Figure 1.

Flowchart of the proposed algorithm. (A) Guided by a preliminary CSF skeleton (red), sampling points in the cortex were identified at a depth of 2 mm. Euclidian distance maps respecting cortical folding were computed and binarized at D _parcel mm. Each parcel underwent a selective morphological dilation adding likely CSF candidates. (B) In each parcel, nonparametric mean shift clustering identified peaks of the intensity histogram, assigning decision boundaries to the sampling points. Final whole‐brain classification was obtained after interpolating decision boundaries from sampling voxels to individual brain voxels, using an inverse distance weighting. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]