What am I looking at?

You are looking at the relative contribution of different cognitive concepts for a single location in the brain, an x,y,z coordinate called a "voxel." The size of any word in the cloud reflects the relative contribution of the cognitive concept for the model in that voxel location.

What do you mean "relative contribution?"

We used a bunch a brain maps labeled with cognitive concepts to generate a sparse Logistic Regression model at each voxel (think an x,y,z coordinate in a 3D brain map), meaning that we found an optimal weighting of each cognitive concept to predict the voxel values across a large set of images. You are looking at the "relative contribution" (e.g., importance) of each concept for a particular voxel, represented by the weight (the beta or regression parameter) produced by the model.

What do these voxel-wise models do?

Having a model at each voxel means that we can use cognitive concepts to predict brain maps, and predict cognitive concepts from a new brain map. The model is sparse because the algorithm sets the weights of as many of the regression parameters (one for each cognitive concept) to zero. Thus, although there are a total 132 cognitive concepts, you will only see a subset in the word cloud.

How do I explore different regions?

If you use the region selector in the bottom left, you will be taken to a random voxel within the region. Some regions are very large, meaning the model looks very different between voxels in the same region, and for this reason we encourage you to reload the same region multiple times to see the variance. Brain regions correspond to randomly selected MNI (x,y,z) coordinates from the AAL2 atlas resampled to 4mm with nearest interpolation. Data and labels were obtained courtesy of NeuroVault. Full script to generate coordinates is available.