基于神经网络的手骨特征区域自适应提取方法  

An adaptive extraction method of carpal feature region based on neural network

作　　者：丁维龙[1] 丁潇 池凯凯[1] 陈琦[1] 毛科技[1,2] DING Weilong;DING Xiao;CHI Kaikai;CHEN Qi;MAO Keji(College of Computer Science&Technology,Zhejiang University of Technology,Hangzhou 310023,China;Zhejiang Kangtihui Technology Company,Hangzhou 311215,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,浙江杭州310023 [2]浙江康体汇科技有限公司,浙江杭州311215

出　　处：《浙江工业大学学报》2020年第5期562-569,共8页Journal of Zhejiang University of Technology

基　　金：国家自然科学基金资助项目(61702456,61571400);浙江省自然科学基金资助项目(LY18C130012);浙江省重点研发项目(2018C01082);浙江省公益技术研究计划/工业项目(LGG20F020018)。

摘　　要：准确分析手腕骨特征骨块的成熟等级是骨龄判别的关键。在骨龄的自动评估中,通过多特征区域识别最终得到骨龄是目前研究的主要方法。钩骨和桡骨在中国人手腕骨发育标准CHN法骨龄评价中所占的权重较大。针对手腕骨特征骨块周边干扰骨块多、识别困难的问题,提出了一种基于BP神经网络的手腕骨特征区域自适应提取方法,可以根据手部X光片所属者的身高和年龄信息,自适应提取出特征骨块,最后搭建卷积神经网络对钩骨、桡骨的成熟等级进行评价。实验结果表明:利用基于BP神经网络的区域自适应提取方法,能够精确提取手腕骨特征骨块区域,在保留完整特征信息的基础上减少周边骨块、肌肉组织的干扰,提高了图片识别的质量;进而提高了卷积神经网络对钩骨、桡骨成熟等级的识别准确率(分别达到了87.83%和85.51%)。所提方法对骨龄的自动识别有重要意义,对临床医生评价骨龄也有较大的参考意义。Accurate analysis of the maturity grade of the characteristic bone fragments of the wrist is the key to discriminate the bone age.In the automatic assessment of bone age,it is the main method to get the final bone age through multi-feature region recognition.The uncinate bone and radius play an important role in the evaluation of bone age of Chinese hand and wrist bone development standard CHN method.Aiming at the problem that there are many disturbing bones around the characteristic bone fragments and they are difficult to be recognized,this paper presents an adaptive extraction method of wrist feature area based on BP neural network.According to the height and age information of the owner of the hand X-ray film,the feature bones can be extracted adaptively.Finally,a convolution neural network is built to evaluate the maturity grade of uncinate bone and radius.The experimental results show that the region adaptive extraction method based on BP neural network can accurately extract the characteristic bone regions of wrist bone,reduce the interference of peripheral bone and muscle tissue on the basis of retaining the complete feature information,and improve the quality of image recognition;furthermore,the recognition accuracy for the maturity grade of uncinate bone and radius based on convolution neural network is improved,reaching 87.83%and 85.51%respectively.The method proposed in this paper is of great significance to the automatic recognition of bone age.It also has great reference significance for clinicians to evaluate bone age.

关 键 词：BP神经网络 区域自适应提取 CHN法 骨成熟等级 医学图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]