A new system to segment and label CT/MRI brain slices using feature ex
traction and unsupervised clustering is presented. Each volume element
(voxel) is assigned a feature pattern consisting of a scaled family o
f differential geometrical invariant features. The invariant feature p
attern is then assigned to a specific region using a two-stage neural
network system. The first stage is a self-organizing principal compone
nts analysis (SOPCA) network that is used to project the feature vecto
r onto its leading principal axes found by using principal components
analysis. This step provides an effective basis for feature extraction
. The second stage consists of a self-organizing feature map (SOFM) wh
ich automatically clusters the input vector into different regions. A
3D connected component labeling algorithm is then applied to ensure re
gion connectivity. We demonstrate the power of this approach to volume
segmentation of medical images. (C) 1997 Elsevier Science B.V.