To evaluate the value of texture features in diagnosing breast tumor of ultrasonic image.Texture measurement were ob-tained from breast tumor in ultrasonic image.Six parameter consisting of mean,standard deviation,smoothness,third moment,con-formity and entropy resulted in the character vector.The back -propagation(BP)neural network was used to classify tumors into be-nign and malignant.BP neural network yielded the following results:88.4% and 78.6% respectively.The proposed system based on texture features performs well in the ultrasonic classification of breast tumors as benign or malignant.%探讨纹理特征在超声乳腺肿瘤诊断中的价值。提取超声图像中乳腺肿瘤的纹理度量,得到由均值、标准差、平滑度、三阶矩、一致性和熵组成的特征矢量,最后用反向传播人工神经网络(BP)对96幅乳腺肿瘤的良恶性进行分类识别。BP 神经网络对良、恶性肿瘤的正确识别率分别为88.4%和78.6%。基于乳腺肿瘤超声图像的纹理特征建立的神经网络系统对肿瘤的良恶性具有较好的识别能力。
展开▼