Diagnosis of Bipolar Disorder Based on Principal Component Analysis and SVM

In this paper, we present a computer aided diagnosis tool to discriminate between healthy subjects and patients with bipolar disorder (BD) using the deformation Jacobian obtained during the registration process of their structural (T1) magnetic resonance imaging (MRI) acquisition. To be able to compare MRI images of different subjects, first we need to register those images to a common template. We perform a two step registration: affine and nonlinear. To avoid the curse of dimensionality, we reduce the dimensions of each image performing a feature selection and linear transformation process. First, we select the voxels with the square difference above a given threshold and, second, we apply principal component analysis (PCA) for further dimensionality reduction of the selected features. Results are obtained over an on-going study in Hospital de Santiago Apostol collecting anatomical T1-weighted MRI volumes from healthy control subjects and BD patients. We perform several experiments with different thresholds achieving up to 90% of accuracy when classifying the selected features with linear support vector machines (SVM) classifier.