Detection of Schizophrenia in brain MR images based on segmented ventricle region and deep belief networks

Schizophrenia (SZ) is a brain disorder that affects manifold cognitive domains which include language, memory, attention and executive functions. Magnetic resonance imaging is used to capture structural abnormalities in brain regions. Many studies have indicated brain region volume changes in Schizophrenia patients. In this work, an attempt has been made to analyze the schizophrenic subjects based on ventricle region using deep belief networks (DBNs). The effectiveness of the proposed method is evaluated on center of biomedical research excellence database. Initially, the ventricle region from the normal and SZ images is segmented using multiplicative intrinsic component optimization method. The DBN using different learning algorithms such as stochastic gradient descent (SGD), adaptive gradient (Adagrad) and root-mean-square propagation (RMSProp) is used to train the considered region. The effect of number of layers and the learning algorithm used to discriminate the normal and SZ subjects in DBN is analyzed. Then, the DBN model is evaluated on test set using accuracy, precision, sensitivity and specificity measures. The classification performance of the proposed system is analyzed using receiver operating characteristic curve. Further, the performance of DBN based on segmented ventricle is compared with region of interest (ROI) image that consists of ventricle along with other tissues. Here threefold validations are carried out for the same set of images. Results show that DBN with RMSProp learning with two hidden layers gives better performance compared to other learning methods such as SGD and Adagrad. In addition, DBN on segmented ventricle region gives least error compared to ROI image. DBN with segmented ventricle provides better classification accuracy of 90%. The proposed method achieves high area under the curve (0.899) for the segmented ventricle image, which clearly demonstrates its effectiveness. Thus, the DBN with RMSProp learning-based classification of segmented ventricle could be used as a supplement in the investigation of Schizophrenia.