Skip to main content
. 2023 Jan 13;29:100452. doi: 10.1016/j.pacs.2023.100452

Table 2.

Evaluation of model performance on the mouse brain vasculature in ssim metrics at different noise levels.

MBLr FIRe 2S-FIRe TV U-Net
6 dB SNR 0.736 ± 0.048 0.788 ± 0.038 0.797 ± 0.047 0.515 ± 0.031 0.432 ± 0.049
9 dB SNR 0.831 ± 0.044 0.847 ± 0.033 0.853 ± 0.039 0.583 ± 0.032 0.559 ± 0.049
12 dB SNR 0.887 ± 0.037 0.886 ± 0.030 0.886 ± 0.033 0.635 ± 0.034 0.636 ± 0.052
15 dB SNR 0.907 ± 0.033 0.901 ± 0.028 0.915 ± 0.026 0.660 ± 0.034 0.699 ± 0.051
18 dB SNR 0.926 ± 0.027 0.918 ± 0.026 0.912 ± 0.027 0.698 ± 0.035 0.692 ± 0.052
21 dB SNR 0.931 ± 0.025 0.923 ± 0.025 0.916 ± 0.026 0.712 ± 0.036 0.701 ± 0.051
24 dB SNR 0.934 ± 0.024 0.926 ± 0.024 0.919 ± 0.025 0.720 ± 0.036 0.705 ± 0.051
27 dB SNR 0.935 ± 0.024 0.927 ± 0.024 0.920 ± 0.025 0.724 ± 0.036 0.708 ± 0.051
30 dB SNR 0.936 ± 0.024 0.928 ± 0.024 0.920 ± 0.025 0.726 ± 0.037 0.708 ± 0.051
- 0.936 ± 0.024 0.928 ± 0.024 0.921 ± 0.025 0.729 ± 0.037 0.709 ± 0.051

MBLr: model-based learning. FIRe: fast iterative reconstruction. 2S-FIRe: two-stage fast iterative reconstruction.TV: total variation. U-Net: single-step post-processing. -: no noise included in collected sensor data.