MaZda评估动脉粥样硬化斑块的风险性

Evaluating the risk grade of atherosclerotic plaque using MaZda

目的采用MaZda软件对粥样硬化斑块B型超声图像进行纹理分析,实现低回声斑块与混合回声斑块的分类。方法收集218个斑块样本,采用交叉验证法。首先利用MaZda从直方图、绝对梯度、游程矩阵、共生矩阵、自回归模型、小波变换中提取斑块的纹理特征值。然后根据费希尔参数法(Fisher)、最小分类误差与最小平均相关系数法(POE+ACC)、相关信息测度法(MI)分别选择10个最优特征值,并融合3组特征获得交集特征组和并集特征组。最后分别根据线性判别分析法(LDA)和非线性判别分析法(NDA)分析纹理特征值,并利用K邻近分类器和神经网络进分类。结果利用LDA分析交集特征组和并集特征组时,分类获得准确率分别为87.5%和89.0%。结论利用MaZda对低回声和混合回声斑块图像分类具有较高的准确率,可作为评估动脉粥样硬化斑块风险性的新工具。

Objective To identify the soft and intermediate plaques atherosclerotic plaque using MaZda software that can extract texture features from ultrasound images.Methods A total of 218 carotid plaque ultrasound images were collected and cross validation was performed.Texture features were extracted from histogram,absolute gradient,run-length matrix,gray-level co-occurrence matrix,autoregressive model and wavelet transform using MaZda.Ten optimal texture features were chosen based on three different methods：Fisher coefficient（Fisher）,minimization of both classification error probability and average correlation coefficients（POE＋ACC）and mutual information measure（MI）,respectively.Two features sets were obtained from the intersection and union of the features chosen by the foregoing three methods.Each texture features set was analyzed by linear（LDA）and nonlinear discriminant analysis（NDA）.The K-nearest-neighbor classification（K-NN）and artificial neural network（ANN）were used to classify the plaques.Results When the features intersection and union were analyzed by LDA,the classification accuracy were 87.5% and 89.0%,respectively. Conclusion It is feasible to identify the high-risk plaques based on texture features extracted from ultrasound images using MaZda.MaZda may be a useful tool for early detection and diagnosis of atherosclerosis.