Study on Diffusion MR Referential Image of Human Brain

摘要

Fig.5 is the diffusion MR referential image obtained from diffusivity images of 7 volunteers. Obviously, in this image, the contour line of brain is blurry and the contrast of image is low. There are two reasons on this problem. On one hand, the precision of normalization is not high enough to match the structure of different volunteers with that of the template image. To solve this problem, more normalizing parameters and high-precision image resample method are required. On the other hand, in this experiment, we only acquired two sets of image data of every volunteer, one of which is T_2-weighted image (no diffusion-sensitive gradient), the other is diffusion-weighted image (diffusion sensitive gradient incorporated). So we do not have enough data to use regression method to obtain the corresponding diffusivity image. Therefore, the diffusivity image shown above is the processing result in use of equation (4) but without regression analysis. For this reason, the diffusivity image has much background noise outside the brain that makes the brain image blurry. D is often very small, in other word. (ln(S_0 (x, y, z) / S_1 (x, y, z))/(b_1 - b_0) is very small. Hence even small background noise will badly affect the processing result of image. To solve this problem, one method is acquire many sets of image data, then obtain the diffusivity image using regression analysis, the other is extracting contour line of brain, then only process the image data inside the contour line of brain. Therefore, the background noise out of brain will not have any effect on the brain image. We can find that the contrast of normalized image is decreased from above figures. The reason is that in the process of normalization, the algorithm we used should do image resampling to the original image. Image resample is actually a process of interpolation to the original data. Hence the contrast of original image will be inevitably decreased.