生成对抗网络在医学图像转换领域的应用

常晓,蔡昕,杨光,聂生东

Applications of Generative Adversarial Networks in Medical Image Translation

Xiao CHANG,Xin CAI,Guang YANG,Sheng-dong NIE

表1 基于GAN的医学图像转换研究

Table 1 Summary of medical image translation research based on GAN

应用场景 文献 图像类型 网络架构 损失函数 评价标准 含噪图像 ↓ 去噪图像 [20] CT WGAN LWGAN + Limage +Lperceptual M8, 9, 14 [21] CT Pix2Pix+ LGAN + Lperceptual M5, 6, 8, 9 [22] CT LSGAN, PatchGAN, LAPGAN LGAN + Limage+Lperceptual M7, 8, 9 [26] MRI WGAN LWGAN + Limage+Lperceptual M8, 9 低分辨图像 ↓ 高分辨图像 [27] MRI Pix2Pix+ LGAN + Limage M7, 8, 9 [28] MRI DCGAN LGAN + L1 M7, 8, 9, 10, 11 [30] PET CGAN, U-Net LGAN + L1 M1, 7, 8 模态转换 [33] T1→FLAIR CGAN LGAN + Limage M7, 8, 17 [34] T1→T2，T1→FLAIR CGAN LGAN + Ledge M7, 8, 9 [35] MRI→CT DCGAN LGAN + Limage + Lgradient M7, 8 [37] MRI→CT Pix2Pix+ LGAN M7, 8 [39] MRI→PET CycleGAN LGAN + Limage + Lcycle M15 [40] MRI→PET CGAN LGAN M1, 2, 3 [41] X-ray→CT DCGAN, WGAN LWGAN M1 小样本 ↓ 大样本 [42] MRI CGAN+PGGAN LWGAN-GP M12 [43] MRI PGGAN LWGAN-GP M12, 13, 16 [44] MRI PGGAN LGAN+Lcycle M8, 9 [45] MRI PGGAN LGAN + LSSIM + L1 M4, 16 [46] MRI CGAN LGAN + L1 + Lseg M17 [47] 病理图像 GAN LGAN M15 [50] X-ray PGGAN LGAN + Limage+Lfrequency M15 [51] ECG Pix2PixHD LGAN+Limage+Lperceptual -