I’m working on Alzheimer’s classification using the ADNI dataset and need advice on two fronts:

Dataset Distribution:
Different studies seem to use varying subsets of the ADNI dataset for their research. Is there a widely accepted or optimal distribution of the ADNI dataset for Alzheimer’s classification tasks?

Preprocessing Challenges:
The original ADNI images are in DICOM format, making conversion to NIFTI format challenging. Meanwhile, processed images are in NIFTI format, but each has undergone different preprocessing steps. Are there standard preprocessing pipelines for ADNI data, considering the legal constraints preventing the sharing of raw data?

I have already tried to use ADNI screening data which resulted in very low accuracy for domain generalization case. I also tried ADNI data with raw MP-RAGE images but they were in DICOM format. The processed images were in NIFTII which is the desired format but there were no common pre-processing steps. Hence for each subject, i didnt have common pre-processed image. If i introduce images per subject with different pre-processing steps, i believe that would be an issue for the model.

Any insights or recommendations on these issues would be greatly appreciated. Thanks!