Validation of vessel-based registration for correction of brain shift

Abstract

The displacement and deformation of brain tissue is a major source of error in image-guided neurosurgery systems. We have designed and implemented a method to detect and correct brain shift using pre-operative MR images and intraoperative Doppler ultrasound data and present its validation with both real and simulated data. The algorithm uses segmented vessels from both modalities, and estimates the deformation using a modified version of the iterative closest point (ICP) algorithm. We use the least trimmed squares (LTS) to reduce the number of outliers in the point matching procedure. These points are used to drive a thin-plate spline transform to achieve non-linear registration. Validation was completed in two parts. First, the technique was tested and validated using realistic simulations where the results were compared to the known deformation. The registration technique recovered 75% of the deformation in the region of interest accounting for deformations as large as 20 mm. Second, we performed a PVA-cryogel phantom study where both MR and ultrasound images of the phantom were obtained for three different deformations. The registration results based on MR data were used as a gold standard to evaluate the performance of the ultrasound based registration. On average, deformations of 7.5 mm magnitude were corrected to within 1.6 mm for the ultrasound based registration and 1.07 mm for the MR based registration.

Mutual‐information‐based image to patient re‐registration using intraoperative ultrasound in image‐guided neurosurgery

“…This is achieved by maximizing the mutual information ͑MI͒ between the iUS and pMR. Other feature-based methods of registering these two image modalities require segmentation of anatomical structures ͑e.g., surfaces 11 or vessels 12 ͒, which can be difficult to automate and usually requires manual ͑time consuming͒ intervention especially in the US images. Application of MI-based registration of US images is relatively unexplored, 13 and only a limited number of studies have been reported ͑e.g., between US and MR for brain, 14,15 liver, 16 and phantom images; 17 between US and MR angiography of carotid arteries; 18 between US and CT for kidney; 19 between cardiac US and SPECT; 20 and between US and US for abdominal and thoracic organs 21 and breast 22 ͒.…”

Registration of MR to Percutaneous Ultrasound of the Spine for Image-Guided Surgery

“…In order to reduce the influence of possible outliers on the registration result, the algorithm was modified by incorporating the Least Trimmed Squares (LTS) robust estimator as described by Reinertsen et al [10].…”

Papillary Muscle Segmentation from a Multi-atlas Database: A Feasibility Study

“…This is a big challenge as the 3D registration of TTE to MRI remains to be an open problem. To date, most existing approaches for echo to MRI (or CT) registration have been used in neurosurgical applications [2,3]. Cardiac echo-MR registration has been addressed to a lesser extent [4][5][6].…”