基于特征融合和判别式学习的胎盘成熟度自动分级

Placental Maturity Evaluation via Feature Fusion and Discriminative Learning

胎盘成熟度分级错误可能会导致小于胎龄儿、死产、死胎等的发生。目前,胎盘成熟度分级主要依赖于临床医生的经验和观察,主观性很强,分级准确性易受医生工作强度、工作时长和工作经验的影响。提出一种基于特征融合和判别式学习的胎盘成熟度自动分级算法。首先,对共544例的B型超声图像和彩色能量多普勒(CDE)胎盘图像,采用提取关键点、对关键点提取特征、进行融合并加以判别式特征编码的方法,形成码书,经过归一化,最后用支持向量机(SVM)进行分类,得到胎盘成熟度分级结果。在测试阶段,将胎盘成熟度测试结果与临床医生的分级结果进行对比,得到如下结果:准确率92.7%,敏感性91.1%,特异性97.6%,平均精度97.3%。结果表明,该方法对胎盘成熟度自动分级具有较高的指导意义。

The error of placental maturity classification may lead to the occurrence of small gestational age（ SGA）,stillbirth,dead fetus,etc. Currently,placental maturity evaluation mainly depends on the clinician＇s experience and observation. In this paper,we proposed a novel method to evaluate the placental maturity automatically by feature fusion and discriminative learning. Specifically,we extracted both the gray-scale intensity and blood flow information by the visual feature detector and descriptor from a total of 544 B-mode gray-scale ultrasound（ US） images and color doppler energy（ CDE） images. After fusing information,we applied the feature encoding method to improve the staging performance using discriminative learning technique.Comparing the test results with the result of the clinicians,we obtained a result with the accuracy of 92. 7%,the sensitivity of 91. 1%,a specificity of 97. 6%,and a mean average precision of 97. 3%. The experimental results showed that the proposed method achieved promising performance for placental maturity automatic classification.