引用本文
  • 刘鸣谦,兰钧,陈旭,于广军,杨秀军.基于多维度特征融合的深度学习骨龄评估模型[J].第二军医大学学报,2018,39(8):909-916    [点击复制]
  • LIU Ming-qian,LAN Jun,CHEN Xu,YU Guang-jun,YANG Xiu-jun.Bone age assessment model based on multi-dimensional feature fusion using deep learning[J].Acad J Sec Mil Med Univ,2018,39(8):909-916   [点击复制]
【打印本页】 【下载PDF全文】 【HTML】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 798次   下载 584 本文二维码信息
码上扫一扫!
基于多维度特征融合的深度学习骨龄评估模型
刘鸣谦1,兰钧1,陈旭1,于广军2*,杨秀军2
0
(1. 卫宁健康科技集团股份有限公司人工智能实验室, 上海 200072;
2. 上海交通大学附属儿童医院儿童精准医学大数据工程技术研究中心, 上海 200040
*通信作者)
摘要:
目的 通过基于特征提取的深度卷积神经网络,结合关键区域特征和人口学信息,评估儿童骨龄。方法 自动识别左手X线图像数据,对图像进行预处理,使用基于深度神经网络的X线图像分析方法,实现左手关节骨龄17个关键区域特征的自动提取,再将骨龄影像特征与临床大数据(人口统计、性别)融合训练骨龄评估模型,测试模型的评估效能。结果 使用基于深度学习的特征区域提取方法比传统图像分析方法可以更好地提取特征信息,结合临床信息从另一维度补充了骨龄发育信息。基于多维度数据特征融合的骨龄评估模型检测得到的骨龄平均绝对误差为0.455,优于传统方法和仅端到端的深度学习方法。结论 相较传统的机器学习特征提取方法,基于特征提取的深度卷积神经网络在骨龄回归模型上有更好的表现,结合人口和性别信息可进一步提升基于图像的骨龄预测准确率。
关键词:  人工智能  深度学习  医学影像  大数据分析  骨骼年龄测定
DOI:10.16781/j.0258-879x.2018.08.0909
投稿时间:2018-08-06修订日期:2018-08-16
基金项目:上海交通大学医工交叉重点项目(项目编号YG2017ZD08)
Bone age assessment model based on multi-dimensional feature fusion using deep learning
LIU Ming-qian1,LAN Jun1,CHEN Xu1,YU Guang-jun2*,YANG Xiu-jun2
(1. Winning Artificial Intelligence Research, Winning Health Technology Group Co. Ltd., Shanghai 200072, China;
2. Big Data Eengineering and Technology Research Center for Pediatric Precision Medicine, Children's Hospital of Shanghai Jiao Tong University, Shanghai 200040, China
*Corresponding author)
Abstract:
Objective To evaluate the bone age of children using deep convolutional neural network based on feature extraction combined with key features and demographic information.Methods Left hand X-ray images were automatically recognized and preprocessed, and then the 17 key region features of bone age in the left hand joint were automatically extracted by X-ray image analysis method based on deep convolutional neural network. The image features of bone age were combined with clinical data (population statistics and gender) to train and test the bone age assessment model.Results The feature region extraction method based on deep learning had better efficiency in extracting feature information than traditional image analysis method, and the feature information combined with clinical information supplemented the information of bone age from another dimension. The average absolute error measured by bone age assessment model based on multi-dimensional data feature fusion was 0.455, which was better than traditional methods and only end-to-end deep learning method.Conclusion Compared with traditional machine learning methods, the deep convolutional neural network based on feature extraction has better performance, and can improve the predicting accuracy of image-based bone age by combining with population information such as gender and age.
Key words:  artificial intelligence  deep learning  medical imaging  data analysis  age determination by skeleton