Using deep learning for undersampled reconstruction has shown advantages for structural MRI and dynamic MRI, but is not commonly used for fMRI. In this work, we propose a neural network based reconstruction with temporal regularization to exploit the temporal redundancy of oscillating steady-state fMRI images. With a factor of 6 undersampling, the proposed method outperforms other approaches such as cascade of convolutional neural networks with high-resolution and high-quality fMRI results.
This abstract and the presentation materials are available to members only; a login is required.