A novel approach for cirrhosis recognition via improved LBP algorithm and dictionary learning

Abstract Early diagnosis of cirrhosis has been increasing the interest of medical specialists and engineers. Cirrhosis diagnosis is difficult to distinguish with naked eyes, and it depends on subjectivity of physicians largely. In this paper, the improved Local Binary Pattern(LBP) algorithm called T-LBP(total LBP) and its corresponding T-LBPs(T-LBP spectrum) feature were proposed to describe cirrhosis texture and to solve the edge blurring problem caused by cirrhosis effectively. We applied fusion of T-LBPs, two-dimensional Gabor transform and K-SVD(single value decomposition which generalizes K-means clustering process) based dictionary learning methods in cirrhosis recognition of ultrasound(US) images for the first time. Advantages of proposed algorithms include, firstly, to our best knowledge, proposed T-LBPs feature outperforms the traditional features using support vector machine(SVM), and it has also been proved that the consuming time of kernel extreme learning machine(kernel-ELM) is less than that of basic ELM in this issue; secondly, dictionary learning based recognition method through T-LBP has obtained the highest recognition rate of 99.69% compared with state-of-the-art methods, and dictionary updating error decreased sharply via T-LBP. Therefore, the proposed algorithm will contribute to the clinical cirrhosis diagnosis.