Joint Neural Network for Fast Retrospective Rigid Motion Correction of Accelerated Segmented Multislice MRI

We demonstrate a deep learning approach for fast retrospective intraslice rigid motion correction in segmented multislice MRI. The proposed neural network architecture combines convolutions on frequency and image space representations of the acquired data to produce high quality reconstructions. Unlike many prior techniques, our method does not require auxiliary information on the subject head motion during the scan. The resulting reconstruction procedure is more accurate and is an order of magnitude faster than GRAPPA. Our work offers the first step toward fast motion correction in any setting with substantial, unpredictable, challenging to track motion.