基于Fisher向量和混合描述符的胎盘成熟度分级  

Placental Maturity Grading via Hybrid Descriptors and Fisher Vector

作　　者：姜峰 倪东[1] 陈思平[1] 姚远[2] 汪天富[1] 雷柏英[1] Jiang Feng;Ni Dong;Chen Siping;Yao Yuan;Wang Tianfu;Lei Baiying(School of Biomedical Engineering,Health Science Center,Shenzhen University,National-Regional Key Technology Engineering Laboratory for Medical Ultrasound,Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging Shenzhen 518060,Guangdong,China;Department of Ultrasound,Affiliated Shenzhen Maternal and Child Healthcare,Hospital of Nanfang Medical University Shenzhen 518060,Guangdong,China)

机构地区：[1]深圳大学生物医学工程学院广东省生物医学信息检测和超声成像重点实验室,广东深圳518060 [2]深圳妇幼保健院超声科南方医科大学附属医院,广东深圳518060

出　　处：《中国生物医学工程学报》2018年第5期521-528,共8页Chinese Journal of Biomedical Engineering

基　　金：国家重点研发计划项目(2016YFC0104703)

摘　　要：胎盘成熟度分级(PMG)对于评估胎儿生长和孕妇健康来说至关重要。目前,PMG主要依赖于临床医生的主观判断,不仅十分耗时,而且由于工作的重复性和冗余性,常会产生误判。传统机器学习中使用的手工特征提取方法,不能很好解决PMG的分级问题,因此提出从B超图像和彩色多普勒能量图像中提取深度混合描述符进行胎盘成熟度自动分级的方法。从深度卷积神经网络中提取卷积特征,并将其与手工特征结合形成混合描述符来提高模型性能。首先,将多个特征层的不同模型进行融合,从图像中获取混合描述符。同时,考虑到深度表达特征,使用迁移学习策略来增强分级性能。然后,用Fisher向量(FV)对提取的描述符进行编码。最后,使用支持向量机(SVM)分类器对胎盘成熟度进行分级。用医生标注好的数据进行测试,在基于19层网络的混合特征模型获得高达94.15%的精确度,比单一使用手工特征模型提升3.01%,比CNN特征模型提升7.35%。实验结果证明,所提方法能够有效应用于胎盘成熟度自动分级。Placental maturity grading（PMG） is very essential to assess fetal growth and maternal health. However, PMG has mostly relied on the clinician＇s subjective judgment, which is time-consuming and subjective. A dditionally it may cause wrong estimation because of redundancy and repeatability of the process. Traditional machine learning-based methods capitalize on handcrafted features, but such features may be essentially insufficient for PMG. In order to tackle it, we proposed an automatic method to stage placental maturity via deep hybrid descriptors extracted from B-mode ultrasound（BUS） and color Doppler energy（CDE） images. Specifically, convolutional descriptors extracted from a deep convolutional neural network（CNN） and hand-crafted features were combined to form hybrid descriptors to boost the performance of the proposed method. Firstly, different models with various feature layers were combined to obtain hybrid descriptors from images. Meanwhile, the transfer learning strategy was utilized to enhance the grading performance based on the deep representation features. Then, extracted descriptors were encoded by Fisher vector（FV）. Finally, we used support vector machine（SVM） as the classifier to grade placental maturity. We used placental data labeled by doctors to test models. The accuracy of the model with hybrid descriptors based on the 19-layer network was 94.15%, which was 3.01% higher than that of the model with hand-crafted features and 7.35% higher than the CNN feature model. The experimental results demonstrated that the proposed method could be applied to the automatic PMG effectively.

关 键 词：胎盘成熟度分级 超声图像 深度卷积网络 混合描述符 Fisher向量 

分 类 号：R318[医药卫生—生物医学工程]