基于卷积神经网络的胎儿二维超声图像丘脑横切面的自动识别  

Automatic recognition of thalamic section in fetal two-dimensional ultrasound images based on convolutional neural network

作　　者：丁春霞 屈若为[2] 殷颖 DING Chun-xia;QU Ruo-wei;YIN Ying(Ultrasound Department,Zhangjiakou Maternal and Child Health Hospital,Zhangjiakou Hebei 075000,China;State Key Laboratory for Reliability and Intelligence of Electrical Equipment Co-constructed by the Provincial Department of Electrical Engineering College of Hebei University of Technology,Tianjin 300130,China)

机构地区：[1]张家口市妇幼保健院超声科,张家口河北075000 [2]河北工业大学电气工程学院省部共建电工装备可靠性与智能化国家重点实验室,天津300130

出　　处：《中国临床医学影像杂志》2020年第5期358-362,共5页Journal of China Clinic Medical Imaging

基　　金：国家自然科学基金(51377045)。

摘　　要：目的:胎儿二维超声图像不同切面的自动识别分类对于提高医生的工作效率具有十分重要的意义。方法:本文针对传统自动分类方法中需要先对图像进行细致分割再进行特征提取和分类识别、分类速度慢等问题,提出了一种基于深度卷积神经网络(Convolutional neural network,CNN)的胎儿二维超声图像中丘脑横切面的自动识别方法。通过对收集到的胎儿二维超声丘脑横切面进行图像增强等预处理,提出了改进的卷积神经网络算法。结果:该算法避免了对于二维超声图像复杂的前期预处理,可以直接输入原始的二维超声图像,具有很强的适应性和泛化能力。试验结果表明,该方法的识别准确率能达到94.81%。结论:此模型的提出,为医学影像自动识别技术提供了新的参考。Objective: Automatic recognition and classification on fetal ultrasound image is significant to make doctors’ work efficiency. Method: Being different from traditional automatic classification method whose images should be segmented in detail, feature extracted manually and then classified, we proposed a deep convolutional neural network(CNN) based fetal thalamus plane ultrasound image recognition method. First, the images were pre-processed, such as image enhancement;then, we proposed an improved CNN algorithm. Result: This algorithm avoids the complex pre-processing of two-dimensional ultrasound image, and can input the original two-dimensional ultrasound image directly. It has strong capacity of adaption and generalization. The experimental results show that the recognition accuracy of this method can reach 94.81%. Conclusion: The proposed model provides a new reference for automatic medical image recognition technology.

关 键 词：脑 超声检查 产前 

分 类 号：R714.53[医药卫生—妇产科学] R445.1[医药卫生—临床医学]