quality metrics

[WorldForever1]

quality metrics

[WorldForever1]

Should I download the preprocessed data？ such as, 100307_3T_Structural_preproc.zip/T1w/T1w_acpc_dc_restore_brain.nii.gz. Thanks!

[1632325673]

Is your problem solved？ Can you tell me how to get the test set？

[sinAshish]

any solutions for the above questions?
for the pre-processing I can see that they have provided the data_bulid.py.

[iwuqing]

Should I download the preprocessed data？ such as, 100307_3T_Structural_preproc.zip/T1w/T1w_acpc_dc_restore_brain.nii.gz. Thanks!

Yeah, the HCP-1200 dataset we download is pre-processed. We use a Python script (https://github.com/huawei-lin/HCP_Dataset_Download_Automatically_Script) to perform downloading. To download the image modality we want (e.g., T1w in our paper), we modify the lines 39-51 in this script as below:

Original (https://github.com/huawei-lin/HCP_Dataset_Download_Automatically_Script/blob/master/script.py)

    keyList = bucket.objects.filter(Prefix = prefix + '/{}/MNINonLinear/Results/tfMRI'.format(subject_number))
    keyList = [key.key for key in keyList]
    keyList = [x for x in keyList if '_LR.nii.gz' in x or '_RL.nii.gz' in x or 'EVs' in x]
    
    if not os.path.exists(output_path):
        os.makedirs(output_path)
        
    totalNumber = len(keyList)
    trycnt = 0
    
    
    
    tempKeys = [x for x in keyList if '_LR.nii.gz' in x or '_RL.nii.gz' in x]

Modified

    keyList = bucket.objects.filter(Prefix = prefix + '/{}/MNINonLinear'.format(subject_number))
    keyList = [key.key for key in keyList]
    keyList = [x for x in keyList if 'T1w_restore_brain.nii.gz' in x]
    
    if not os.path.exists(output_path):
        os.makedirs(output_path)
        
    totalNumber = len(keyList)
    trycnt = 0
    
    tempKeys = [x for x in keyList if 'T1w_restore_brain.nii.gz' in x]

After the downloading, we only conduct intensity normalization of [0, 1]. Thanks for your attention!

[iwuqing]

Second one, when I try to reproduce this experiment, there are some questions about quality metrics, LPIPS, PSI,LPC-SI. How to use the slice-by-slice strategy to compute them? Would mind sharing the code about this partment?

The three metrics are employed from their official implementations, their code links are:
LPIPS：https://github.com/richzhang/PerceptualSimilarity
PSI：https://github.com/feichtenhofer/PSI
LPC-SI：https://ece.uwaterloo.ca/~z70wang/research/lpcsi/

Moreover, the slice-by-slice strategy includes three steps:

1. Given a 3D volume, we extract its 2D MR slices from three orthogonal directions (axial, sagittal, and coronal directions).
2. We compute the scores for each 2D MR slice.
3. We average the scores of all the 2D MR slices to calculate the final scores.

The following code is to compute LPIPS using the slice-by-slice strategy:

import SimpleITK as sitk
import numpy as np
import torchs
import lpips

def compute_lpips(gt_slice, recon_slice):
    lpips_alex = lpips.LPIPS(net = 'alex')
    h, w = gt_slice.shape
    gt_image = np.zeros((1, 3, h, w))
    recon_image = np.zeros((1, 3, h, w))
    for c in range(3):
        gt_image[:, c, :, :] = gt_slice
        recon_image[:, c, :, :] = recon_slice
    lpips_value = lpips_alex(torch.tensor(gt_image).to(torch.float32), torch.tensor(recon_image).to(torch.float32))
    return lpips_value

gt_volume_path = ''
recon_volume_path = ''

gt_volume = sitk.GetArrayFromImage(sitk.ReadImage(gt_volume_path))
recon_volume = sitk.GetArrayFromImage(sitk.ReadImage(recon_volume_path))

# In any volume, the center region often includes more image information than the margin region.
# Therefore, we extract 10 2D slices in the center from three directions, respectively.
# Here the volumes are 264*264*264 size
i_strart, i_end = 127, 137

lpips_value = 0.

for i in range(i_strart, i_end):
    # direction 1
    gt_slice = gt_volume[i, :, :]
    recon_slice = recon_volume[i, :, :]
    lpips_value += compute_lpips(gt_slice, recon_slice)
    # direction 2
    gt_slice = gt_volume[:, i, :]
    recon_slice = recon_volume[:, i, :]
    lpips_value += compute_lpips(gt_slice, recon_slice)
    # direction 3
    gt_slice = gt_volume[:, :, i]
    recon_slice = recon_volume[:, :, i]
    lpips_value += compute_lpips(gt_slice, recon_slice)

lpips_value = lpips_value / 30
print('LPIPS:', lpips_value)

The rest two metrics (PSI and LPC-SI) are also computed by using similar coding in MATLAB

[WorldForever1]

Thanks, I will have a try!