基于支持向量机的超声乳腺肿瘤图像计算机辅助诊断系统  

Support vector machines based computer-aided diagnosis system of breast tumor with ultrasound images

作　　者：李晓峰[1] 沈毅[1] 

机构地区：[1]哈尔滨工业大学控制科学与工程系检测技术与自动化装置实验室,哈尔滨150001

出　　处：《光电子．激光》2008年第1期115-119,共5页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金资助项目(60573071)

摘　　要：支持向量机与其他16种常见分类器相比具有优异的分类性能。本文提出一个基于支持向量机的超声乳腺肿瘤图像计算机辅助诊断系统。该系统提取了乳腺肿瘤图像归一化自相关系数、均方根系数、紧凑度等共26个纹理和几何特征。这些特征不但反映乳腺良性肿瘤和恶性肿瘤本质的区别,还考虑到医学超声图像为相干图像的特点。实验结果表明,该系统诊断精确度为93.03%,敏感性为94.30%,特殊性为91.59%,错误正比率92.80%,错误负比率为93.33%,接受特性曲线面积AUC为0.9669,性能优于较有影响的文献[1]提出的方案。A computer-aided diagnosis system of breast tumor with ultrasound images based on is proposed. Based on the criterions used by physicians, total twenty-six features about textures and geometric feature including autocorrelation coefficients, mean square deviation coefficients, and compact ratios are extracted. These features reflect the essential differences between malignant breast tumors and benign tumors,and concern about the characters of ultrasound images as the coherent images. Experiment results show that the classification accuracy is 93.03%, sensitivity is 94.30%, specificity is 91.59%, the positive predictive value is 92.80% ,negative predictive value is 93.33% ,the areas under the ROC curve is 0. 9669.

关 键 词：支持向量机 医学超声图像 计算机辅助诊断 小波变换 特征提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]