BOLD Acquisitions and GAN Synthetized VASO Contrasts for Rapid Layer-dependent fMRI

Motivation: Though useful, implementing VASO for layer activity analysis requires specialized pulse sequences, which can be more complex than standard BOLD fMRI sequences. Goal(s): To develop a GAN model that synthetically generate VASO contrast images from acquired BOLD images. Approach: We trained a GAN model from paired BOLD and VASO image dataset. Model was evaluated using metrics like PSNR, SSIM, and MAE and layer profiling. Results: Our GAN model translates the acquired BOLD contrast images into VASO contrast images with an average SSIM of 0.85 ± 0.02. Further, brain layer profiling shows agreement between acquired and GAN-assisted VASO images. Impact: We present a method to eliminate the need for implementing VASO pulse sequence by synthetically generating VASO images from acquired BOLD images.