双侧特征融合的乳腺肿块检测

Breast Tumor Detection on Fusion of Bilateral Feature

乳腺癌是妇女最常见的恶性肿瘤之一,面向乳腺钼靶X线图像的计算机辅助肿块检测技术可以帮助影像科医师早期发现乳腺病变.针对于单侧的乳腺肿块检测中准确率有待提升的问题,提出双侧特征融合的乳腺肿块检测算法.首先,进行图像预处理,并利用相干点漂移完成乳腺轮廓配准;然后,利用配准得到的变换矩阵获得双侧乳腺感兴趣区域,再在其中提取左右侧乳腺的单侧特征向量和双侧对比特征向量,从而建立融合的特征模型,并采用遗传选择算法对特征向量进行特征选择;最后利用极限学习机基于选择后的特征进行乳腺肿块检测.实验结果表明,与传统的基于单侧的乳腺肿块检测算法相比,文中算法能有效地提高检测准确率.

Breast cancer is one of the most common malignant tumors of women, computer-aided breast tumor detection technology faced on mammogram can help radiologists to detect breast lesions early. Breast tumor detection （BTD） algorithm on fusion of bilateral feature is presented, which aims at the problem that the accuracy of BTD on unilateral feature can be improved. First, preprocess the image and registration the breast contour using the coherent point drift. Then, use transformation matrix obtained from registration to acquire regions of interest （ROIs） of bilateral breast. After that, extract unilateral feature vector of left and right breast and bilateral contrast feature vector in ROIs. Thereby, the fusion feature model is set up, and then the features are selected by genetic algorithm selection. Finally, breast tumor is detected by extreme learning machine based on the selected features. Experimental results show that the proposed BTD algorithm on fusion of bilateral features can improve the accuracy of the detection effectively, compared to the traditional BTD algorithm on unilateral feature.