image.

CNR enhancement in the presence of multiple interfering processes using linear filters

Given several images of the same slice, a linear filter can produce an image in which the contrast-to-noise ratio (CNR) between pathological and normal tissues is greater than in any of the initial images. To distinguish the pathology from more than one tissue, the filter should optimize the set of CNRs between the pathology and each of the interfering tissues. We define the optimal filter as the one which provides the largest value for the minimum CNR in the set and show how it is selected from a field of only four possibilities. The filter is demonstrated with both experimental phantom studies and clinical cases. Filter performance is compared with that of other techniques for distinguishing a desired feature from more than one interfering process.

Optimization of survey protocols for MRI

A method for evaluating the sensitivity of MRI protocols to changes in an arbitrary number of tissue parameters over a broad range of parameter values is presented. This analysis is useful for choosing an optimal basis set of images for either a "survey" protocol or tissue segmentation algorithms. A survey protocol is required when searching for a lesion of unknown type or location. Segmentation requires unique signal signatures for tissues that may cover a broad range of tissue parameter values. Data acquisition is modeled as a mapping of a domain of tissue parameter values into a signal manifold in a signal strength space defined by the MRI protocol. The efficacy of the protocol is evaluated by investigating the characteristics of the signal manifold. A figure of merit which maximizes the probability of discriminating each point in the domain of tissue parameters from all others is developed.

Computing material-selective projection images in MR

We detail a robust, general method for computing projection images of individual materials in a volume as linear combinations of MR projection images with different material-dependent weightings. Signal per unit volume for each material in each raw image is acquired directly for accurate cancellation of undesired, overlapping materials. The weighted sum of the input images is determined to maximize the signal-to-noise ratio (SNR) and minimize inhomogeneity effects in the material-selective images. We tested the implementation experimentally in both phantom and human studies, producing selective images with reasonable SNRs and material isolation. With further development of sequences to rapidly acquire input images having greater material differentiability, we envision the application of the selective projection imaging format to screening studies searching over large volumes for diseased tissues.

MR imaging of cerebral tumors: state of the art and work in progress

MR imaging of the brain has made tremendous progress during the last years. This technique is generally superior to computed tomography (CT) in brain tumors, due to its capability for direct imagining in various planes and its high tissue contrast. Moreover, the detectability and differentiation of extraaxial tumors, previously the domain of CT, has been improved with paramagnetic contrast agents (PCA). Although, the sensitivity of MRI for intracranial tumors is unchallenged, the specificity for such tumors is not remarcably greater than that of CT. Differentiation between high grade glioma, abscess and metastasis still requires biopsy for definitive diagnosis. Methods for improvement of specificity--tissue characterization--are currently being evaluated in a clinical setting. Further development in this field is necessary before such methods can be applied on a routine basis.

Optimization of MR protocols: a statistical decision analysis approach

A new method of optimizing MRI data acquisition protocols is presented. Tissues are modeled with probability density functions (PDFs) of tissue parameter values (such as T1, T2). The imaging data acquisition process is modeled as a mapping from a tissue parameter space to a signal strength space. Tissue parameter PDFs are mapped to signal strength PDFs for each tissue in a clinical problem. The efficacy of an MRI protocol is evaluated using the methods of statistical decision analysis applied to the signal strength PDFs, including the propagation of noise. This procedure evaluates the ability to discriminate different tissues based on the signal strengths produced with the protocol. The model can incorporate an arbitrary number of tissues, parameters, and pulse sequences in the protocol. The multivariate nature of MRI and the observed broad distribution of tissue parameter values makes this model more appropriate for optimizing data acquisition protocols than methods which maximize the signal-difference-to-noise ratio between discrete values of the tissue parameters. It is shown that these two methods may calculate different optimal protocols. The method can be used to optimize data acquisition for quantitative computer-based tissue classification, as well as imaging. Data acquisition and image processing philosophies are discussed in light of the method.