This week I talked at the Biophysics department, KTH about the interpretation of CT scans of fish brains.  We are working on a project looking at brains and social behavior, in small fish known as guppies, using artificial selection. And my primary role is to find ways of identifying differences in three-dimensional brain scans.

Here I'll describe a few of the steps we take. First all of the brain samples are arranged and scanned in a transparent plastic tube. The whole scan is then divided into portions. Each portion contains the brain of one individual. With a down-sampling of 4x4x4 and after removing the artifacts, each brain image consist of around 42 million voxels (a voxel is a three dimensional pixel). An example of guppy brain scan is shown in the video below.

My aim is to identify regions of interest, which change significantly under different treatments. Two different techniques are employed to assist further interpretation: a parametric and a non-parametric one.

In the parametric approach, our guppy brain expert, Alex Kotrschal, constructed a brain-atlas of the important different regions. This reference image is constructed using a blend of segmentation, smoothing and manual annotation. Then this reference image is aligned ('registered') to the different samples, so that the regions of interest are mapped to each individual. This gives us the precise size in volume, shape and morphological information of these regions in each sample.  An example is given below.

The non-parametric approach is entirely based on processing the intensities of the scans using multivariate pattern analysis. For this, a combination of Support Vector Machines and permutation testing is employed. This results in a so-called P-map: a map of statistical P-values for every part of the brain testing the null hypotheses that brains from two treatment groups are not different. We can plot these P-values for every voxel in a three dimensional brain structure. Below is one example

While the multitude of the P-values corrupts direct statistical interpretation, the P-map is used to indicate where parts of the brain might differ.

The parametric and non-parametric tools indicate where differences occur. We are currently writing up an article on the difference between the detailed brain structure of guppies bred for small and large brains.