Cortical Peeling: CSF/Grey/White Matter Boundaries Visualized by Nesting Isosurfaces

Source: Lecture Notes in Computer Science 1996;1131:99-104.
Author: Holmes CJ, MacDonald D, Sled JG, Toga AW, Evans AC.

Abstract:
ABSTRACT: One computational tool of particular interest in the analysis of MR images of the cerebral cortex has been the automatic extraction of surfaces. Recently, enhanced MRI images have been produced whose quality presented the opportunity to extract several of these surfaces from the same volume. We present preliminary data using this approach to visualize the boundaries of the cortex from the dura to the white matter. A series of isosurfaces were extracted from a signal-enhanced T1 MRI. The lower MR-intensity isosurfaces featured cortical vessels and dura, in contrast to higher intensity isosurfaces, in which the deeper structure of the superficial gyri became apparent. Two surfaces were chosen, one representative of the outer cortical surface and the other of the inner grey/white boundary, and their separation was computed to construct a cortical thickness map. The use of nested isosurfaces with high quality MR images can differentially segment surface features and reveal the underlying structure of gyri and sulci, aiding anatomical delineation, and may ultimately result in the automatic extraction of cortical thickness maps. METHODS: To produce an MRI volume with extremely high signal to noise ratio, 27 individual high contrast T1 volumes were acquired from a single subject. These volumes were automatically registered to a standard, Talairach-based space where they were intensity-averaged. Although a number of post-processing methods for the compensation of field non-uniformity exist, the method we used iterates between segmentation and estimation of the corrupting filed, with intensity non-uniformity modelled as a smoothly varying multiplicative field. Initially the skull is stripped and the remaining voxels are segmented into CSF, grey matter, and white matter using their intensity and the TPM for each class. The corrected intensity of each voxel is estimated from a weighted average of the mean intensities for the respective classes. The ratio of this estimate to the actual intensity is used to estimate the non-uniformity field (NUF) at that location. The NUF is then smoothed by a filter whose region of support is restricted to the region of the brain mask. Iteration proceeds by correcting the original volume using the NUF estimate and repeating the process of classifying and filtering. After convergence, thin plate splines are fit to the NUF to extrapolate it to regions outside the brain mask. RESULTS: Averaging 27 scans produced a single MRI volume whose enhanced signal made it possible to observe fine structure in the thalamus and brainstem. The homogeneity of the cortical grey matter was greatly enhanced, as was that of the white matter. We produced a series of surfaces that, when viewed in sequence, give the impression that the layers of the meninges, CSF and cortex are peeled away from the underlying white matter. As the isosurface intensity approached that of the white matter, the extracted surface lay closer to the grey/white boundary, and the deeper structure of the superficial gyri became apparent. The extraction did not, however, appear as uniform as expected after the RFI correction. CONCLUSION: Through posthoc averaging we produced a MRI volume whose characteristics permitted the extraction of nested isosurfaces. The extracted surfaces tracked the changing features of the grey/CSF/ and grey/white interfaces as the isovalue target was incrementally increased, and the generated thickness map was superficially similar to that of von Economo. With improved RFI correction and surface extraction techniques, it ought to be possible to produce reliable cortical thickness maps using nested isosurfaces.