Distinguishing schizophrenic patients from healthy controls based on MRI data: A tensor linear discriminant approach

• Published in: 2010 9th IEEE International Conference on Cognitive Informatics pp. 524 - 529
• Main Authors: Linchuan Zhang, Lubin Wang, Hui Shen, Dewen Hu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2010
• Subjects: Accuracy; Classification algorithms; Feature extraction; Magnetic resonance imaging; MRI; nearest-neighbor classifier(NNC); Pattern recognition; Schizophrenic; Tensile stress; Tensor Linear Discriminant Analysis (Tensor LDA); Training
• ISBN: 9781424480418; 1424480418
• DOI: 10.1109/COGINF.2010.5599683

Recently, there are available laboratory procedures providing useful information to psychiatric diagnostic systems. In this paper, a tensor-based pattern recognition system was used to classify schizophrenic patients and healthy controls. The novel tensor approach is an extension of linear discriminant analysis (LDA). In this method, each subject' structure MRI image was viewed as a tensor sample. After splitting samples into training and testing data, we obtained a series of projecting matrix through Tensor LDAalgorithm, and the feature matrix obtained can be used in the testing data to get the class labels. The performance of our system was tested by the leave-one-out cross-validation strategy. Experimental results showed that the sensitivity, specificity, and overall classification accuracy of our system were 86.36%, 94.44%, and 90%, respectively. Moreover, we compared the classification of Tensor LDAwith the traditional LDA. The results showed that the tensor method outperformed on this task than the traditional LDA.