基于B超图像分析的脂肪肝辅助诊断方法研究  

Computer-aided Diagnosis of Fatty Liver Based on Ultrasonic Images

作　　者：汪小毅[1] 林江莉[1] 李德玉[1] 汪天富[1] 郑昌琼[1] 程印蓉[2] 

机构地区：[1]四川大学373信箱生物医学工程中心,成都610065 [2]成都市第一人民医院超声科,成都610016

出　　处：《生物医学工程学杂志》2006年第4期726-729,共4页Journal of Biomedical Engineering

基　　金：四川省青年科技基金资助项目(05ZQ026-019);四川省应用基础研究资助项目(03JY029-072-2)

摘　　要：本研究为B超诊断脂肪肝建立计算机辅助诊断手段。通过分析正常肝和脂肪肝B超图像的图像特征,包括图像的近远场灰度比特征,以及灰度共生矩阵的角二阶矩、熵和反差分矩统计特征,组成特征矢量,再分别用κ-平均聚类算法、自组织特征映射人工神经网络和反向传播人工神经网络对特征矢量进行分类处理。κ-平均聚类算法对正常肝的识别率为100%,对脂肪肝的识别正确率为63.6%;自组织特征映射人工神经网络对正常肝的识别正确率达100%,对脂肪肝的识别正确率达93.94%;反向传播人工神经网络对正常肝和脂肪肝的识别率均为100%。本文建立的方法能较肉眼更精确地反映正常肝和脂肪肝B超图像的特征,如果再结合医生的临床经验能大大提高脂肪肝的诊断准确性。This study aims to provide a computer-aided method for the diagnosis of fatty liver by B-scan ultrasonic imaging. Fatty liver is referred to the infiltration of triglycerides and other fats of the liver cells ;which affected the texture of liver tissue. In this paper, some features including mean intensity ratio, as well asi angular second moment, entropy and inverse differential moment of gray level co-occurrence matrix were extracted form Bscan ultrasonic liver images. Feature vectors which indicated two classes of images were created with the four features. Then we used re-means clustering algorithm, self-organized feature mapping （SOFM） artificial neural network and back-propagation （BP） artificial neural network to classify these vectors. The accuracy rate of κ- means clustering algorithm was 100% for normal liver and 63. 6% for fatty liver. The results of SOFM neural network showed that the accuracy rate was 84.8% for normal liver and 90. 9% for fatty liver. The accuracy rate of neural network was 100% both for normal liver and fatty liver. This technology could detect the characteristics of B-scan images of normal liver and fatty liver more accurately. It could greatly improve the accuracy of the diagnosis of fatty liver.

关 键 词：脂肪肝 图像分析 辅助诊断 人工神经网络 

分 类 号：R318[医药卫生—生物医学工程] R575.5[医药卫生—基础医学]