Answers are grounded in this paper and cite their sources.
Ultrasound is a fast, widely available, and cost-effective imaging modality. These properties make it well-suited for intra-operative use, where it can provide surgeons with near real-time updated visualization of anatomical structures and tissue deformations. In brain tumor resection, this capability is especially valuable due to the delicate nature of the operation and presence of brain shift. A common strategy to exploit intra-operative ultrasound is to register pre-operative images, typically high-resolution MRI, to intra-operative ultrasound, thereby combining detailed anatomical information with up-to-date deformation estimates. In this work, we present three registration methods based on distinct models of the relationship between ultrasound intensity and the intensities of other imaging modalities. At the core of each method is a similarity metric, derived within a unified analytical framework. We evaluate these methods using rigid and nonlinear deformation models across six neurosurgical registration tasks: three involving pre-resection ultrasound volumes, which are well studied in the literature, and three involving post-resection ultrasound volumes, a setting that has received limited attention in prior work. Averaged over all datasets, our proposed methods reduce the target registration error from 5.49 mm to 2.96 mm, 2.85 mm, and 3.07 mm, respectively. Using a single, fixed set of parameters, our approaches consistently improve accuracy across all tasks, with total runtimes below 40 s and performances comparable to state-of-the-art methods. Finally, we demonstrate how different assumptions about ultrasound intensity formation influence registration performance and identify the models that lead to more accurate and robust alignment.
No takes yet. Share an insight, caveat, or question.
Mohammadreza et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: