We propose a novel method that integrates generative image priors with subspace constrained MRSI reconstruction. A sufficient and flexible image representation was first generated by adapting a pretrained StyleGAN to subject-specific anatomical images. We validate that StyleGAN can be flexibly adapted to accurately represent different contrast of the same subject. The adapted GAN prior is then used to model the spatial coefficients in the subspace-based reconstruction. Improved performance over the original subspace reconstruction in the SPICE framework is demonstrated using simulation and in vivo data.
This abstract and the presentation materials are available to members only; a login is required.