Fast Voxelwise SNR Estimation for Iterative MRI Reconstructions.
Dalmaz Onat O, Abraham Daniel D, Toews Alexander R AR, Chaudhari Akshay S AS et al.
To develop a fast, general-purpose framework for voxelwise noise characterization in iterative MRI reconstructions, recovering the image-domain noise variance, the primary quantity from which SNR, g -factor, and related image-quality metrics are derived. The framework addresses both the intractability of closed-form formulas beyond Cartesian sampling and the long runtime of pseudo multiple replica (PMR). A stochastic noise estimator is proposed that operates in the image domain by Probing the Image-space noise COvariance matrix (PICO) with random probe images. Complex random-phase probes attain minimum kurtosis and reduce estimator variance in the tested reconstruction. PICO was validated against analytical benchmarks and PMR references using Cartesian knee data ( R = 2 ), physical brain phantom data ( R = 2 , 3 , 4 ), in vivo non-Cartesian brain data ( R = 2 ), and compressed-sensing knee reconstructions ( R = 2 ). In Cartesian reconstructions, PICO accurately reproduced analytical SENSE g -factor maps. In non-Cartesian reconstructions, it achieved 1 % g -factor NRMSE ≈ 11 . 9 × faster than PMR in a phantom and ≈ 2 . 81 × faster in vivo, with the efficiency advantage persisting at higher acceleration. For compressed sensing, the Jacobian-based estimator produced noise maps consistent with PMR and computed them 1 . 84 × faster at equal sample counts. Agreement worsened at large added-noise amplitudes and moderate TV strengths. PICO targets the exact covariance of fixed linear reconstructions and a first-order covariance for nonlinear reconstructions, providing a computationally efficient alternative to PMR. By reusing existing reconstruction primitives, it could help incorporate voxelwise noise maps into reconstruction workflows.