Result of code

[wuxiyu64]

Hi, after running this program, I found a discrepancy between the test results and the paper. For DIRLAB data set, TRE mean and std results were not consistent with the results in the literature, and the results were much worse than those in the paper. Especially in 4dct 08,09,10. Have you made any adjustments during training？

[jelmerwolterink]

Hi @wuxiyu64 , thanks for reaching out. I just reran the code for pt 8 and 9 and get

8 [1.21 0.46 0.47 0.57] [1.31 0.66 0.72 1.21]
9 [1.2 0.43 0.52 0.52] [0.99 0.55 0.6 1.01]

Is this in line with the discrepancy you found?

[wuxiyu64]

Hello, thank you for your reply. My implementation results are as follows:
8 [3.53 0.83 1.25 2.79] [4.83 1.16 2.06 4.491]
9 [12.18 7.77.21 3.74] [4.7 4.74 3.95 2.66]
10 [3.97 1.09 2.15 2.57] [4.08 1.43 2.7 3.27]
This is inconsistent with the results of the article and there is a big gap. Do you set other parameters or have special skills in the training process?

[jelmerwolterink]

That is indeed a wide gap. I used the code that is on here, so no other parameters or special skills that are not in here already :) I can imagine that maybe the masks you use for the lungs are different? Are you sure those are properly aligned with the images?

[wuxiyu64]

Thanks again for your reply. I will check the accuracy of the data set again