Development of real time co-registration and visualization of multimodal neuroimages and intra-operative MRI

Abstract

[한글]

[영문]Recently, the registration of high resolution (HR) pre-operative (OP) MRI and intra-OP MRI (iMRI) is one of the hot issues. iMRI has been beneficial for reliable surgical operation. Since it has a low image quality, co-registration and visualization of pre-op HR image data and iMRI is clinically important. However, obscure contours, intensity inhomogeneity as well as small field of view (FOV) is challenges to overcome. The aim of this study was to investigate how multimodal images could be better registered into iMR image for image-guided neurosurgical operation. Nine patients (4 female, 5 male, age range: 32-73 years; mean age: 53 years) with glioma were scanned using pre-operative 3T MRI. During surgical operation, patients were scanned using a vertically opened Polestar 0.15T iMRI. We utilized automatic segmentation method using level set algorithm for extracting a region of tumor from whole brain image. Before registration, brain extraction using BET (brain extraction tool) and intensity heterogeneity correction procedures were done first to increase registration performance. To co-register pre-iMR image at the beginning and iMR images during operation, we used both linear and non-linear registration. At first, pre-iMR image is registered to iMR images by rigid registration, followed by non-linear registration which does not include affine transformation. In this study, we used two kinds of non-linear registration methods to select suitable method for our data. One is non-parametric non-rigid registration algorithm (Diffeomorphic demons) using diffusing models to perform image to image matching. The other is a GPU-based non-linear registration algorithm (AIRWC) which maximizes normalized mutual information using B-splines and consumes brief registration time. And we used landmark based registration for avoiding the anatomical information of whole brain. As a result, we found that the utilization of the BET was effective in the registration between iMR images with small FOV. In the non-linear registration process from pre-iMR image to iMR image during operation, the registered pre-iMR image showed acceptable registration. When the region of CSF (cerebrospinal fluid) was only used as the landmark of registration, the registration performance of registration was increased compared to registration with full image and anatomical information was preserved outstandingly. Also, the result image of diffeomorphic demons registration algorithm was more geometrically similar to the brain structure of iMR image during operation than that of AIRWC registration algorithm. In evaluation section, diffeomorphic demons method showed prominent registration performance without change. The transformation derived from the non-linear registration was applied to Pre-OP HR T1 image and diffusion tensor data were transformed to iMRI space and was visualized. According to the preliminary results, registration method using multimodal MR images can be used to reliably detect tumor boundary and facilitate tumor resection. We believe that our method will provide a clinical supplement to neurosurgical operation. The contributions of this experiment are an optimizing algorithm for intra-OP registration and visualizing deformation of brain tissue and DTI. Our method can be used for surgical decision making to minimize post-OP neurological deficits.