A New Method for Quantification of Age-Related Brain Changes

Vassili Kovalev and Frithjof Kruggel

A new method is proposed for quantification of age-related brain changes. The method includes calculating 3D volumetric texture descriptors, extracting principal components, and assessing the significance of brain changes using multivariate analysis techniques. Structural changes were evaluated using high resolution anatomical MRI-T1 brain images of a group of 152 healthy subjects aged from 18 to 70 years (76 males and 76 females). The Talairach parcellation system was applied to study normal brain aging on four scale levels: the whole cerebrum, the nine coronal sections, the twelve axial sections, and 108 box-shaped sections resulting from both subdivisions. Statistical analysis has revealed significant brain deteriorations with age at different scale levels. Most of the brain regions are affected with a slight predominance in the frontal lobes. We concluded that 3D texture analysis followed by statistical evaluation procedures is a robust technique for detecting age-related changes in the anatomical MR images of the human brain.

Keywords: brain, ageing, 3D volumetric texture, co-occurrence

age-image00

 

Fig. 1. General scheme of quantification of brain changes based on 3D volumetric texture analysis using multi-sort co-occurrence matrices followed by corresponding multivariate (ANOVA) statistical assessment of structural brain image features.

Note that structural features are derived from original co-occurrence matrices directly using Principal Component Analysis method. Such features (i.e., the resultant Principal Components) are uncorrelated by definition and constitute objective, unbiased, and minimal feature set as opposed to the subjective and mutually-correlated Haralick’s features which are still used by researchers for several decades since early ages of the image analysis methods. (See other sections of the site for a technique which allows mapping the significant principal components back to co-occurrence matrix elements and further to original image pixels/areas for highlighting the key image regions where discovered findings are coming from).

age-image01

Fig. 2. Example slices of MRI-T1 brain images of young, mid-aged, and aged healthy subjects.

age-image02

Fig. 3. A regression diagram plotting the predicted age vs. actual age for the left (green circles) and right (red triangles) brain hemispheres (152 subjects aged 18-70 years).

age-image03

Fig. 4. Talairach brain partitioning for coronal slices (top) and significance of age-related changes in corresponding coronal brain sections on the bottom (significance scores are based on the analysis of 152 subjects aged 18-70 years).

age-image04

Fig. 5. Talairach brain partitioning for axial slices (top) and significance of age-related changes in corresponding axial brain sections on the bottom (significance scores are based on the analysis of 152 subjects aged 18-70 years).

age-image05

Fig. 6. Significance of age-related brain changes in Talairach corono-axial blocks (152 subjects aged 18-70 years).  Significance scores are coded using the color scale provided underneath.