CLASSIFICATION OF BRAIN COMPARTMENTS AND HEAD-INJURY LESIONS BY NEURAL NETWORKS APPLIED TO MRI

An automatic, neural network-based approach was applied to segment nor mal brain compartments and lesions on MR images. Two supervised networ ks, backpropagation (BPN) and counterpropagation, and two unsupervised networks, Kohonen learning vector quantizer and analog adaptive reson ance theory, were trained on registered T2-weighted and proton density images. The classes of interest were background, gray matter, white m atter, cerebrospinal fluid, macrocystic encephalomalacia, gliosis, and ''unknown.'' A comprehensive feature vector was chosen to discriminat e these classes. The BPN combined with feature conditioning, multiple discriminant analysis followed by Hotelling transform, produced the mo st accurate and consistent classification results. Classifications of normal brain compartments were generally in agreement with expert inte rpretation of the images. Macrocystic encephalomalacia and gliosis wer e recognized and, except around the periphery, classified in agreement with the clinician's report used to train the neural network.