Calibrationless joint compressed sensing reconstruction for rapid parallel MRI

Abstract 3D magnetic resonance imaging (3D MRI) or multi-slice MRI involves significant data acquisition time. Traditionally, scanning rate of conventional MRI is restricted due to inherent physiological and instrumental limitations. In multi-slice parallel MRI (multi-slice pMRI) adjacent slices are highly correlated, so one can interpolate missing k-space data of any slice from its adjacent slices. Moreover, images corresponding to different coils are also highly correlated as they represent the same field of view (FoV) with different spatial sensitivity functions. Exploiting above redundancies in multi-slice pMRI, we propose a joint compressed sensing (CS) reconstruction model, which contains two joint sparsity promoting regularization terms based on wavelet forest sparsity and joint total variations (JTV). Extensive experiments are carried out for performance evaluations and compared with existing methods using real multi-slice pMRI datasets. Results in terms of CS reconstruction quality show significant improvements compared to the state-of-the-art. Additionally, implementation of the proposed method is done using multi-core CPU and GP-GPU in a hybrid computing environment to study its clinical feasibility in terms of computational time.