Deformable MRI-Ultrasound registration using correlation-based attribute matching for brain shift correction:...

Area of Science:

• Medical image analysis
• Neurosurgical navigation
• Image registration

Background:

• Brain shift, or intraoperative tissue deformation, reduces the effectiveness of preoperative images in guiding neurosurgery.
• Non-rigid registration between preoperative magnetic resonance (MR) and intraoperative ultrasound (iUS) is a proposed solution to mitigate brain shift.
• Existing MR-iUS registration methods often lack accuracy and generality across diverse clinical datasets.

Purpose of the Study:

• To develop and validate an accurate and generalizable MR-iUS registration algorithm for neurosurgical guidance.
• To address the challenges of accuracy and generality in multi-site clinical data for MR-iUS registration.
• To improve the compensation for brain shift during neurosurgery.

Main Methods:

• Employed high-dimensional texture attributes instead of image intensities for MR-iUS registration.
• Utilized correlation-based attribute matching, replacing standard difference-based methods.
• Developed a strategy to manage significant field-of-view mismatches between MR and iUS images.
• Optimized key parameters across multi-institutional brain tumor datasets (43 patients, 758 landmarks).

Main Results:

• The algorithm consistently reduced landmark errors across three independent datasets (from ~5-6 mm to ~2 mm).
• Achieved competitive accuracy against 15 other algorithms, demonstrating state-of-the-art performance.
• Exhibited high accuracy and generality, maintaining performance with fixed parameters across multi-site data, unlike other methods requiring dataset-specific tuning.

Conclusions:

• The proposed MR-iUS registration algorithm offers a robust solution for accurate and generalizable neurosurgical navigation.
• It effectively compensates for brain shift, enhancing the utility of preoperative imaging during surgery.
• The study provides a detailed characterization of landmark errors by brain region and tumor type, filling a gap in the literature.