基于改进的多尺度深度残差网络肝包虫超声影像诊断方法

Ultrasonic image diagnosis of hepatic hydatid based on improved multiscale depth residual network

为了提高肝包虫病的诊断效率和诊断精度,分析肝包虫病超声图像影像学特点.使用自动预处理方法进行图像预处理,收集符合模型输入规范的肝包虫病超声影像数据,构建中等规模的地区包虫病超声影像数据集.此外,针对肝包虫病超声图像语义信息提取困难的问题,提出一种基于自注意力机制的多尺度金字塔分型网络,与常见的基准网络相比,所提出的模型能够达到最好的分类效果,准确率为91.18%.最后,使用梯度加权类激活映射图进行可视化分析,根据实验结果探索五类肝包虫病在超声影像组学中的规律,为肝包虫病的临床诊断提供可靠的参考依据.

In order to improve the diagnostic efficiency and accuracy of hepatic hydatidosis,the imaging characteristics of hepatic hydatidosis were analyzed.In this study,the automatic preprocessing method was used for image preprocessing,and the ultrasonic image data of hepatic hydatidosis that met the model input specifications were collected to construct a medium-sized regional ultrasonic image data set of hydatidosis.In addition,in view of the difficulty in extracting semantic information from ultrasonic images of hepatic hydatid disease,a multi-scale pyramid classification network based on self attention mechanism is proposed.Compared with common benchmark networks,the model can achieve the best classification effect,with an accuracy rate of 91.18%.Finally,the gradient weighted class activation map was used for visual analysis to explore the rules of five types of hepatic echinococcosis in ultrasound imaging omics according to the experimental results,and provide a reliable reference for the clinical diagnosis of hepatic echinococcosis.