乳腺肿瘤超声图像中感兴趣区域的自动检测

Automatic Detection of the Region of Interest from Breast Tumor Ultrasound Images

由于斑点噪声、伪影以及病灶形状多变的影响,乳腺肿瘤超声图像中肿瘤区域的自动检测以及病灶的边缘提取比较困难,已有的方法主要是由医生先手工提取感兴趣区域(ROI)。本研究提出一种乳腺肿瘤超声图像中感兴趣区域自动检测的方法,选用超声图像的局部纹理、局部灰度共生矩阵以及位置信息作为特征,采用自组织映射神经网络进行分类,自动识别乳腺肿瘤区域。对包含168幅乳腺肿瘤超声图像的数据库进行识别的结果表明:该方法自动识别ROI的准确率达到86.9%,可辅助医生提取肿瘤的实际边缘以及进一步诊断。

Due to the speckle noise,shadowing artifacts and the variance in shape of sonographic lesions,it is difficult to detect breast tumors and extract lesion boundaries from ultrasound images automatically.Previous breast tumor detection methods have been based on the manual extraction of the region of interest（ROI）.In this paper,a computer-aided automatic method was proposed to detect the ROI of breast tumors from ultrasound images.A self-organizing map（SOM） neural network was used for the classification of the area of breast tumor.The ROI could be extracted automatically employing the local texture,the local gray level co-occurrence matrix,and positions as features.Experimental results showed that the method could recognize breast tumors in our 168-image database with an accuracy of 86.9%,which may assist physicians to extract the boundary and make further diagnose.