Linear Fusion of Reconstructed Magnitude and Phase part of MRI

Poorvi Hasmukhlal Patel


The main objective of my research work is to fuse the denoised medical images, as the two parameters which plays important role in MRI (magnetic resonance imaging), are feature extraction and object recognition, which would be difficult if the images are corrupted with noise. Denoising is a challenging problem in MRI (acquired as a complex signal) as the magnitude and phase error both are to be corrected. Now for detailed diagnosis the MR Images can be taken more than once say one taken six months ago and one taken currently, so it is impossible to have the same alignment of the body part same as the previous scan so it would be difficult to compare MRI with naked eyes. So after getting denoised MRI, I am fusing the MR Images. I have developed an algorithm for MRI denoising in which I am using non-linear adaptive gradient smoothing filter for magnitude reconstruction and non-linear anisotropic diffusion for phase reconstruction and affine transformation for fusing the MR images linearly.  I have got satisfactory results from my algorithm as applied on various live MR Images taken from the radiologist. Instead of using Gaussian filter or linear filter the non-linear gradient filter gives the better result as its smoothes the edges little hence preserving the fine details and borders. Anisotropic diffusion of the magnitude reconstructed MR Image gives sharp edges and better visualization than the simple FFT transform, hence phase part is reconstructed. Linear fusion using affine transform gives image with better features identified; hence better diagnosis and treatment can be done.


MRI denoising, magnitude and phase reconstruction, non-linear gradient filter, affine transform

Full Text: PDF


  • There are currently no refbacks.