Three-dimensional segmentation of CT images using neural network

The segmentation is an important part of the automatic or semi-automatic analysis systems of CT images. In this paper, the scheme of segmentation directly based on three-dimensional gray volume is presented. The CT slices digitized by scanner or digital camera are normalized and reconstructed to output the 3D gray volume. The 3D feature extraction reduces the correlation among voxels and emphasizes the continuity of voxels of the same tissue and the discontinuity of voxels between different tissues. A neural network of feature map classifier is used to cluster the feature volumes. The results show the difference between the segmentation based on 3D and 2D features.