Direct estimation of metabolite maps from undersampled k-space data using linear tangent space alignment

Conventional MRSI methods perform MRSI image reconstruction and spectral quantification in two separate steps. This work presents a novel direct estimation method for MRSI that reconstructs high-resolution metabolite concentration maps from undersampled k-space data, leveraging a linear tangent space alignment (LTSA) model for spectral quantification and a low-rank (LR) model for denoising. Furthermore, the proposed framework allows estimating the temporal basis functions of the LR model from the undersampled, noise-corrupted k-space data, thus eliminating the need for experiment-dependent or pre-acquired spectral training data. The performance of the proposed method was validated using numerical simulation phantom and in vivo MRSI data.