Breast cancer remains a leading cause of cancer deaths among women in many parts of the world. It has been reported that early detection of breast cancer in asymptomatic women can reduce breast cancer mortality. Mammography is considered to be the most effective technique for early detection of breast cancer. Mass automatic detection is the first step of computer-aided diagnosis. An algorithm was developed to detect masses in digital mammograms. Mass detection is very difficult because of weak contrast to background and varieties of masses in size, location, and intensity. Preprocessing method, feature extraction and classifier design are key problems. In this paper, we present a binary decision tree based method. After mammograms are enhanced, an adaptive threshold algorithm is applied to segment suspicious regions. Six features , such as area, compactness, circularity, gray variance, gray mean value and deviation, are extracted to represent suspicious region. Binary decision tree is used to classify the suspicious regions to either mass or normal tissue. A set of 50 mammograms were used to verify the performance of this scheme. Results were achieved with a sensitivity of 86% at the 1. 18 false positive (FPs)/image. Experimental results demonstrate the effectiveness of the proposed method.%乳腺癌是严重威胁女性健康的重要疾病,乳腺癌计算机辅助诊断能够提高乳腺普查的效率和精度.乳腺肿块的自动检测是实现乳腺癌计算机辅助诊断的重要一步.由于肿块和背景之间的对比度低,肿块大小、位置、灰度不确定等,肿块的准确检测非常困难.预处理、疑似区域分割、特征提取以及分类器设计是乳腺肿块分割的关键.本文对经过增强的乳腺X光图像采用一种自适应阈值方法分割出疑似区域,提取疑似区域表征乳腺肿块 的面积、紧凑度、圆形度、灰度方差、灰度均值以及偏离度六种特征,最后利用二叉决策树把疑似区域分为两类:肿块和正常乳腺组织.利用50幅图像测试系统的性能,肿块的检测率(TP)为86.18％,且每幅图像的平均误检(FP)为1.18个.实验结果证明了本文提出方法的有效性.
展开▼
