Probabilistic Atlas-Based Image Reconstruction for Ultrafast MR Spectroscopic Imaging

Motivation: Ultrafast MRSI significantly accelerates MRSI scans but faces challenges in image reconstruction due to high dimensionality and low SNR. Goal(s): To improve image reconstruction for ultrafast MRSI. Approach: A new method is proposed to effectively incorporate spatiospectral priors, combining spectral basis functions for spectral constraints, a generative atlas for spatial priors, and a sparse component for subject-specific image features. Results: The proposed method has been applied to SPICE. Both simulation and in vivo experiments showed significantly improved accuracy and reduced uncertainty over existing subspace-based and edge-preserving regularization-based reconstruction methods. Impact: With the capability to reduce imaging dimensionality and improve SNR, the proposed method can further accelerate MRSI scans, making high-speed high-resolution MRSI possible for practical applications.