使用非对称数据进行MR到CT的图像生成 Deep MR to C

作者采用cycle-GAN的思想进行MR和CT图像的模态转换。利用循环一致性实现在非对称数据下图像的模态转换。

损失

方法

backward cycle

评价指标

1）mean absolute error

2）peak-signal-to-noise-ratio (PSNR)

实验结果

Fig. 4: From left to right Input MR image, synthesized CT image, reference real CT image, and absolute error between real and synthesized CT image.
Fig. 5: From left to right Input MR image, synthesized CT image with paired training, synthesized CT image with unpaired training, reference real CT image.
Fig. 6: From left to right Input MR image, synthesized CT image, reconstructed MR image, and relative error between the input and reconstructed MR image