摘要
本研究以灰度共生矩阵描述乳腺钼靶X线影像中结构扭曲的纹理特征。对学习样本(乳腺结构扭曲样本44个,正常样本78个),计算五个反映纹理性质的特征参数,根据相应的Fisher系数,确定最适合作为分类依据的特征参数或特征参数组合。用线性判别分析对测试样本(乳腺结构扭曲样本43个,正常样本78个)进行分类。分类结果表明本研究确定的纹理特征熵(ENT)是识别乳腺结构扭曲的最佳统计参数(分类正确率达78.5%、ROC曲线下的面积为0.786)。
The gray-level co-occurrence matrix was employed to describe the texture features of architectural distortion in mammograms. Based on a training dataset composed of 44 architectural distortions and 78 normal mammograms, 5 parameters were calculated and analyzed to represent the texture features. The best features for classification were determined according to the corresponding Fisher indices of these parameters. A linear model was designed to classify a test dataset composed of 43 architectural distortions and 78 normal mammograms with the features determined above. The test results showed that the feature entropy （ENT） was selected as the best parameter to detect breast architectural distortion with an accuracy of 78.5 %, and the area under the ROC curve was 0.786.
出处
《中国生物医学工程学报》
CAS
CSCD
北大核心
2007年第4期503-507,共5页
Chinese Journal of Biomedical Engineering
基金
上海市教委科研基金(04BB11)
上海市教委E-网格研究院项目基金(200304)。
