将超分辨率重建技术引入医学领域,提升医学图像分辨率辅助医生分析病情,针对目前深度学习方法特征提取感受野低,训练难度过大的问题,提出了一种轻量级的医学图像超分辨率方法.首先多尺度特征提取模块,使用多个小卷积核级联方式扩大感受野;其次通道搅乱注意力模块,通过通道搅乱方式增强各个信息通道之间的联系,通过空间注意力模块进行特征聚合;最后使用亚像素卷积层进行上采样操作得到最终输出的高分辨率图像.经实验验证,所提出方法与RDN(Residual Dense Network for Image Super-Resolution)相比,参数量显著下降,同其他轻量级算法相比,图像的重建质量更好.
CT image super-resolution reconstruction based on multi-scale channel scrambling network
In order to introduce super-resolution reconstruction technology into the medical field and improve the resolution of medical images to assist doctors in analyzing the disease,a light-weight medical image super resolution method is proposed.Firstly,a multi-scale feature extraction module,which uses multiple small convolution kernel cascading methods to expand the receptive field;Secondly,a channel disturbance attention module,which enhances the connection between various information channels through channel disturbance;Finally,the sub-pixel convolutional layer used for upsampling operation to obtain the fi-nal output high-resolution image.It is verified by experiments that compared with RDN(Residual Dense Network for Image Su-per-Resolution),the proposed method has significantly reduced parameters,and compared with other lightweight algorithms,the quality of image reconstruction is better.
medical imagesuper-resolution reconstructionattention mechanismmulit-scale feature
万方数据
维普
