Answers are grounded in this paper and cite their sources.
Registration of longitudinal brain MRI scans containing pathologies remains a challenging problem, due to substantial changes in tissue appearance. While medical image registration methods have advanced considerably, they often lack the required precision and robustness. To address this gap, we introduced the brain tumor sequence registration (BraTS-Reg) challenge, the first public benchmark specifically designed for deformable registration between pre-operative and follow-up scans of patients with adult-type diffuse gliomas. Conducted in conjunction with ISBI-2022 and MICCAI-2022, BraTS-Reg provides de-identified, multi-institutional, multi-parametric MRI data accompanied by expert-annotated landmark points across time. Training data with ground truth annotations were publicly released for algorithm development, while validation data without annotations enabled pre-testing. Submitted containerized algorithms were evaluated on hidden testing cases. Quantitative evaluation was based on median Euclidean error (MEE), Robustness, and the determinant of Jacobian of the displacement field. Top-ranked methods demonstrated comparable performance and shared key features, including pre-alignment of scans, deep neural networks, inverse consistency analysis, and per-case test-time optimization. The best method achieved MEE comparable to inter-rater variability for 63% of landmarks, underscoring the need for further improvements in accuracy and robustness relative to human experts. BraTS-Reg serves as a three-fold active resource for research on image registration, providing a publicly available dataset, a framework for a community benchmark, and online evaluation tools, accessible at https://bratsreg.github.io/ .
No takes yet. Share an insight, caveat, or question.
Baheti et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: