Computer aided diagnosis of liver ultrasound images using K-nearest neighbour approach

Fatty liver disease has become a major public health concern due to its high prevalence. In order to diagnose fatty liver diseases, Computer Aided Diagnosis (CAD) systems have been proposed by various researchers in the recent past. Despite the fact that much work has been done in the domain of CAD of liver ultrasound images, it remains a topic of considerable importance due to the rising prevalence of fatty liver around the world. As a result, in biomedical imaging, a quantitative method with higher accuracy is still needed to assist radiologists in making better diagnoses. In this paper, a CAD method for the classification of liver ultrasound images is proposed. Texture features using seven different texture models have been extracted from region of interest. The extracted forty-five texture features are passed to K-Nearest Neighbour (K-NN) classifier for the classification purpose. Further, in this work, 10 different distance metrics have been evaluated to obtain the maximum accuracy. Additional experiments were also conducted to find the best value of K which reduces the number of errors we encounter. It has been found from the experimental results that the proposed method using K-NN classifier yielded best results with City Block distance metric and a value of k as 7. The overall accuracy obtained with these parameter settings is 96.7% with area under ROC curve as 0.99. The experiments demonstrated that the proposed method gave better results for the classification of liver ultrasound images into two categories.