Weighted Sum of Squares Reconstruction Based on Regional Mutual Information in parallel MRI

The quality of image reconstruction is very important in parallel magnetic resonance imaging (MRI). Sum of Squares (SOS) reconstruction is considered as a popular reconstruction method in parallel MRI, however, it does not have a good suppression on motion artifacts caused by the autonomous and involuntary movement of patient. This paper presents an improved weighted sum of squares (WSOS) reconstruction algorithm based on regional mutual information (RMI). Every phased array coil image is divided into several small regions. The mutual information of regions is selected as the similarity metric between images to calculate the damage degree of coil data. Different coil image data are given different weights, and then the suitable optimal weighted coefficients will be redistributed to the WSOS algorithm formula to reconstruct the images. The purpose is to minimize the impact of coil damage data on the final reconstructed image, eliminate artifacts effectively, and strive to overcome the shortcomings of the SOS algorithm. To demonstrate the validity of the proposed improvement algorithm to eliminate artifact, two groups of 8-channel brain and cardiac receiver coils k-space under-sampling data have been reconstructed by this proposed method. Experimental results show that the reconstructed images based on the regional mutual information (RMI) have a better uniformity over the field of view than the conventional WSOS reconstruction based on mutual information (MI). In addition, the performance of the two reconstruction methods is also quantitatively evaluated by the normalized mean squared error (MSE), and the MSE value further confirm the feasibility and effectiveness of the improved algorithm.