摘要
为改善低剂量CT图像的质量,提出一种基于改进型Wasserstein生成对抗网络(WGAN-gp)的低剂量CT图像去噪方法。WGAN-gp在WGAN网络的基础上加入梯度惩罚项,解决了WGAN训练困难,收敛速度慢的问题,进一步提高网络的性能。同时加入新感知损失度量函数,使对人眼而言敏感的纹理信息得到保留。实验结果表明,与目前相关的方法相比,PSNR提高了1.83 dB,SSIM表达能力增加了约3.5%,降低了时间复杂度,显著改善了低剂量CT图像的可视质量。
For improving the low-dose CT image quality,the alow-dose CT image denoising method based on WGAN-gp.is put forward.A gradient penalty used by the WGAN network is introduced.The WGAN training costs is reduced and the convergence rate is improved.Meanwhile,a novel perceptual loss function is exploited for enhancing the low-dose images sensitivity to human eyes.Experimental results demonstrate that PSNR is increased by about 1.5 dB and SSIM rises 1.5 on average,moreover show that comparison with conventional methods,the time complexity is reduced,and the visual quality of low-dose CT images is significantly improved.
作者
徐曾春
叶超
杜振龙
李晓丽
XU Zeng-chun;YE Chao;DU Zhen-long;LI Xiao-li(School of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,China)
出处
《光学与光电技术》
2019年第3期101-107,共7页
Optics & Optoelectronic Technology
