Improving the Textural Model of the Hepatocellular Carcinoma Using Dimensionality Reduction Methods

The diagnosis of the malignant tumors is one of the major issues in nowadays research. We aim to elaborate a computerized, non-invasive method, for detecting the Hepatocellular Carcinoma (HCC), based on information from ultrasound images. For performing automatic detection of HCC, we elaborated the imagistic textural model of this malignant tumor, consisting in the relevant textural features and in their specific values for HCC. In this paper, we enhance the imagistic textural model of HCC, by using dimensionality reduction methods, the final purpose being that of obtaining an improvement of the classification process. Principal Component Analysis is a well known dimensionality reduction method, which maps the data into a new space, lower in dimension by finding the principal directions of variation. We experiment this method, studying its influence on the automatic diagnosis accuracy and we also try to combine it with Correlation based Feature Selection, for adding class label sensitivity.