计算机辅助系统在乳腺钙化性病变X线摄影诊断中的应用

Application of Computer-aided System in X-ray Diagnosis of Breast Calcification Lesions

目的探讨计算机辅助检测算法在乳腺X线摄影钙化检测诊断中的应用。资料与方法回顾性分析经乳腺X线摄影检测且经病理学证实的病例样本316例,其中钙化病变样本112例、无钙化病灶样本204例。利用多尺度滤波去噪和自适应局部直方图均衡化预处理增强乳腺X线图像的对比度。通过K-means算法进行乳腺组织的多阈值迭代腺体分割。利用局部二值模式和灰度共生矩阵2个纹理特征进行乳腺钙化性病变真假阳性检测。比较计算机自动诊断结果与2名放射学专业医师诊断结果。结果利用支持向量机、随机森林和自适应增强算法自动分类结果的精确率分别达到90.0%、81.5%和87.5%。从事乳腺X线诊断工作约1年的低年资住院医师诊断准确率为80.0%,从事乳腺X线诊断工作约5年的低年资主治医师诊断准确率为85.0%。结论计算机自动支持向量机分类检测算法优于放射科医师的经验诊断结果。计算机辅助乳腺X线钙化病变检测算法具有较高的准确率,可为乳腺X线摄影放射科医师提供良好的第二观感。

Purpose To discuss the application of computer-aided detection algorithm in X-ray detection diagnosis of breast calcification.Materials and Methods Retrospective analysis was conducted for 316 case samples confirmed by X-ray detection and pathology,including 112 cases of calcified lesions and 204 cases of calcified lesions.The contrast of breast X-ray images was enhanced by multi-scale filter de-noising and adaptive local histogram equalization preprocessing.Multi-threshold iterative gland segmentation of breast tissue was conduct by applying K-means algorithm.The local binary model and the gray level co-occurrence matrix were used to detect the true and false positive of calcified breast lesions.The automatic diagnosis results of computer were compared with the results of 2 radiologists.Results The precision rates of automatic classification results of computer using support vector machine,random forest and adaptive enhancement algorithm reached 90.0%,81.5%and 87.5%,respectively.The diagnostic accuracy rate of low-grade residents who engaged in X-ray breast diagnosis for about 1 year was 80.0%,and that of low-grade attending physicians who engaged in X-ray breast diagnosis for about 5 years was 85.0%.Conclusion The automatic SVM classification detection algorithm of computer is superior to the empirical diagnosis results of radiologists.The computer-aided X-ray breast calcification lesion detection algorithm has a high accuracy rate and provided a good second impression for mammography radiologists.