Add condition to process '.txt' files with list of fMRI files from different subjects

[rsilva8]

MISA-pytorch/dataset/dataset.py

Line 30 in de9e5ca

if '.mat' in data_file:

add an elif conditional statement for '.txt' extension.
Inside, use os.open in reading mode to read in the list of files (use with clause).
Each filename goes in as an item in the list.

[rsilva8]

After reading all file names, load the .nii files from each subject using nib.load in a for loop.
For each subject, their Time-by-3D fMRI array needs to be masked and reshaped into a 2D Time-by-Voxel matrix.
These 2D matrices should then be stacked into a 3D array (one slice per subject).
Finally, convert the array to torch tensor.

[rsilva8]

You will need this code to load the files and arrange into the 3D matrix (fnames is the list of file names).
You will also need an input argument for the mask filename.

import nibabel as nib
import numpy as np

# Example maskfname:
# "/data/mialab2/competition2019/UKBiobank/process/results/fmri_mask_generation_after_QC_etc/ica_analysisMask_intersect_NMKtemplate.nii"

def mask_data_faster(img, mask):
    container = nib.Nifti1Image.get_fdata(img,dtype= np.float64)
    # mask = nib.Nifti1Image.get_fdata(masker, dtype=np.float32)
    # mask = mask == 1
    print(container.dtype)
    return container[mask].T # returns T x N (time x samples/voxels)

def load_and_mask_nii(fnames, maskfname):
    # Load mask and change to logical (bool):
    masker = nib.load(maskfname)
    mask = nib.Nifti1Image.get_fdata(masker, dtype=np.float32)
    mask = mask == 1
    
    # Load and mask .nii files, reshaping and stacking
    imgs = [nib.load(fname) for fname in fnames]
    X = np.stack([mask_data_faster(img, mask) for img in imgs])
    return X

[rsilva8]

Put these subject list files in the slurm folder:

ukb_fMRIpaths_unaffected_11754.txt
ukb_fMRIpaths_unaffected_100.txt

[rsilva8]

This is the file with the Wpca matrix we will use: /data/users3/rsilva/dask-iva/W_gpca_M100_K12_ukb_fMRI.npy