摘要
针对医学电子计算机断层扫描(CT)图像方向校正问题,提出一种并行卷积回归(PCRN)多任务深度学习网络.通过侧旋角度正回归和翻转概率逻辑回归,求得校正参数来精准地校正图像.进一步,针对医学图像训练样本稀缺的情况,提出一种串行回归(SCRN)的深度学习架构,弥补并行卷积回归网络在小样本情况下校正精度不足的问题.实验结果表明:在样本充分,并行卷积回归网络和样本稀缺情况下,串行卷积回归网络对小角度偏转、大角度偏转和翻转的腹部CT图像校正结果都优于传统的配准方法.
To address the problem of orientation correction of CT image,a parallel convolution regression network(PCRN)with multitask deep learning is proposed.Orientation parameters are learned by the positive regression of the lateral rotation angle and the reversal probability logistic regression to accurate calibration images.Furthermore,in view of the lack in training samples for medical images,a deep learning frame named serial convolution regression network(SCRN)is introduced,which makes up for the inadequate correction accuracy of parallel convolutional regression network in the case of small samples.The experimental results show that the PCRN method with sufficient samples and the SCRN method with scarce samples are superior to the traditional registration methods in correcting CT images with small,large angles and flipped situation.
作者
林家庆
韩娟
袁直敏
彭佳林
LIN Jiaqing;HAN Juan;YUAN Zhimin;PENG Jialin(College of Computer Science and Technology,Huaqiao University,Xiamen 361021,China)
出处
《华侨大学学报（自然科学版）》
CAS
北大核心
2020年第3期366-373,共8页
Journal of Huaqiao University(Natural Science)
基金
国家自然科学基金资助项目(11771160)。
关键词
CT图像
方向校正
深度卷积网络
多任务回归网络
CT image
orientation correction
deep convolution network
multitask regression network
