lmmmeng / lld-mmri2023 Goto Github PK

Dear Competition Organizer's,

I am writing to inquire about the release of the annotations for the validation set. As per the schedule provided, it was mentioned that the annotations would be released on July 8th. However, to date, I have not been able to access the annotations.

I would be grateful if you could kindly provide an update regarding the release of the annotations for the validation set.Having access to the validation set annotations is crucial for participants like me to evaluate and refine our models effectively.

Thank you very much for your attention to this matter. I look forward to your response and appreciate your assistance in resolving this issue.

I wonder if you've provided a version of baseline for windows

Are the training and test sets for the competition allowed to be publicly disseminated?

Hi! My team "Liuhuhu" already filled in the registration information, but we have not received the data link, plz tell us if there is anything wrong.Thanks!

Like this issue "#1", I used the command "type lld_mmri2023_part0* > lld_mmri2023.zip" to concatenate the dataset files to a new dataset file. But I also can't decompress the new dataset file. Additionally, I use the command "unzip lld_mmri2023.zip", it shows a error "Trying to read large file (> 2 GiB) without large file support".
The Windows system is used to conduct these operations.

Hi!
I used the code 'image.GetSpacing()' and found that the voxel of all data is (1.0, 1.0, 1.0). Has the data been resampled?

Hello,
I checked the full image using ITK-SNAP and I found that the shown spacing is 1x1x1mm which is inconsistent with the true data volume. Maybe the authors only change the nii files' head but not resample the nii volume?

After I download the dataset, I want to decompress the dataset, but they all prompt that the file is empty. Is it my problem or is this a common problem?

Hello,
I apologize for bothering again.
I have noticed that some cases exhibit repetitive slices in both their "in" and "out" phases.
For example, in case MR2603, the upper part of the kidney is repeated twice.


Also cases like MR3573 MR13219 MR17022 MR17217 （maybe others） have the same issue.

I am feeling somewhat confused because if I intend to incorporate a registration method into my pipeline, these repetitive slices could potentially impact its performance. Do you have any suggestions on how to address this issue? Is it necessary to repair it?

Hello,
I recovered the original image spacing by using the given annotations. I found the Z spacing of the contrast-enhenced phases are incorrect, as shown in the below image. Other phases' spacing are visually correct.

I run the baseline, but the results seem a bit incorrect.
I achieve the best value in validation:
acc: 0.36923076923076925
f1: 0.24474850546279117
recall: 0.2687074829931973
precision: 0.2650609080841639
kappa: 0.16718749999999993
The performance is too low. Can you provide the classification performance of baseline you running?

The raw data has been divided into 5-fold sub-datasets. However, the original code (main.py) could only process one-fold data every time. What is the actual process of training and validating? How can it be validated by randomly selecting 5-fold, only 1-fold, or the provided 5-fold?

Dear LLD-MMRI 2023 Team,

could you please provide more information about how to access the labels of the validation set?

Thanks in advance!