Observations on maximum entropy processing of MR images

Beyond theg-factor limit in sensitivity encoding using joint histogram entropy

The maximum practical speed-up that can be achieved using parallel imaging methods is widely accepted to be limited by g-factor noise. An approximate expression for the g-factor noise as a function of the principal eigenvector of the inverse sensitivity matrix is derived. This formulation allows g-factor enhanced noise to be reduced by a constrained optimization procedure with joint image histogram entropy between a reference image and a SENSE image as an image quality metric. The reference image does not need to have identical resolution or contrast. The reference image may also be used for coil calibration. The limits of the method are explored using simulated and real array coil data with high g-factor using a variety of contrast and resolution combinations. The method preserves image structure, contrast, and lesions even when these were not observable in the reference data. In all cases g-factor was dramatically reduced.

A method of radio-frequency inhomogeneity correction for brain tissue segmentation in MRI

An automatic method of correcting radio-frequency (RF) inhomogeneity in magnetic resonance images is presented. The method considers that image intensity variation due to radio-frequency inhomogeneity contains not only low frequency components, but also high frequency components. The variation is regarded as a multiplication of low frequency (capacity variation of coil) and the frequency of object (true image). The efficiency of the proposed method is illustrated with the aid of both phantom and physical images. The impact of the inhomogeneity correction on brain tissue segmentation is studied in detail. The results show significant improvement of the tissue segmentation after inhomogeneity correction.

Image feature based automatic correction of low-frequency spatial intensity variations in MR images

An automatic method of compensating for low-frequency variations in magnetic resonance images is presented. Small variations within a tissue type are modelled and a correction function is generated. The method is based completely on image features and does not need a phantom or user interaction to generate the compensation function. This image correction simplifies digital image analysis and may enhance clinical evaluation. As a result, the correction technique reduces inhomogeneity and improves contrast. Our results show that the radiofrequency response variation of coils can be reduced. The segmentation process, even with a simple threshold method, produces more reliable results when corrected images are used. The presented method is most useful for images acquired in the sagital and coronal planes with circular local coils, or using surface coils, e.g., spine coils.

MRI segmentation: methods and applications

The current literature on MRI segmentation methods is reviewed. Particular emphasis is placed on the relative merits of single image versus multispectral segmentation, and supervised versus unsupervised segmentation methods. Image pre-processing and registration are discussed, as well as methods of validation. The application of MRI segmentation for tumor volume measurements during the course of therapy is presented here as an example, illustrating problems associated with inter- and intra-observer variations inherent to supervised methods.