Self-organized feature detection and segmentation of magnetic resonance images

Unsupervised, competitive learning was applied to a self-organizing map for feature detection, and tissue segmentation of magnetic resonance images of the brain. The multi-spectral input data were the individual pixel intensities from T1-weighted, T2-weighted, and proton density MR images. The technique trained quickly and generalized to other slices from the same study. Pathologies were detected, and white matter, gray matter, and cerebral spinal fluid were segmented.<
>