Texture analysis of liver fibrosis microscopic images: a study on the effect of biomarkers

Abstract

Chronic hepatic injury results in liver fibrosis with eventual progression to irreversible cirrhosis. Liver fibrogenesis involves the activation of the quiescent hepatic stellate cell into an activated myofibroblast that is characterized by α-smooth muscle actin (α-SMA) expression and the production of collagens (types I and III). In the present study, rats were randomly divided into three groups: (i) control group, where rats were only treated with a vehicle; (ii) fibrosis group, where rats were treated with carbon tetrachloride (CCl(4)) to induce liver fibrosis; and (iii) silymarin group, where rats were protected with silymarin during CCl(4) treatment. Rats were sacrificed and sections of liver tissue were counterstained with hematoxylin and eosin and Masson's trichrome. Other sections were immunostained using collagens and α-SMA primary antibodies. Fibrosis was confirmed using serum marker measurements. Microscopic images of the stained sections were acquired and digitized. The Biomarker Index of Fibrosis (BIF) was calculated from the images by quantifying the percentage of stained fibers. Statistical methods of texture analysis (TA), namely co-occurrence and run-length matrices, were applied on the digital images followed by classification using agglomerative hierarchical clustering and linear discriminant analysis with cross validation. TA applied on different biomarkers was successful in discriminating between the groups, showing 100% sensitivity and specificity for classification between the control and fibrosis groups using any biomarker. Some classification attempts showed dependence on the biomarker used, especially for classification between the silymarin and fibrosis groups, which showed optimal results using Masson's trichrome. TA results were consistent with both BIF and serum marker measurements.