A Robust Majorization-Minimization Algorithm for Sequential Neuroimage Registration
Abstract. Robust registration of a sequence of brain images is critical for tasks such as three-dimensional (3D) reconstruction, time-series alignment, and common coordinate mapping. Here, we propose a robust registration algorithm based on the majorization-minimization (MM) optimization principle. This algorithm produces both 3D reconstructed images and transformation parameters for one diffeomorphism and one rigid transformation per two-dimensional (2D) sequential image. To achieve the strong robustness necessary for working with heterogeneous images in a high-throughput setting, we minimize a loss that is quadratic for small errors and almost constant for large errors, making it even more robust than [Formula: see text] and Huber losses. We motivate this choice and demonstrate that it can be effectively implemented in an MM algorithm. We then test our algorithm on simulated images with a known gold standard and on two different modalities of microscopy images.
Authors
- Kenneth L. Lange (ORCID: https://orcid.org/0000-0002-1313-5030)
- Daniel Jacob Tward (ORCID: https://orcid.org/0000-0002-4607-6807)
- Gaiting Zhou
Institutions
- University of California, Los Angeles (US)
Publication Details
- Journal
- SIAM Journal on Imaging Sciences
- Published
- 2026-11-16
- DOI
- https://doi.org/10.1137/25m1791755
- Primary Topic
- Medical Image Segmentation Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Karen Toffler Charitable Trust
- National Institutes of Health
- National Institute of Mental Health
- National Human Genome Research Institute
- National Institute of General Medical Sciences
- National Institute of Neurological Disorders and Stroke