A new rapid landmark-based regional MRI segmentation method of the brain

Abstract

Background: Neurodegenerative and cerebrovascular diseases show a distinct distribution of regional atrophy and subcortical lesions. Objective: To develop an easily applicable landmark-based method for segmentation of the brain into the four cerebral lobes from MRI images. Method: The segmentation method relies on a combination of anatomical landmarks and geometrical definitions. It is applied on the surface reconstruction of the MRI volume. The internal borders between the lobes are defined on the axial slices of the brain. The reliability of this method was determined from MRI scans of 10 subjects. To illustrate the use of the method, it was applied to MRI scans of an independent group of 10 healthy elderly subjects and 10 patients with vascular dementia to determine the regional distribution of white matter hyperintensities (WMH). Results: The intra-rater relative error (and intra-class correlation coefficient) of the lobe segmentation ranged from 1.6% to 6.9% (from 0.91 to 0.99). The inter-rater relative error (and intra-class correlation coefficient) ranged from 1.4% to 5.2% (from 0.96 to 0.99). Density of WMH was significantly higher in all four lobes in VD patients compared to controls (p<0.05). Within each group, WMH density was significantly higher in frontal and parietal than in temporal and occipital lobes (p<0.05). Conclusion: This landmark based method can accommodate age and disease-related changes in brain morphology. It may be particularly useful for the study of neurodegenerative and cerebrovascular disease and for the validation of template-based automated techniques.

Introduction

Neuropathological studies suggest a differential susceptibility of cerebral cortical regions for different neurodegenerative processes [1], [2] and cerebro-vascular disease [3], [4]. A major challenge in clinical research on neurodegenerative or cerebrovascular disorders of the brain is the reliable identification of regional pattern of disease-related cerebral atrophy and the regional distribution of vascular cerebral lesions.

To show the differential involvement in-vivo, two basic methodological approaches exist to segment cerebral regions (or calculate lobar volumes) on MRI scans: (a) an automated method relying on template and (b) a manual method based upon landmarks.

There have been several reports on automated methods [5], [6], [7], [8] for quantification of the lobe volumes by first normalizing the magnetic resonance images to the Talairach and Tournoux template [9] and segmenting the brain regions based on the template. The template defines every pixel as a member of a specific region and that segmentation is then transferred to the MRI scan of interest. The advantage of automated methods is speed, reliability, and ease of use by the operator. The disadvantage is that it does not take into account differences in shape and variability of the cortex, which are more pronounced in patient populations. In addition, the normalization process will inevitably cause an averaging process to take place and individual differences may be lost, which can hide differences in structure between normal populations and patient populations.

There are several reports on manual methods for tracing a single or two lobes of the brain, typically the temporal and/or frontal lobe, on a slice by slice basis (see for example [10], [11], [12], [13]) or on a reconstructed 3-D model of the cortex [14]. There is a report on the regional segmentation of the five lobes in the cortex [15]. These methods trace each individual region and segmentation methods are used to delineate brain from CSF. The main advantage of these methods is that differences in shape and variability can be accounted for in application to normal subjects and patient populations. The disadvantages are that it can be time consuming, landmarks can be difficult to detect, and the reliability between and within raters over time has to be confirmed.

The objective of this work is the development of a rapid and easy to use landmark-based technique for regional segmentation of the brain into different lobes. The segmentation is manual, with the traces done with reference to prominent sulcal points and drawing well-defined lines when no sulci and/or gyri reference landmarks are available. We will demonstrate the usefulness of the technique with its application to regional measures of white matter hyperintensities (WMH) in a group of patients with vascular dementia and an age-matched healthy control group.