Multimodal image registration for preoperative planning and image-guided neurosurgical procedures

Abstract

Image registration is the process of transforming images acquired at different time points, or with different imaging modalities, into the same coordinate system. It is an essential part of any neurosurgical planning and navigation system because it facilitates combining images with important complementary, structural, and functional information to improve the information based on which a surgeon makes critical decisions. Brigham and Women's Hospital (BWH) has been one of the pioneers in developing intraoperative registration methods for aligning preoperative and intraoperative images of the brain. This article presents an overview of intraoperative registration and highlights some recent developments at BWH.

Copyright © 2011 Elsevier Inc. All rights reserved.

Figures

Figure 1

Anatomical and functional MR information acquired pre-operatively. (A) An anatomical T1 scan of a patient with a lesion in the left frontal region. (B) Extracted white matter tracts using tractography from DTI. (C) Speech areas found with fMRI overlaid on an axial anatomical slice. Notice that one of the speech centers is adjacent to the tumor and should be conserved during surgery. (D) Composite view of the anatomical MR image and functional DTI and fMRI.

Figure 2

Pre- and intra-operative MR images that shows brain-shift and resection. (A) Pre- operative image. (B) Intra-operative image. Notice the cavity which is a result of the tumor resection. (C) Absolute difference image of the images in (A) and (B). Notice the large intensity differences around the tumor site. Large image gradients and image intensity differences are often the driving force in image registration methods. However, in the case of resections, these image forces should be contained to prevent them from driving the registration into false minima.

Figure 3

Transformation cascade. Three steps are needed to map the functional images, fMRI and DTI, into the intra-operative physical space. First, rigid, or affine, transforms, RfMRI and RDTI are estimated to align the functional images with the pre-operative anatomical image. A rigid transform is generally adequate because the brain is still enclosed in the skull and assumed to move rigidly up until the start of the surgery. Secondly, a non-rigid transform U is estimated to align the anatomical pre-operative image with the anatomical intra-operative image. A non-rigid transform is required to capture the non-linear movements due to for instance brain-shift. A rigid transform Rpoint is estimated from homologous points extracted from the patient space and the intra-operative image. In some intra-operative MR scanners, for instance the GE Signa SP 0.5T which was in use at BWH until 2007, the physical tracking space and the image space were aligned and the estimation of Rpoint was unnecessary. Consequently, points X in the pre-operative functional space can be mapped to the patient (physical) space with the following transformation cascade: Rpoint (U(RfMRI(X))). This enables navigation of the functional images in a neuro-surgical navigation system.

Figure 4

This diagram shows the general flow of a registration algorithm. It is generally an iterative process where the transformation parameters are estimated to minimize the “distance” between two images. The distance measure can measure anything from image intensities to the distance between homologous points in the two images.

Figure 5

Illustration of the effect of registration uncertainty on mapping of functional areas from the pre-operative to the intra-operative space. The anatomical image is a post-operative image, acting as a proxy for an intra-operative image, acquired from the same patient we show pre-operative images of in Figure 1. (A) A coronal MRI slice with the estimated boundary of a speech activated functional area after registration. (B) Takes into account the registration uncertainty, and rather than visualizing the point-estimate as in (A), it visualizes the marginal probability that a certain voxel is inside the functional area (purple denotes a probability of one which falls off normal to the iso-contours down to red which is close to a zero probability). If we assume that the hyper intense area is the site of the resection, then we can conclude with a small probability that the functional area was touched during surgery because the outer rim of the color map touches on the “resected” area. Image (C) shows an uncertainty map. The uncertainty is measured as the dispersion of the marginal posterior distribution over deformation parameters. We have colored high uncertainty with purple and low uncertainty with red. It can be seen that the registration uncertainty is higher close to the resection than away from the resection. It is important to keep in mind that a high uncertainty in the registration does not necessarily mean that the registration is inaccurate, but the chance that it is less accurate is higher.