摘要
骨龄预测时,手骨X光片经常存在标尺、伪影、噪声、曝光不当等缺陷.采用常规的滤波加深度学习神经网络等模型进行预测往往正确率不高.提出一种X光片骨龄辅助预测的预处理方法,包括使用专门用于生物医学图像分割的U-Net网络将X光片中手骨分割出来,使用图像二值化对U-Net生成的掩模进行去除背景处理,使用灰度直方图均衡的办法解决图像过亮或过暗的问题.经上述精细预处理后,再进行深度学习神经网络预测,实验结果表明精细的预处理对实验结果有很好的改进作用.
Problems of scale,artifact and noise in the X-ray pictures of hand bone always exist during the bone age prediction.The accuracy of prediction by normal filter and end-to-end deep learning neural network model is not so high normally.This paper uses the method of U-Net,which is used for biomedical image segmentation,to split hand bone X-ray film.The method uses image binarizaion to remove the background of mask generated by U-Net,and uses grayscale straight square equalization to solve the problem of brightness of images.Deep learning neural network prediction after fine pretreatments mentioned above can significantly improve the results of bone age prediction.
作者
苏叶
李婧
徐寅林
Su Ye;Li Jing;Xu Yinlin(School of Computer and Electronic Information,Nanjing Normal University,Nanjing 210023,China)
出处
《南京师范大学学报（工程技术版）》
CAS
2021年第2期54-59,共6页
Journal of Nanjing Normal University(Engineering and Technology Edition)
关键词
骨龄预测
预处理
灰度直方图均衡
U-Net网络
深度学习
bone age prediction
pretreatment
grayscale histogram equalization
U-Net network
deep learning
