Removing images where segmentation files are empty.

[plbenveniste]

We should remove images were the segmentation files are empty, indicating no lesion.

Related to #7

We should also ask the team why some images have empty segmentations. They might still be useful. (maybe previous timepoints?)

[plbenveniste]

I used the following piece of code to list all empty label files.

Code used for listing empty label files:

import os 
from pathlib import Path
from tqdm import tqdm
import nibabel as nib
import numpy as np

path_data = "/home/GRAMES.POLYMTL.CA/p119007/critical_lesions/MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06"

# List all .nii.gz images in the folder
list_images = list(Path(path_data).glob("*.nii.gz"))
# Keep only the label files
list_labels = [str(img) for img in list_images if "label" in str(img)]

# Init list of empty labels
list_empty_labels = []

print("Nb of labels", len(list_labels))

# For each label, we check if the label is not empty
for label in tqdm(list_labels):
    # Load the image
    label_nib = nib.load(label)
    # Get the data array
    data_array = label_nib.get_fdata()
    # If the data array is empty, we remove the file
    if np.all(data_array == 0):
        print(f"Empty label found: {label.replace('/home/GRAMES.POLYMTL.CA/p119007/critical_lesions/','')}")
        list_empty_labels.append(label.replace('/home/GRAMES.POLYMTL.CA/p119007/critical_lesions/',''))
    else: 
        # We check the values included in the label
        unique_values = np.unique(data_array)
        if len(unique_values) != 3:
            print(f"Label {label} contains unexpected values: {unique_values}")
        elif np.all(unique_values != [0,1,2]):
            print("unique values in label", label, ":", unique_values)

print("Number of empty labels:", len(list_empty_labels))
print("Empty labels:")
# Display it one per line
for label in list_empty_labels:
    print(label)

Here is the output out of 241 labels:

Number of empty labels: 20
Empty labels:
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.244.3596.11880548.22001.1411646030.637-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.12.2.1107.5.2.30.25581.2014121717283919365953665.0.0.0-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.12.2.1107.5.2.6.23059.30000018060112025935900002915-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.46.670589.11.4416.7.1294169700001501300601-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.25.38818.105.1282163488-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.177.3596.7771142.23288.1185283544.464-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.244.4120.7582168.13628.1385124590.565-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.244.3596.11880548.21832.1433500207.767-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.12.2.1107.5.2.31.31356.2013092013385691542534123.0.0.0-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/2.16.840.1.113662.4.1458149515632.1364562815.3936236000945345684-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.12.2.1107.5.2.31.31117.2016091316501391100043821.0.0.0-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/2.16.840.1.113662.4.1458149515632.1142444521.5908943698116625731-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.244.3596.11880548.22154.1486126053.377-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.244.3596.13616250.25608.1421752491.352-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.244.3596.13616250.25529.1410548851.731-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.12.2.1107.5.2.6.23059.30000010012613565376500006701-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/2.16.840.1.113662.4.1458149515632.1364564325.6155896298951484853-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.46.670589.11.21609.5.0.7184.2009040611422171185-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.2.840.113619.2.410.6574342.2816289.26158.1495646347563-label.nii.gz
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.12.2.1107.5.2.30.26888.2018122607404527812711010.0.0.0-label.nii.gz

Furthermore, we checked the values included in non-empty label files: only 1 image had only 2 unique values:
MRI_Mayo_Clinic_for_Montreal/Upload-2025-05-06/1.3.12.2.1107.5.2.19.45236.2013081307033471778404060.0.0.0-label.nii.gz : [0. 1.]. All other non-empty files had [0,1,2] as unique values.

We should ask them what to do with the corresponding images. Should we consider that the images do not contain a critical lesion?