Data-Driven Regularization Parameter Selection in Dynamic MRI

In dynamic MRI, sufficient temporal resolution can often only be obtained using imaging protocols which produce undersampled data for each image in the time series. This has led to popularity of compressed sensing (CS) based reconstructions. One problem CS approaches is determining regularization parameters, control balance between fidelity and regularization. We propose a data-driven approach total variation parameter selection, where reconstructions yield expected sparsity levels domains. The are from measurement reference spatial Two formulations proposed. Simultaneous search pair yielding both domains (S-surface), sequential selection S-curve method (Sequential S-curve). evaluated simulated experimental DCE-MRI. test case, methods reconstruction that close root mean square error (RMSE) optimal reconstruction. almost equal high perceived quality. Both lead highly feasible parameters cases while computationally more efficient.