对比双侧视图信息的致密型乳腺X线图像肿块检测

Bilateral Analysis of Mass Detection for Dense Mammograms

针对现有双视图肿块检测方法存在的问题,提出一种适用于致密型乳腺X线图像的直接对比双侧视图信息的计算机辅助肿块检测方法.为提高双侧图像对称区域的匹配精度,分割图像中的胸肌区域及腋窝区域,建立仅包含乳房区域的生理坐标系;综合乳腺生理特征及肿块病理性质提取感兴趣区域,以梯度图像的局部三元模式特征距离作为尺度测量对称像素的相似度,有效地降低了肿块检测假阳性率.采用北京大学人民医院乳腺中心提供的临床图像进行算法性能实验,结果表明,生理坐标系在定位与匹配对称区域方面具有良好的性能;与现有双视图肿块检测方法相比,在相同的肿块检测正确率下,文中方法获得更低的检测假阳性率.

A mass detection method based on bilateral information was proposed for dense mammograms.To improve the matching accuracy of symmetric regions,breast regions were extracted after removal of pectoral muscle and axilla part,and an anatomic coordinate system that only consisted of breast region was constructed.To reduce the false positive rate of mass detection,regions of interest(ROIs)were determined using both physiological and pathological characters besides the inherent information of the images.Feature distance of the local ternary pattern of breast gradient amplitude was employed to measure the similarity between matching pixels.The proposed method is tested on the clinical images provided by the breast center of Peking university people’s hospital.Experimental results show that the constructed coordinate system obtains better ROI matching performance.Compared with existing bilateral mass detection methods,the proposed method yields a lower false positive rate with the same mass detection accuracy.