High-Resolution Optogenetic Functional Magnetic Resonance Imaging Powered by Compressed Sensing and Parallel Processing

Optogenetic functional magnetic resonance imaging (ofMRI) is a powerful new technology that enables precise control of brain circuit elements while monitoring their causal outputs. To bring ofMRI to its full potential, it is essential to achieve high-spatial resolution with minimal distortions. With our proposed compressed sensing (CS) enabled method, high-spatial resolution ofMRI images can be obtained with a large field of view (FOV) without increasing spatial distortions and the amount of acquired data. The ofMRI data were sampled with passband balanced steady-state free precession (b-SSFP) fast stack-of-spiral sequence in order to achieve ultra-high-spatial resolution images in a short amount of time. Interleaves of data were randomly collected. The images were recovered from the undersampled k-space data by solving an unconstrained convex optimization problem, which balances the trade-off between data consistency and sparsity. The optimization problem can be solved by gradient descent combined with backtracking line search algorithms. Discrete cosine transform (DCT) were chosen as a sparsifying transform. The ofMRI image reconstruction was processed in parallel on a graphics processing unit (GPU) using C/C++ language supported by NVIDIA CUDA engine in order to achieve short reconstruction time. An existing nonequispaced fast Fourier transform (NFFT) algorithm was modified for our GPU parallel processing purpose. The results demonstrate that the compressed sensing reconstructed image has higher resolution while maintaining a precise activation map, compared to a fully sampled low-resolution image with the same amount of data and scan time. A 4-D image can be reconstructed in less than fifteen minutes, which allows compressed sensing ofMRI to become a practical application.