Wavelet Tree Support Detection for Compressed Sensing MRI Reconstruction

A priori knowledge of the signal/image support based on its statistical and structural information in the transformed domain improves the quality of compressed sensing (CS) reconstruction. Hidden Markov tree models the wavelet domain support of magnetic resonance images obtained from undersampled k-space data very well. With the support information, parent-child pairs in the wavelet tree are detected accurately, and hence, iterative regularization problems for CS-based magnetic resonance imaging reconstruction are solved with high throughputs. Simulation results with artificial and real MRI datasets show significant improvements over existing methods in terms of just noticeable difference scanning besides other objective and subjective metrics.