Phase sensitive reconstruction of magnetic resonance images
LccRC 78.7 N83 M34 1991
LcshMagnetic resonance imaging
Image processing - Digital techniques
MetadataShow full item record
AbstractMagnetic resonance (MR) imaging is a recently developed technique that allows physicians to get high-quality images of living tissue without harmful radiation. MR images are complicated by phase errors, which cause a real image to become complex. In the standard reconstruction technique, the phase error is simply discarded. It is the purpose of this thesis to show how estimating and then compensating for phase errors can help improve the performance of the reconstruction process. Finding a method of estimating the phase error is the first problem to overcome. Existing algorithms are reviewed and a new algorithm based on polynomial modeling is introduced. The algorithms are test d by applying them to three clinical MR images. The desirable properties of a phase estimate are identified and the output of each algorithm is measured for these properties. The advantages of the new polynomial phase estimate are that it has a smooth shape that has good frequency domain properties and a high immunity to image noise. Its disadvantages are that it does not account for localized phase changes and it does not work on all types of images. Partial Fourier reconstruction is a technique that uses phase correction to reduce the raw data requirements of an image. Four existing algorithms are analyzed, and from this analysis, a new algorithm, based on a finite impulse response (FIR) phase correction, is introduced and then modified. The algorithms are applied to both artificial data with artificial phase estimates and clinical data with the phase estimates acquired from the algorithms above. The overall error introduced by the modified FIR algorithm is only slightly poorer than the best of the existing algorithms. Its advantages are that it executes faster and is better at confining errors to the image region where they occur. The magnitude operation that is normally used before displaying MR images produces undesirable side effects in low signal-to-noise ratio (SNR) images and inversion recovery images. In low SNR images, the magnitude operation reduces the contrast of the image. When the magnitude of an inversion recovery image is taken, the sign of the image is lost and the contrast is reduced. Both of these problems can be overcome by using a phase-corrected real image. When tested on clinical images, the images corrected with the polynomial phase estimate produce the best results in both cases. Modeling is a procedure that attempts to reduce rippling artifacts that appear in severely truncated data sets. By modeling phase-corrected data, the algorithm only requires half the computer resources that it needs for complex image modeling. It is shown that modeling and partial Fourier algorithms can be combined to reconstruct images from very small data sets. An image processing package, developed to implement the algorithms in this the is, is described. The package creates a high-speed, interactive image processing environment. An interface was developed to allow researchers to easily add their own procedures to the package.
Bibliography: p. 179-183.