Data driven MRI inhomogeneity correction

Abstract

A data driven method for correcting low The basic assumption for this work was that the correcfrequency inhomogeneity in magnetic resonance images tion function can be found from small variations within is presented. Small variations within a tissue type are these homogeneous areas by extensive averaging. The efmodeled and a correction function is generated. The spefect of image of the magnetic properties on the correction ciality of this method i s that it is based on image feafunction must be minimized. lures and does not need a phantom nor user interaction We decided to approximate the correction function by to generate the correction function. The image correction separating its spatial variables (3). This works well for simplifies digital image analysis and can enhance clinical head coils, but can fail for surface coils. However, we sugevaluation. A s a result, the correction technique reduces gest that this approximation is good enough when only ihe inhomogeneity and improves the contrast of magnetic one surface coil is used. resonance images. M(p3 = ~ O ~ Z ( P . ) ~ , b J p , ) ~ ( 3 ) The images are first averaged spatially by using 3 x 3 block neighborhood. This removes some of the noise and Although magnetic resonance imaging (MRT) can provide a more Of and is generated. When accurate infomation from the human the correction function is calculated it is applied to the ous diagnostic PUrPoses, the analysis of the quantitative Original image and thus no blurr or lack of detail is visible infomation of Mm is still in its early stage 111. Image in the corrected images. The filtered images IF are j u s t inhomogeneity is the greatest single bad characteristic of used to calculate the correction function. MR images to i n d i t d e Some easy methods of image The average absolute difference between 4-connected segmentation such as tresholding 121. pixels is calculated. A small “variation” is defined to a The messured image I can be thought of being corndifference smaller than the average. A homogeneous area posed from three elements: ideal magnetic image M, is defined to an area in which d l the pixels have small N and radio frequency (m) disto&ion F. Letting difference with their connected pixels in the area. Let ~7 5 be a location in the image space, 8 following estimative and ?be 4-connected pixels and can be calculated from: equation can be written: