摘要
利用人工智能中的深度学习方法自动检测并评价西南地区青少年左手腕关节X线片的骨龄。在四川大学华西第二医院共收集2426例1-18岁青少年左手腕X线片,利用YOLOv3框架和少部分数据进行标定、训练以检测X线片上传统骨龄评价方法需要的区域,将关键区域截图并调整姿态组成新的图片。再利用caffe框架将扩展后的数据集分成训练集、验证集、测试集,以骨龄为标签对不同性别数据分别进行训练以获得男性和女性骨龄预测的模型,并计算误差在±1岁以内的准确率。选择caffe框架训练出来的最好模型,测试出测试集中±1岁的准确率为男性81.06%,女性85.08%。利用深度学习中简单的神经网络训练少量数据即可得到不错的骨龄评价准确率,表明了深度学习方法在西南地区青少年骨龄评价的可行性以及在数据增加和网络优化之后准确率存在的极大提升空间。
Deep learning method in artificial intelligence is used to automatically detect and evaluate the bone age of X-rays of wrist bones in young children from Southwest China.A total of 2426 cases of left wrist X-rays of 1-18 years old are collected from West China Second Hospital of Sichuan University.The YOLOv3 framework and a small amount of data are used for calibration and training to detect the areas needed by traditional bone age assessment methods on the X-rays,and the key areas are captured and adjusted to form a new picture.Then the extended data set is divided into training set,verification set and test set by the caffe framework.With bone age as a label,the models for predicting bone age of men and women are obtained by training different sex data,and the accuracy of the error within ±1 year is calculated.The best model trained by the caffe framework is selected to test the accuracy of ±1 year old in the test set,which is 81.06%for males and 85.08%for females.A simple neural network in deep learning can be used to train a small amount of data to obtain a ideal accuracy rate of bone age evaluation,which indicates the feasibility of deep learning method in bone age evaluation of adolescents in southwest China and the great room for improvement of accuracy after data increase and network optimization.
作者
李睿
张世杰
黄奥云
陈虎
LI Rui;ZHANG Shi-jie;HUANG Ao-yun;CHEN Hu(School of Computer Science,Sichuan University,Chengdu 610065,China;National Key Laboratory of Fundamental Science on Synthetic Vision,Sichuan University,Chengdu 610065,China;Wisesoft Co.,Ltd.,Chengdu 610045,China)
出处
《计算机技术与发展》
2020年第1期124-128,134,共6页
Computer Technology and Development
基金
国家重点研发计划(2016YFC0801100)
关键词
人工智能
深度学习
骨龄
放射学
artificial intelligence
deep learning
bone age
radiology
