Motion-Corrected MR Image Reconstruction
(Funded by EPSRC grant EP/P001009/1)
Motion, such as that caused by breathing or the heart beating, can cause artefacts in reconstructed MR images. In this work, we developed techniques to detect these artefacts and even correct for them during the reconstruction process. This was made possible by exploiting deep learning methods. In our initial work (Oksuz et al, Proc MICCAI MLMIR) we trained a variant of the AUTOMAP reconstruction network to perform motion artefact correction. Subsequently, we developed separate artefact detection and correction networks in a cascaded architecture and traned them in an end-to-end fashion (Oksuz et al, Proc MICCAI).
- l. Oksuz, J. Clough, B. Ruijsink, E. Puyol-Antón, A. Bustin, G. Cruz, C. Prieto, D. Rueckert, A. P. King, J. A. Schnabel, "Detection and Correction of Cardiac MR Motion Artefacts during Reconstruction from K-space", Proceedings MICCAI, 2019. (Arxiv paper)
- I. Oksuz, J. Clough, A. Bustin, G. Cruz, C. Prieto, R. Botnar, D. Rueckert, J. A. Schnabel, A. P. King, "Cardiac MR Motion Artefact Correction from K-space Using Deep Learning-Based Reconstruction", Proceedings MICCAI MLMIR, pp21-29, 2018. (paper)
In later work, we extended the motion artefact detection and correction network to also perform image segmentation. This resulted in a deep learning based solution that could both reconstruct and segment cine MR data, whilst accounting for the possible presence of corruption in the data due to motion and mistriggering. The model had state-of-the-art performance in the presence of corrupted data as well as when presented with uncorrupted data.
- I. Oksuz, J. Clough, B. Ruijsink, E. Puyol-Antón, A. Bustin, G. Cruz, C. Prieto, A. P. King, J. A. Schnabel, "Deep Learning Based Detection and Correction of Cardiac MR Motion Artefacts During Reconstruction for High-Quality Segmentation", IEEE Transactions on Medical Imaging, 2020. (paper)