原发性肝癌CT图像纹理分析-现代仪器与医疗.PDF

研制开发 2013 年 第 19 卷 第 4 期 原发性肝癌CT图像纹理分析* Hepatocellular Carcinoma CT Image Texture analysis 摘  要  目的 探索图像纹理分析方法对原发性肝癌CT 图像的识别能力。方法 首 先从一阶统计特征、灰度共生矩阵、灰度行程矩阵3 方面提取正常肝脏和原发性肝 癌CT 图像的纹理特征, 然后采用t 检验进行特征选择, 最后利用BP 神经网络对保留 郝 涛1 2    ** 张 智 ***  的特征进行分类识别。结果 BP 神经网络对正常肝的识别率是 100 ±0.00%, 对原发 Hao Tao Zhang Zhi 性肝癌的识别率是93.60 ±5.35%。结论 BP 神经网络经设计优化后能达到较高的识 别准确率, 对于原发性肝癌的计算机辅助诊断具有一定实际意义和理论价值。 (1.湖北省文理学院附属襄阳中 关键词  原发性肝癌 特征提取 特征选择 BP 神经网络 心医院 襄阳 441021；2.华 中科技大学生命科学与技术学 Abstract Purpose Explore the recognition capability of the texture analysis of Hepatocel- 院 武汉 430074） lular Carcinoma CT images. Methods The normal liver and Hepatocellular Carcinoma CT (1.Xiangyng Central Hospital Xiangyang 441021;2.Huazhong images texture features extraction is based on first order statistics, gray-level co-occurrence University of Science and Technology matrix and gray-level run-length matrix, and the feature selection based on t-test is applied. College of Life Science Wuhan Finally, classify and recognize the characteristics of reservations by BP neural network. 430074) Results the BP classifier produces about accuracy rate of 100±0.0% for normal liver, 93.60±5.35% for Hepatocellular Carcinoma. Conclusion BP neural network optimized to achieve high recognition accuracy, the study is consider to be helpful for practical aided- diagnosis ． Keywords Hepatocellular Carcinoma Feature extraction Feature selection BP neural