基于改进生成对抗网络的低剂量CT去噪算法

针对低剂量计算机断层扫描(computerized tomography,CT)在图像采集过程中引入较多噪声,造成图像质量严重下降的问题, 提出一种基于残差注意力机制与复合感知损失的低剂量CT去噪算法。在该算法中,利用生 成对抗网络完成对低剂量CT图像的去噪,在网络框架中引入多尺度特征提取及残差注意力 模块,以融合图像中不同尺度的信息,提高网络对噪声特征的区分能力,避免在去噪过程中 丢失图像细节信息。同时采用复合感知损失函数,以加快网络收敛速度,促使去噪图像在感 知上与原图像更接近。实验结果表明:与现有的算法相比,所提算法能够有效抑制低剂量 CT图像中的噪声,并恢复更多的纹理细节；对比低剂量CT图像,所提算法处理后的CT 图像峰值信噪比(peak signal-to-noise ratio,PSNR) 值提高了31.72%, 结构相似性(structural similarity,SSIM)值提高了13.15%,可以满足更高的医学影像诊断要求 。

Low-dose CT denoising algorithm based on improved generative adversarial networ k

In order to solve the problem that low-dose computerized tomography (CT) introduces a lot of noise in the proqcess of image acquisition,which leads to the serious degradation of image q uality,a low-dose CT denoising algorithm based on residual attention mechanism and composite perceptu al loss is proposed in this paper.In this algorithm,the Generative Adversarial Networks i s used to complete the denoising of low-dose CT images.The multi-scale feature extraction and re sidual attention module are introduced into the network framework to fuse the information of diff erent scales in the image,improve the ability of the network to distinguish noise features,and avoid the loss of image details in the process of denoising.At the same time,the composite perce ptual loss function is used to accelerate the convergence speed of the network and promote the denoi sing image to approach the original image perceptually.Experimental results show that the pro posed algorithm can effectively suppress noise and recover more texture details in low-dose CT images compared with existing algorithms.Compared with the low-dose CT images,the peak signal-to-noise ratio (PSNR) value and structural similarity (SSIM) value of the CT images processed by the proposed algorithm are increased by 31.72% and 13.15%,which can meet the higher requirements of medical imaging diagnosi s.