Classification of liver diseases using ultrasound images based on feature combination

In this paper a computer-aided diagnostic system for classification of normal, fatty and cirrhotic liver ultrasound images using feature combination is proposed. The features including spatial domain and transform domain features. Comparative studies have shown that both spatial-domain based and transform-domain based features have good effects on classification. So, to have them both in the CAD system, it is preferred to have combination of these two feature spaces. Here we have extracted gray level cooccurrence matrix features as spatial domain feature and also energy and energy deviation of 2-D WPT and 2-D Gabor filter banks sub-images as transform domain features, then three feature vectors have been combined to achieve classification accuracy of 96.1% by K Nearest neighbor (KNN) classifier.