Liu Yiwen, Liu Xiang, Wang Shuohong, Song Jialin, Zhang Jianquan
School of Electric and Electronic Engineering, Shanghai University of Engineering Science, Shanghai, 201620, China.
School of Electric and Electronic Engineering, Shanghai University of Engineering Science, Shanghai, 201620, China.
Comput Biol Med. 2020 Oct;125:104002. doi: 10.1016/j.compbiomed.2020.104002. Epub 2020 Sep 18.
Liver cirrhosis is a common chronic progressive disease with a high mortality rate. The early diagnosis and treatment of liver cirrhosis is an important research subject in the medical field. In this paper, a novel method is proposed for the accurate extraction of the liver capsule and auxiliary diagnosis of cirrhosis based on high frequency ultrasound images. First, a self-developed method is used to extract the predictive capsule of ultrasound images, which involves the detection of liver ascites with sliding windows, image enhancement with multiscale detail and fuzzy set, structure segmentation with morphological processing, and predictive capsule detection with traversal search method. Thereafter, the real capsule is obtained by the gray difference method according to different gray values between the liver capsule region of the original ultrasound images and the set threshold. Finally, according to the analysis of smoothness, as well as the continuity and fluctuation of predictive and real capsule, four novel features called NoL, VoS, CV, and NoF are proposed for the computer auxiliary diagnosis model. This model is designed on the basis of support vector machine and k-means clustering and can classify normal liver and three liver cirrhosis stages. The experimental results reveal that the accuracy of the liver capsule extraction using this model is 95.13% and final classification accuracy of four stages can reach 92.54%, 88.46%, 89.23% and 94.55%, respectively. The results also indicate that the method proposed in this paper can achieve the classification of liver cirrhosis stages much more accurately and efficiently compared with previously utilized methods.
肝硬化是一种常见的慢性进行性疾病，死亡率很高。肝硬化的早期诊断和治疗是医学领域的一个重要研究课题。本文提出了一种基于高频超声图像准确提取肝包膜并辅助诊断肝硬化的新方法。首先，采用自主研发的方法提取超声图像的预测包膜，包括用滑动窗口检测肝腹水、用多尺度细节和模糊集进行图像增强、用形态学处理进行结构分割以及用遍历搜索法检测预测包膜。此后，根据原始超声图像的肝包膜区域与设定阈值之间的不同灰度值，通过灰度差法获得真实包膜。最后，根据对预测包膜和真实包膜的平滑度、连续性和波动情况的分析，为计算机辅助诊断模型提出了四个新特征，即NoL、VoS、CV和NoF。该模型基于支持向量机和k均值聚类设计，可对正常肝脏和三个肝硬化阶段进行分类。实验结果表明，使用该模型提取肝包膜的准确率为95.13%，四个阶段的最终分类准确率分别可达92.54%、88.46%、89.23%和94.55%。结果还表明，与以前使用的方法相比，本文提出的方法能够更准确、高效地实现肝硬化阶段的分类。
