Abstract-We evaluate a combined discrete wavelet transform (DWT) and wavelet packet algorithm to improve the homogeneity of magnetic resonance imaging when a surface coil is used for reception. The proposed algorithm estimates the spatial sensitivity profile of the surface coil from the original anatomical image and uses this information to normalize the image intensity variations. Estimation of the coil sensitivity profile based on the wavelet transform of the original image data is found to provide a robust method for removing the slowly varying spatial sensitivity pattern.Keywords -MRI, wavelet packet, discrete wavelet transform, surface coil
I. INTRODUCTIONSurface coils offer the potential for a 2 to 5 fold increase in sensitivity compared to volume coils. Despite of this signal-to-noise advantage, surface coils are not used regularly for many applications because of their small region of coverage and their intrinsically inhomogeneous reception profile. The limited coverage of a single small surface coil can be extended by using a phased array of multiple coils [1].The phased array technique also improves the homogeneity of the images in the plane of the array, but the image intensity is still significantly brighter near the coils than deeper in the brain. Thus, the surface coil detector has an inherently inhomogeneous reception profile that leads to a variation in intensity across the image. This significantly degrades the utility of the images for evaluation of pathology in the cortex.To correct the inhomogeneity of the surface coil MRI due to the modulation of coil reception sensitivity profiles, several approaches have been studied to estimate the coil profile, which is used to correct the measured inhomogeneous MRI [1][2][3][4][5][6][7][8][9][10][11][12]. These methods use either a theoretically generated model [1,3,13] of the coil or the information in the image itself [2,4,[6][7][8] to generate the expected coil sensitivity map. In the first case, knowledge of the location and orientation of each surface coil is required in addition to a B 1 field map generated from the coil geometry. In the second case, the coil intensity profile can be approximated by a low-pass filtered version of the original image. The low pass filter based approximation of the surface coil profile requires a priori knowledge of the anatomy and coil fall-off in order to determine the appropriate cut-off spatial frequency which separates the low frequency variations due to coil fall off from the higher spatial frequency variations due to the anatomy.Here we propose a solution to correcting surface coil image intensity variations using post-hoc processing of the original surface coil image. The method identifies edges in the image and uses this information to improve the estimation of the coil sensitivity map. The slowly varying intensity changes comprising the estimated coil sensitivity map are determined from a filter bank implementation. This method allows the comparison of multiple levels of spatial filtering. The optimum level...