In order to classify the minimal hepatic encephalopathy （MHE） patients from healthy controls, the independent component analysis （ICA） is used to generate the default mode network （DMN） from resting-state functional magnetic resonance imaging （fMRI）. Then a Bayesian voxel- wised method, graphical-model-based multivariate analysis （GAMMA）, is used to explore the associations between abnormal functional integration within DMN and clinical variable. Without any prior knowledge, five machine learning methods, namely, support vector machines （SVMs）, classification and regression trees （ CART ）, logistic regression, the Bayesian network, and C4.5, are applied to the classification. The functional integration patterns were alternative within DMN, which have the power to predict MHE with an accuracy of 98%. The GAMMA method generating functional integration patterns within DMN can become a simple, objective, and common imaging biomarker for detecting MIIE and can serve as a supplement to the existing diagnostic methods.
In order to facilitate the extraction of the default mode network（DMN）, reduce the data complexity of the functional magnetic resonance imaging （fMRI）and overcome the restriction of the linearity of the mixing process encountered with the independent component analysis（ICA）, a framework of dimensionality reduction and nonlinear transformation is proposed. First, the principal component analysis（PCA）is applied to reduce the time dimension 153 594×128 of the fMRI data to 153 594×5 for simplifying complexity computation and obtaining 95% of the information. Secondly, a new kernel-based nonlinear ICA method referred as the kernel ICA（KICA）based on the Gaussian kernel is introduced to analyze the resting-state fMRI data and extract the DMN. Experimental results show that the KICA provides a better performance for the resting-state fMRI data analysis compared with the classical ICA. Furthermore, the DMN is accurately extracted and the noise is reduced.
