摘要
目的采用多重分形谱纹理分析法以及模式识别技术,建立一种识别正常肝脏和不同程度脂肪肝B超图像的计算机辅助诊断方法。方法通过提取每幅B超图像多重分形谱曲线的奇异标度差和多重谱面积两个特征量,再结合近远场灰度比特征量,组成三维特征矢量输入BP人工神经网络进行分类识别。结果正常肝脏正确识别率96.00%,轻度脂肪肝识别率80.00%,中度脂肪肝识别率88.00%,重度脂肪肝识别率92.00%。结论特征矢量结合BP网络的识别方法在B超图像上能较好地识别肝脏脂肪化程度,可作为一种辅助诊断方法。
Objective To develop a recognition method of liver steatosis degree on type-B ultrasonic images based on multi-fractal spectrum texture analysis method and pattern recognition. Methods Features of singularity strength width and multi- spectrum area were extracted from the curve of multi-fractal spectrum of each liver ultrasonic images. These two features and the feature of mean intensity ratio comprised a three-dimensional feature vector, which would be classified by BP neural network. Results The classification accuracy was 96.00% for normal liver, 80.00% for mild fatty liver, 88.00% for moderate fatty liver and 92.00% for severe fatty liver. Conclusion Feature vector combined with BP neural network can identify the steatosis degree of liver on the ultrasonic images and can be used as an assistant diagnostic method.
出处
《中国医学影像技术》
CSCD
北大核心
2009年第7期1289-1292,共4页
Chinese Journal of Medical Imaging Technology
关键词
超声检查
脂肪肝
多重分形谱法
计算机辅助诊断
人工神经网络
Uhrasonography
Fatty liver
Multi-fractal spectrum
Compute-aided diagnosis
Artificial neural network
