图像特征参量分析方法及其在胎盘分级中的应用

目的 针对孕妇妊娠期胎盘功能分级这一具体应用，建立自动诊断系统，并探讨图像特征提取及选择、模式识别的方法。方法 应用灰度统计量分析、二维滤波器和小波分析理论等图像处理理论，从胎盘B型超声图像中提取出用以胎盘功能分级的多个特征参数，然后利用前向搜索法进行参数的有效性分析，并分别通过线性回归法和数量化理论，建立特征参量与胎盘分级之间的联系。结果 前向搜索法可以较好地分析各特征参量的应用价值，而且训练集的模式识别效果好。结论 在样本集较小的情况下，采用线性回归进行模式识别得到的结果有较大的临床价值。

Image Analysis in Terms of Characteristic Parameters and Its Application in Placenta Grading.

Objective To develop a Computer Aided Diagnosis system to grade the placenta automatically using methods of image feature extraction and pattern recognition. Method Features from B mode ultrasound images were extracted by applying image processing methods, such as gray value statistics, 2D filter and wavelet analysis. Then the validity of the features was analyzed by sequential forward selection (SFS) algorithm. The relationship between the features and the placental grade was obtained using linear regression and quantitative method. Result The SFS algorithm was suitable for analyzing the validity of features and the results of training set were perfect. Conclusion Automatic placenta grading using linear regression was valuable in clinic when the number of placenta cases was limited.