Data Mining in Brain Imaging

  Investigators

    Lazarevic Aleksandar

    Megalooikonomou Vasileos

    Obradovic Zoran

    Pokrajac Dragoljub

  Problem

    To facilitate the process of discovering brain structure-function associations from image and clinical data and to make retrieval of similar brain scans possible, we have developed a statistical method for classification of brain image data based on measures of dissimilarity between three dimensional probability distributions.

  Results

    We propose a method for classifying regions of interest in brain images. The method is based on computing the Mahalanobis distance between a new sample and data sets related to each considered class (condition).    The proposed method is compared to an alternative method for classifying a new subject based on computing the Kullback-Leibler probabilistic distance between distributions estimated through a non-parametric procedure.  In addition, supervised neural network models were compared with previous two methods.