Development of a New Sub-Volume Probabilistic Atlas of the Elderly Brain: Focus on the Frontal Lobe

Martina Ballmaier, Ivo Dinov, Daniel Pham, Helen Lavretsky, Arthur W Toga, Anand Kumar

LONI, Department of Neurology, UCLA School of Medicine, Los Angeles, CA 90095, USA.
Neuropsychiatric Institute, UCLA School of Medicine, Los Angeles, CA

Subject: Modeling & Analysis

Abstract
The development of structural probabilistic brain atlases provides the framework for new analytic methods capable of understanding normal variability within a particular population as well as differentiating between normal and diseased populations. The goal of this study is to map normal and depression-specific variability of the elderly brain by introducing a new sub-volume probabilistic atlas (SVPA) method (Dinov et al., 2000). The frontal lobe was chosen as the region of interest, since initial MRI results demonstrated loss in prefrontal lobe volume in late-onset depression in the absence of generalized atrophy (Kumar et al., 1998). More recently, in addition to a reduction in frontal lobe volumes, an attenuation of the normal right-left asymmetry was also observed, with frontal asymmetry decreasing with increasing severity of depression (Kumar et al., 2000).

Methods
In this paper we begin by constructing the left and right frontal lobe parts of the elderly SVPA atlas using T1-weighted MRI volumes of 10 elderly normal control subjects. The probabilistic distribution of global tissue segmentation maps for each voxel in the frontal lobe are computed from manually labeled individual brains and combined to form probabilistic (cloud-like) fields. The constructed SVPA atlas was then used to automatically extract the white matter, gray matter and CSF volumes from a set of MRI volumes. One-way analysis of variance (ANOVA) was employed to examine differences between the automatically and manually obtained tissue volumes for the left and right frontal lobes of the 10 subjects initially used to construct the atlas.

Results
No significant effects were found between the SVPA automatic volume extraction approach and the manually drawn masks (F-statistics=2.008; P=0.13), suggesting that the SVPA atlas method can be reliably applied to compute regional white matter and gray matter volumes in a robust and automated manner.

Conclusions
The construction and validation of the SVPA approach on the frontal lobe implies that this methodology can be used to automatically extract volumetric measures for a variety of regions of interest, thus providing the investigator with robust and fast tools for assessing temporal, population or treatment effects on the elderly brain. Another potential usage of the constructed SVPA atlas includes its applications to analyzing regionally the statistical differences in anatomically constrained functional images. In these situations one would apply the underlying probability maps in computing more accurate statistics of group differences.

Kumar A, Jin Z, Bilker W, Udupa J, Gottlieb G (1998). Late-onset minor and major depression: early evidence for common neuroanatomical substrates detected by using MRI. Proc Natl Acad Sci USA 95: 7654-7658.

Dinov I, Mega MS, Thompson PM, Lee L, Woods RP, Holmes CJ, Sumners DL, Toga AW (2000) Analyzing functional brain images in a probabilistic atlas: a validation of subvolume thresholding. Neuroimage 24(1): 128-138.