Automatic selection of ROIs using a model-based clustering approach

This paper presents a new method for automatic selection of Regions of Interest (ROIs) of functional brain images based on a Gaussian Mixture Model (GMM). This method allows avoiding the so-called small sample size problem in the construction of a CAD system that performs the automatic diagnosis of Alzheimers disease (AD). First we generate an image that holds the differences between normal and AD subjects and then, we model the ROIs from this image by using GMM and the Expectation Maximization algorithm. These regions are used to select a reduced set of features from the activation map of each patient and allow us to train statistical classifiers such as Support Vector Machines (SVMs). We have tested this approach on a SPECT images database and the accuracy rate achieved by the CAD system was 94.5%. This value significantly improves the results obtained by previously developed approaches.