Analysis of Ultrasound Images of Superficial Organs Based on the Combination of Global Features and Local Features

Abstract

Ultrasound imaging has become the preferred method for early detection of superficial organ lesions due to its non-invasive, economical, convenient, and radiation-free features. First, in the superficial organ ultrasound image analysis module based on local features, using Dense Sift as the local feature descriptor, the similarity sequence of the local features of the input image and the images in the sample library is calculated by the BOW algorithm and sorted in descending order of similarity. Secondly, using the wavelet transform’s local feature information representation capabilities in the time and frequency domains, the wavelet transform is performed on the small and identical feature information contained in the image to remove redundant information in each feature map to obtain a salient part of the image’s local features. Finally, through the analysis of elastic ultrasound images, a quantitative index that can be used to evaluate ultrasound images of superficial organs is proposed, which has higher accuracy and reliability than the current clinical methods for evaluating ultrasound images of superficial organs. By analyzing the features of superficial organ ultrasound images, it is proposed that the overall features of superficial organ ultrasound images are more conducive to distinguishing benign and malignant lesions than local features. Based on the initial localization of the lesion, the global features of the superficial organ ultrasound image were combined with the local features of the B-ultrasound image, and the method of combining global features with local features was used to classify and achieved good results.