. 2021 May 15;23:100271. doi: 10.1016/j.pacs.2021.100271

Table 1.

Comparison of reconstruction performance for synthetic vasculature in different models trained with either textural images or both the textural images and measured time-series signals.

Dataset	Synthetic Vasculature
Network	Parameters	PSNR	SSIM	SSIM Luminance	SSIM Contrast	SSIM Structure	Exec.
FD-UNet (TR)	37,888,449	20.802 ± 1.835	0.584 ± 0.053	0.741 ± 0.074	0.907 ± 0.021	0.832 ± 0.047	42 ms
FD-UNet (BP)	37,888,449	21.923 ± 1.959	0.637 ± 0.052	0.779 ± 0.069	0.928 ± 0.017	0.852 ± 0.043	42 ms
Y-Net (TR & Raw)	37,506,354	20.778 ± 2.233	0.713 ± 0.064	0.850 ± 0.034	0.913 ± 0.017	0.851 ± 0.049	48 ms
Y-Net (BP & Raw)	37,506,354	21.856 ± 2.159	0.736 ± 0.065	0.868 ± 0.031	0.920 ± 0.020	0.863 ± 0.045	48 ms
FD-YNet (TR & Raw)	34,852,306	21.064 ± 2.150	0.722 ± 0.060	0.853 ± 0.031	0.910 ± 0.019	0.855 ± 0.047	70 ms
FD-YNet (BP & Raw)	34,852,306	22.204 ± 2.209	0.739 ± 0.061	0.868 ± 0.034	0.927 ± 0.018	0.865 ± 0.043	70 ms

The abbreviations in parentheses represent the data form used by the models. TR: time-reversal image. BP: backprojection image. Raw: measured time-series signals. PI: pixel-interpolated data.