QianyeYang / mpmrireg

The source code for paper "Cross-Modality Image Registration using a Training-Time Privileged Third Modality"
MIT License
8 stars 1 forks source link

Data issues #1

Open qilin-world opened 6 months ago

qilin-world commented 6 months ago

https://drive.google.com/file/d/15l4IBfNUTdOwQL6rY2H6ekpwfaeNIfPj/view?usp=sharing "The fully-preprocessed data" provided by the author includes | -- t2.npy |---Dwi.npy |---Dwi_ldmk_1. npy Dwi_ldmkn.npy |---T2uldmk_1. npy T2uldmk.npy; But it's missing | -- dwi_b0.npy that causes the test.py file to not run properly

QianyeYang commented 4 months ago

https://drive.google.com/file/d/15l4IBfNUTdOwQL6rY2H6ekpwfaeNIfPj/view?usp=sharing "The fully-preprocessed data" provided by the author includes | -- t2.npy |---Dwi.npy |---Dwi_ldmk_1. npy Dwi_ldmkn.npy |---T2uldmk_1. npy T2uldmk.npy; But it's missing | -- dwi_b0.npy that causes the test.py file to not run properly

Hi,

Thanks for the feedback.

Actually it does not require dwi_b0 during the inference - the dwi_b0 is only used for training, but will be not used in the inference.

In the code of the dataloader at: https://github.com/QianyeYang/mpmrireg/blob/92077f41c3a9c646f0b5a009da55895097dfe507/src/data/dataloaders.py#L38-L42 We can see that when we are using external data set for inference (which do not contain dwi_b0), it will pass a fake dwi_b0 path in, which is the same to the dwi paths. This is for maintaining the data structure of the code without introducing many if-else conditions - it perhaps not a good habit for coding, but should not the reason for causing the mentioned error.

In this case, your error is probably not caused by the missing dwi_b0 images. If you can provide the error informations I'm able to offer more guidance.

qilin-world commented 4 months ago

Hello author, I would like to replicate your training process by attempting to run four scripts in your 'scripts/mpmrireg' folder, all of which reported the same error: FileNotFoundError: [Errno 2] No such file or directory: '/ Data/mpmireg/train/mr-0/dwi_b0. npy '(as shown in Figure 1), may I ask how I can solve it, or could you provide more comprehensive data, as shown in Figure 2. Thanks! Figure 1 Figure 2

QianyeYang commented 4 months ago

Hello author, I would like to replicate your training process by attempting to run four scripts in your 'scripts/mpmrireg' folder, all of which reported the same error: FileNotFoundError: [Errno 2] No such file or directory: '/ Data/mpmireg/train/mr-0/dwi_b0. npy '(as shown in Figure 1), may I ask how I can solve it, or could you provide more comprehensive data, as shown in Figure 2. Thanks! Figure 1 Figure 2

Hi,

The purpose of the provided data is only for inference, cause we are not able to find any public data set that contains T2w, low-b diffusion and high-b diffusion MR scans at the same time. The training data used in this paper is collected from either hospital or company, which are not able to be released due to ethical concerns.

However, this method could be applied to applications which have routinely acquired multi-modality scans with public data avaliable. Such as brain MR scan, or ProstateX data set (contains no b0-dwi but also with other modality instead).

And you could create dummy dwi_b0 images, for example, make them identical to corresponding dwi images, just to check if the network could be trained as expected.