论文部分内容阅读
目的:寻找适合识别正常肝、原发性肝癌和肝血管瘤CT图像的特征向量。方法:从一阶统计特征、灰度共生矩阵、灰度行程矩阵三方面提取正常肝、原发性肝癌和肝血管瘤CT图像的纹理特征,然后采用t检验进行特征选择,最后利用神经网络识别系统,把保留的特征作为输入量,对正常肝和原发性肝癌进行识别。结果:所设计的BP神经网络,对正常肝(100±0.00)%、原发性肝癌(91.08±6.96)%,对肝血管瘤(85.76±12.51)%。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的实际意义和理论价值。
Objective: To find suitable eigenvectors for CT images of normal liver, primary liver cancer and hepatic hemangioma. Methods: Texture features of CT images of normal liver, primary hepatocellular carcinoma and hepatic hemangioma were extracted from first-order statistical features, gray level co-occurrence matrix and gray stroke matrix. Then the feature was selected by t-test. Finally, neural network was used to identify The system, using the characteristics of retention as input, identifies normal liver and primary liver cancer. Results: The BP neural network was designed for normal liver (100 ± 0.00)%, primary liver cancer (91.08 ± 6.96)% and hepatic hemangioma (85.76 ± 12.51)%. Conclusion: BP neural network can achieve high recognition accuracy after being designed and optimized, and has certain practical and theoretical value for computer-aided diagnosis of primary liver cancer.