Multi-scale information with attention integration for classification of liver fibrosis in B-mode US image

•A novel network for the classification of liver fibrosis into five levels is proposed which can effectively improve the classification performance.•The equipped pyramid structure uses a large receptive field to capture the detailed information from the vague and low contrast and signal-to-noise ratio US images.•The proposed local consistency attention module can suppress the distractors and enhance the representation of the feature.•Experimental results on the liver fibrosis dataset demonstrate the effectiveness of the proposed network with a local consistency attention mechanism.