A Framework for SPICE MRSI Reconstruction Posterior Uncertainty Quantification

A Laplace-approximation framework for SPICE MRSI reconstruction at 7 T estimates posterior uncertainty in spatial coefficients and metabolite concentrations, matching repeated-scan patterns.

• Why it matters: SPICE improves MRSI reconstruction quality but lacks uncertainty estimates, limiting confidence in metabolite concentration maps and how acquisition choices affect reliability.
• What they did: The method uses Laplace approximation to estimate covariance, propagates it through the spectral basis and spectral fitting, and computes Hessian–vector products with voxel-wise conjugate-gradient without forming the full Hessian.
• The result: In repeated in vivo 7 T MRSI acquisitions, estimated posterior uncertainty shows spatial and spectral patterns similar to empirical repeated-scan uncertainty, and varies with spatial regularization and k-space undersampling.

AI summary of the abstract · may contain errors · not medical advice · read the paper.