What is the best way to organize dataset before loading them to pytorch and training?

i am trying to work on 3D-IRCADb database to segment the vessels, where data divided into 20 folders end each folder has 3 folders (Dicom, Mask, vtk file ) , i want to train and test netwok on this dataset but I’m unsure as to how to go about getting these images set up to easily load in PyTorch. My prior experience has been using the CIFAR 10 dataset, which was already set up and easy to load