基于特性模型与神经网络的乳腺图像肿块自动检测技术

Automatic Detection of the Masses in the Mammograms Using Characteristic Modeling and Neural Networks

钼靶X线摄影是最常用的乳腺癌早期诊断手段。该文针对乳腺图像中的肿块提出了一种基于特性模型与神经网络的计算机辅助诊断技术。它首先建立两种特性模型分别描述脂肪组织和腺体组织中的肿块;然后对脂肪中的肿块采用迭代阈值法进行检测,对腺体中的肿块采用小波域黑洞检索法进行标记;接着采用一种基于Canny算子和能量场约束以及ANFIS控制的填充膨胀方法分割疑似肿块;最后使用一种MLP分类器剔除假阳性。实验结果表明,该算法在面对特性迥异的多种肿块时可取得较高的检测精度,并保证较低的假阳性率。

Mammography is a conventional early detection method for breast cancer. A novel Computer-Aided Diagnosis （CAD） method for the masses is proposed in this paper. Two characteristic models are built up to represent the masses with various backgrounds, and iterative thresholding is carried out to detect the masses in the fatty tissue; however, black-hole detection of wavelet-domain is applied to label the masses in the dense tissue. Filling dilation based on ANFIS controller, Canny detector and the energy field constraint is used to segment the suspicious masses, and MLP-based classifier is applied to suppress the false positives. The experiments validate that the proposed algorithm gets high detection precision, as well as low false positive rate.