The Role of the Textural Microstructure Co-occurrence Matrices in the Automatic Detection of the Cirrhosis Severity Grades from Ultrasound Images

Abstract

Cirrhosis is a lethal disease that can also precede liver cancer. The golden standard for diagnosing this affection, the biopsy, is invasive and dangerous. The noninvasive detection of the cirrhosis severity grades is a major challenge in these conditions. The objective of our research is to automatically discover the cirrhosis evolution stages from ultrasound images, through unsupervised classification techniques, based on textural features. In this work, the role of the textural microstructure co-occurrence matrices in the detection of the cirrhosis severity grades was studied. The textural features were provided at the input of unsupervised classification methods.  Through specific techniques, the relevant textural features for class separation and their specific values for each severity grade were determined. The method was validated through cluster visualizations and supervised classification, the resulted accuracy being above 90%. The effect of the Principal Component Analysis (PCA) technique upon discovery process of the cirrhosis grades was also studied.