Sub 2 mm resolution fMRI at 3T using randomly undersampled 3D-EPI with locally low-rank + temporally sparse reconstruction

Motivation: Many fMRI applications require sub 2 mm resolution, but most existing methods lose functional sensitivity at sub 2 mm resolution due to thermal noise, especially at 3T. Goal(s): To investigate and develop acquisition and reconstruction methods for high resolution BOLD fMRI at 3T. Approach: 3D-EPI with randomized sampling was used for acquisition. Images were reconstructed via locally low-rank plus sparse decomposition. A finger tapping task is used for testing. Results: Motor activation was detected at 1.8 mm resolution, with our protocol producing activation maps with higher specificity compared to standard reconstruction methods. Impact: Randomized 3D-EPI and locally low-rank plus temporally sparse decomposition are novel approaches for high resolution fMRI. Other fMRI scientists can use our vendor-agnostic, open-source implementation as a template, adapting it to suit their specific high resolution fMRI needs.