Can I load whole 3D MRI image into my memory without performing patching or subsampling?

I have 17 GB of data collected from ADNI where I am going to classify them without any preprocessing and augmentation steps. I load a single image using nibable library. Now I am trying to load all the .nii format files from my directory.
For this purpose I already studied the basic dataloading method shown in WRITING CUSTOM DATASETS and also studied custom dataloader for 3D files .
Please help me someone with some sort of code which help me to understand that I am on right path or I have to change the direction of implementing it?

Are you planning to train the model with 2d or 3d ?

1 Like

I will train my CNN modal using 3D data

Please Help!!!
I am stuck here to load the data from the directory for that please specify some direction.

Will this help ?

I can’t get this I am not too much expert in pseudocode
I can load a single image but I can save the file in h5 file using HDF5 software which convert and save into h5 file.? Does It will be ok ?
If can share the pseudocode than It will help more to understand your point.
Thanks for fast reply