| Because ultrasound imaging is a simple, reliable and non-invasive modality, it is particularly indicated for assessing the liver fibrosis stage. An ultrasound evaluation of the liver fibrosis stage of chronic liver disease can be performed by assessing various ultrasound factors such as the liver size, the bluntness of the liver edge, the coarseness of the liver parenchyma, nodularity of the liver surface, the size of the lymph nodes around the hepatic artery, the irregularity and narrowness of the inferior vena cava, .... However, the conventional definition of the fibrosis stage of the liver based on evaluation of these ultrasound factors is imperfect and lacks accuracy and reliability. The present work proposes to identify the liver fibrosis stage from a characterisation of the perfusion from ultrasound images. A liver disease leads to changes in the organ perfusion that can be emphasised on ultrasound images by injecting contrast agent. In the evolution of the disease, 5 stages are usually defined and a visual evaluation of these stages is often difficult. In the present study, we propose to characterize the evolution of the perfusion when the contrast agent is injected from image texture analysis and ultrasound tissue characterization techniques. Texture analysis consists in extracting parameters characterizing the arrangements of the more or less regular patterns that constitute the image. Different texture analysis techniques are implemented: statistical methods (co-occurrences matrices), model-based methods, to provide a set of parameters that is a signature of the tissue pathological state. To validate this approach, a set of 21 sequences of images livers at different pathological stages is used. The sequences represent at the beginning the liver without contrast agent and about 20 s later the contrast agent is fully distributed in the tissue. A window of homogeneous tissue is extracted each second for processing. 74 textural features were extracted from 8 different texture analysis methods and were calculated on each image. The features presenting an intercorrelation coefficient larger than 0.95 were eliminated and 30 features remain for the statistical analysis. 8 features were calculated to represent the evolution of each textural parameter during contrast agent diffusion. The results show that the texture analysis of images corresponding to the fully distributed contrast agent is more relevant to identify the stage of fibrosis than the features characterizing the evolution of the texture. In particular, a classification of the samples in two classes (the 2 highest pathological grades versus the 3 lowest) exhibits a sensitivity of 63 % and a specificity of 91 %. The statistical significance was set to p<0.05. These results show the potentiality of combining contrast agent and texture analysis on ultrasound images. A study on a larger number of samples will be conducted. |