Study On Intensity Inhomogeneity Correction And Coil Sensitivity Estimation In Magnetic Resonance Imaging

Nowadays magnetic resonance imaging is widely implemented in clinical and computer aided diagnosis. MR images are, however, often degraded by intensity inhomogeneity, also known as bias field, which is mainly caused by inhomogeneous magnetic field, imperfections of the radio frequency coils or inhomogeneous coil sensitivities of the receiving coils. The intensity inhomogeneity can cause unwanted intensity variations throughout the tissues of the same type. This intensity inhomogeneity may have an impact on clinical diagnose and computer aided analysis. Although it is not a serious problem for qualitative diagnosis, such artificial intensity variation may have serious impact on automatic quantitative image analysis techniques, such as segmentation and registration, which are very sensitive to intensity variation. Moreover, intensity inhomogeneity is particularly severe in MRI at ultra-high fields and sometimes even disables visual diagnosis. Therefore correction of intensity inhomogeneity is usually required for MR images.The correction of intensity inhomogeneity from an MR image is generally a blind separation problem since the true image and the bias field are both unknown. Fortunately the bias field can often be modeled as a spatially slowly-varying field and is a multiplicative component of the measured image. A common assumption for intensity inhomogeneity correction is that the intensities of tissues of the same type are close. Based on this assumption, we can know that the intensity variation inside a region, which is composed of the same type of tissues, mainly results from the intensity inhomogeneity. To obtain an accurate estimation of the intensity inhomogeneity, we proposed two polynomial fitting based correction methods. The methods based on image intensities and image gradients.We first proposed a polynomial fitting based correction method which extract the intensity inhomogeneity features from multiple tissue regions. The traditional intensity based methods estimate the bias field by determining a tissue region by determining thresholds in the histogram, and then fitting a low-order polynomial to the intensities in this region. The estimation accuracy of this method, however, relies on the accuracy of the segmented tissue region, if the determined thresholds are too large, different tissues may included into the same region, which leads to wrong intensity inhomogeneity feature extraction, while small thresholds may not extract enough intensity inhomogeneity information of the image. Besides, if the tissue is not evenly distributed across the whole image, the bias field may overly rely on the extrapolation of the polynomial, which may cause overfitting. To solve this problem, we proposed an improved method, estimating the intensity inhomogeneity surface by determining a set of tissue regions using region growingiteratively, in which the intensity differences represent the local intensity inhomogeneity. The full-region intensity inhomogeneity is then obtained by low-order polynomial fitting.We then proposed another image gradient based correction method. The intensity based correction method has to first determine the tissue regions in the MR image. Since the intensity inhomogeneity function is coupled with the real image intensities, if the tissue regions are not segmented correctly, the intensity differences between different tissue types may have an impact on the bias field estimation. To eliminate this effect, we estimate the intensity inhomogeneity by processing on the gradient maps of the inhomogeneous image, the intensity inhomogeneity surfaces are then obtained by minimizing a specific cost functions of the image’s one-order gradients in both x-direction and y-direction.We also improved the above mentioned methods so they can be applied for coil sensitivity estimation in multi-coil MR systems. The coil sensitivity is an important parameter for both phased array imaging and parallel imaging, so it must be estimated accurately. Same as the intensity inhomogeneity, the coil sensitivity is also a slow-varying surface, so we can apply the intensity inhomogeneity correction algorithms for coil sensitivity estmation. However, the intensity inhomogeneity caused by coil sensitivity is much heavier that some areas in a coil image have very low SNR. It may amplify the noise if an iterative intensity inhomogeneity correction algorithm is used for coil sensitivity.We proposed a novel iterative strategy which can estimate the coil sensitivity iteratively which avoid amplifying noise. This method improves the accuracy of coil sensitivity estimation, and can restrain the intensity inhomogeneity of the fused image in phased array imaging or eliminate the ghost artifact in the reconstructed image in parallel imaging.Our algorithms have been evaluated with both simulated and real images and the results demonstrate good intensity inhomogeneity correction and coil sensitivity estimation accuracy. Also, the methods have been applied on real MRI systems and achieved satisfying performance.