PRUNING DATASETS IN DISCRIMINANT ANALYSIS: A DTI STUDY TO SCHIZOPHRENIA

A comparative study is commonly performed by means of pre-defined or expert selected region of interest (ROI)-analysis or voxel based analysis (VBA). In contrast to these methods, correlations within the data can be modeled by using principal component analysis (PCA) and linear discriminant analysis (LDA). The mapping computed by PCA/LDA is displayed to identify the discriminative regions. A technique called ´pruning´ is introduced to iteratively discard misclassified subjects from the cohort. These subjects reside in the region in feature space where the classes are overlapping. As the exact cause of this overlapping is unknown, it is preferable to base the mapping merely on representative prototypes, residing in the nonoverlapping parts of the feature space. After pruning the PCA/LDA mapping, a more pronounced decrease in FA in larger parts of the corpus callosum was observed, compared to conventional VBA