METHOD FOR EXTRACTING SIGNIFICANT TEXTURE FEATURES OF B-MODE ULTRASOUND IMAGES AND APPLICATION THEREFOR

摘要

A method for extracting significant texture features of B-mode ultrasound images and an application therefor, disclosing a new channel attention mechanism network and context-activated residual network, effectively modelling B-mode ultrasound liver fibrosis texture information, the networks using global context information to enhance important texture features and suppress useless texture features, such that the deep residual network captures more salient texture information in B-mode ultrasound images. The process is divided into two stages: training and testing. In the training stage, B-mode ultrasound images are used as the input, and pathological results of liver puncture biopsies are used as labels for training the context-activated residual network. In the testing stage, a B-mode ultrasound image block is inputted into the trained liver fibrosis non-invasive diagnostic model to obtain a liver fibrosis grade of each ultrasound image. The present method implements rapid and accurate estimation of the liver fibrosis grade of B-mode ultrasound images from a data-driven perspective.