Table 2.

Comparison of reconstruction performance for synthetic vasculature in different models trained with either the pixel-interpolated data or the combination of pixel-interpolated data with another form of input.

Dataset	Synthetic Vasculature
Network	Parameters	PSNR	SSIM	SSIM Luminance	SSIM Contrast	SSIM Structure	Exec.
Y-Net (PI & Raw)	37,524,210	23.747 ± 2.320	0.817 ± 0.037	0.912 ± 0.015	0.957 ± 0.009	0.909 ± 0.029	49 ms
FD-YNet (PI & Raw)	34,861,234	24.400 ± 2.167	0.828 ± 0.044	0.920 ± 0.019	0.954 ± 0.013	0.914 ± 0.029	76 ms
Y-Net (PI & TR)	32,809,717	25.090 ± 2.299	0.821 ± 0.030	0.906 ± 0.013	0.958 ± 0.012	0.924 ± 0.027	27 ms
FD-YNet (PI & TR)	27,790,517	23.800 ± 2.297	0.835 ± 0.043	0.924 ± 0.020	0.958 ± 0.011	0.912 ± 0.030	47 ms
Y-Net (PI & BP)	32,809,717	24.392 ± 2.481	0.826 ± 0.042	0.914 ± 0.018	0.963 ± 0.008	0.915 ± 0.030	27 ms
FD-YNet (PI & BP)	27,790,517	24.804 ± 2.295	0.812 ± 0.038	0.906 ± 0.019	0.959 ± 0.011	0.908 ± 0.030	47 ms
UNet (PI)	31,062,145	24.942 ± 2.234	0.831 ± 0.037	0.918 ± 0.014	0.955 ± 0.014	0.925 ± 0.026	24 ms
Pixel-DL (PI)	37,906,305	24.957 ± 2.204	0.815 ± 0.030	0.902 ± 0.011	0.958 ± 0.013	0.922 ± 0.027	42 ms
PixelGAN (PI)	G: 37,906,305 D: 1,711,041	24.538 ± 2.182	0.822 ± 0.040	0.917 ± 0.020	0.958 ± 0.011	0.907 ± 0.029	42 ms
PixelcGAN (PI)	G: 37,906,305 D: 1,743,809	24.571 ± 2.214	0.813 ± 0.044	0.907 ± 0.029	0.957 ± 0.011	0.907 ± 0.030	42 ms
Model-Based Learninga	198,565	29.590 ± 2.694	0.930 ± 0.026	0.971 ± 0.011	0.985 ± 0.006	0.966 ± 0.014	∼ 6 s
TV	-	23.774 ± 2.403	0.721 ± 0.037	0.869 ± 0.025	0.914 ± 0.023	0.863 ± 0.035	345 s

The abbreviations in parentheses represent the data form used by the models. TR: time-reversal image. BP: backprojection image. TV: total variation. Raw: measured time-series signals. PI: pixel-interpolated data. G: generator. D: discriminator. ^aTrained with the data evaluated by repeated forward and adjoint operators and reconstructed outputs from the previous iteration.