Abstract
Motion artefacts created by patient motion during an MRI scan occurfrequently in practice, often rendering the scans clinically unusable andrequiring a re-scan. While many methods have been employed to ameliorate theeffects of patient motion, these often fall short in practice. In this paper wepropose a novel method for removing motion artefacts using a deep neuralnetwork with two input branches that discriminates between patient poses usingthe motion's timing. The first branch receives a subset of the $k$-space datacollected during a single patient pose, and the second branch receives theremaining part of the collected $k$-space data. The proposed method can beapplied to artefacts generated by multiple movements of the patient.Furthermore, it can be used to correct motion for the case where $k$-space hasbeen under-sampled, to shorten the scan time, as is common when using methodssuch as parallel imaging or compressed sensing. Experimental results on bothsimulated and real MRI data show the efficacy of our approach.