神经网络在超声脂肪肝图像识别中的应用研究

The application research of Neural Network for classifying the ultrasonic liver image

本论文通过利用人工神经网络来研究脂肪肝超声图像的识别算法。通过基于空间灰度独立矩阵,空间频率分解和分形特征的特征提取,采用两层BP神经网络对正常肝脏,轻、中和重度脂肪肝脏共四类超声图像进行分类识别。实验结果表明神经网络分类器对四种肝脏超声图像的分类可以达到95.33%的正确率,其结果对实际辅助诊断很有用。

In this paper, a classification algorithm of ultrasonic fanny liver images is studied by Artificial Neural Network （NN）. The feature extraction is based on the spatial gray-level dependence matrices （SGLDM）, spatial - requency decomposition and fractal feature. Two sets of ultrasonic liver images-normal liver and abnomal liver （including light-fanny liver,moderate-fanny liver and severe- fanny liver） image are successfully classified through NN. The result shows that the NN classifier produces about 95.33% correct classification for the two sets of ultrasonic liver image and our study is considered to be helpful for practical aided-diagnosis.