Attention-Driven Multi-channel Deformable Registration of Structural and Microstructural Neonatal Data
Grigorescu I, Uus A, Christiaens D, Cordero-Grande L, Hutter J, Batalle D, Edwards AD, Hajnal JV, Modat M, Deprez M (2022)
Publication Type: Conference contribution
Publication year: 2022
Journal
Publisher: Springer Science and Business Media Deutschland GmbH
Book Volume: 13575 LNCS
Pages Range: 71-81
Conference Proceedings Title: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Event location: Singapore, SGP
ISBN: 9783031171161
DOI: 10.1007/978-3-031-17117-8_7
Abstract
Image registration of structural and microstructural data allows accurate alignment of anatomical and diffusion channels. However, existing techniques employ simple fusion-based approaches, which use a global weight for each modality, or empirically-driven approaches, which rely on pre-calculated local certainty maps. Here, we present a novel attention-based deep learning deformable image registration solution for aligning multi-channel neonatal MRI data. We learn optimal attention maps to weigh each modality-specific velocity field in a spatially varying fashion, thus allowing for local fusion of structural and microstructural images. We evaluate our proposed method on registrations of 30 multi-channel neonatal MRI to a standard structural and microstructural atlas, and compare it against models trained without the use of attention maps on either single or both modalities. We show that by combining the two channels through attention-driven image registration, we take full advantage of the two complementary modalities, and achieve the best overall alignment of both structural and microstructural data.
Authors with CRIS profile
Involved external institutions
United Kingdom (GB)How to cite
APA:
Grigorescu, I., Uus, A., Christiaens, D., Cordero-Grande, L., Hutter, J., Batalle, D.,... Deprez, M. (2022). Attention-Driven Multi-channel Deformable Registration of Structural and Microstructural Neonatal Data. In Roxane Licandro, Roxane Licandro, Andrew Melbourne, Jana Hutter, Esra Abaci Turk, Christopher Macgowan (Eds.), Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (pp. 71-81). Singapore, SGP: Springer Science and Business Media Deutschland GmbH.
MLA:
Grigorescu, Irina, et al. "Attention-Driven Multi-channel Deformable Registration of Structural and Microstructural Neonatal Data." Proceedings of the 7th International Workshop on Perinatal, Preterm and Paediatric Image Analysis, PIPPI 2022, held in conjunction with the 25th International Conference on Medical Imaging and Computer-Assisted Intervention, MICCAI 2022, Singapore, SGP Ed. Roxane Licandro, Roxane Licandro, Andrew Melbourne, Jana Hutter, Esra Abaci Turk, Christopher Macgowan, Springer Science and Business Media Deutschland GmbH, 2022. 71-81.
BibTeX: Download