Brain Atrophy
Traditionally, multiple sclerosis (MS) has
been characterized mainly in terms of the associated inflammation and
demyelination. However, recent evidence demonstrates that axonal
damage, and cerebral atrophy also occur to a significant extent, even
early in the disease. A number of these studies have established
strong correlations between disability and atrophy of the brain stem
or upper spinal cord, but correlations with cerebral atrophy are not
as clear, at least in part due to the difficulty of measuring cerebral
atrophy and to the fact that clinical deficits only loosely reflect
the extent of the underlying cerebral pathology. In as much as
progressive brain atrophy represents irreversible brain damage and
tissue loss, a metric of brain atrophy is a powerful surrogate for the
accumulated burden of disease. We feel that brain atrophy metrics
will prove to be viable, precise, and clinically relevant measures of
disease burden in MS. Atrophy measures will lead to a better
understanding of MS pathology and will have important implications for
disease prognosis, and monitoring treatment effect in clinical trials.
Our technique is based on the automatic estimation the BICCR
(brain to intracranial capacity ratio). This measure depends on the
estimatation of the grey-matter (GM), white-matter (WM) and
cerebrospinal fluid (CSF) intra-dural volumes from in vivo MRI data
(see Fig. below).
- Atrophy = dural_mask{(GM + WM) / (GM + WM + CSF)}
- Methodology
- image intensity non-uniformity correction
(John Sled's N3)
- stereotaxic registration of T1-weighted data
(mritotal)
- inter-modality (T1,T2 and PD) registration (mritoself)
- tissue classification (using fuzzy C-means)
- surface deformation for identification of dural surface
- volume estimation of GM, WM and CSF within the dural mask.
Fig 1: The first row (left to right) shows four transverse slices,
starting just above the cerebellum, and ending above the centrum
semiovale, taken from the stereotaxically resampled patient
T1-weighted MRI volume. The second row shows the contours of the
automatically extracted dural surface overlaid on the corresponding
classified data. The BICCR is computed as the ratio WM+GM/(WM+GM+CSF)
of all voxels within the dural contour. Note that all computations
are done in 3D even though only 2D images are shown here.
Last modified: October 12th, 1999 Comments or suggestions to louis@bic.mni.mcgill.ca