Neurosynth: Mismatch between image data and activations?




Why is there a discrepancy between ‘activations’ and ‘images’ for the same study? Aren’t both supposed to be voxels in the MNI152 space

I am using the python API for neurosynth.

The code:

dataset = Dataset.load(opj(DATA_DIR, ‘neurosynth’, ‘dataset.pkl’))

activations = dataset.activations
img = dataset.get_image_data(ids=[9065511])

print img.sum()
#prints 1172

#prints (10, 16)

Are activations thresholded?


The image data gives you a vector of voxels, where a value of 1 means the voxel falls within X mm (depending on the radius used at initialization) of a reported activation. The length is equal to the number of voxels within the gray matter mask.

The activation data is just a text table containing information about reported activations. The 16 columns are different variables (e.g., x, y, and z coordinates; detected space; etc.).

The reason the sum is different is that (a) in the image, the activations are convolved with spheres (so the sum will generally be much larger), and (b) it’s a binary map, so if activations overlap, the max value remains 1.