Detecting the Evolution Phases of Hepatocellular Carcinoma from Ultrasound Images, Using Generalized Co-Occurrence Matrices

The hepatocellular carcinoma (HCC) is the most frequent malignant liver tumor. Detecting the HCC evolution phases represents an important research challenge. Our aim is to discover the evolution stages of the HCC tumor, through unsupervised classification techniques, using texture-based image analysis methods. We defined the textural model of these phases, consisting of the most relevant textural features for their characterization and of the specific values of these features: arithmetic mean, standard deviation and probability distribution. The role of the features derived from generalized co-occurrence matrices was emphasized in this work. The obtained results were validated through supervised classification methods, such as Multilayer Perceptron (MLP), Support Vector Machines (SVM), decision trees, respectively multi-class meta-classifiers, in combination with the basic learners, yielding classification accuracy above 95%. The method of Self Organizing Maps (SOM) was also employed for data representation.