Abstract
PURPOSE: Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) is used in cancer imaging to probe tumor vascular properties. Compressed sensing (CS) theory makes it possible to recover MR images from randomly undersampled k-space data using nonlinear recovery schemes. The purpose of this paper is to quantitatively evaluate common temporal sparsity-promoting regularizers for CS DCE-MRI of the breast. METHODS: We considered five ubiquitous temporal regularizers on 4.5x retrospectively undersampled Cartesian in vivo breast DCE-MRI data: Fourier transform (FT), Haar wavelet transform (WT), total variation (TV), second-order total generalized variation (TGV (α)(2)), and nuclear norm (NN). We measured the signal-to-error ratio (SER) of the reconstructed images, the error in tumor mean, and concordance correlation coefficients (CCCs) of the derived pharmacokinetic parameters K(trans) (volume transfer constant) and v(e) (extravascular-extracellular volume fraction) across a population of random sampling schemes. RESULTS: NN produced the lowest image error (SER: 29.1), while TV/TGV (α)(2) produced the most accurate K(trans) (CCC: 0.974/0.974) and v(e) (CCC: 0.916/0.917). WT produced the highest image error (SER: 21.8), while FT produced the least accurate K(trans) (CCC: 0.842) and v(e) (CCC: 0.799). CONCLUSION: TV/TGV (α)(2) should be used as temporal constraints for CS DCE-MRI of the breast.