Preprocess Data

[hoai-thuong]

To reproduce the preprocess of BTCV data, I follow your paper that had mentioned that " re-sample all the CT scans to the voxel spacing [1.5×1.5×2.0]mm3 and normalize them to have zero mean and unit variance".

I make this code bellow and got the shape of image like your preprocessed data (example 0001.h5).

But when i draw the histogram of mine and yours, it's different

I don't know if your team make sth else like clip,remove,....

I hope you will reply me or just suggest me some idea. Thank you so much

This is my code:

import os
import nibabel as nib
import numpy as np
import h5py
from scipy.ndimage import zoom

data_dir = "data"
images_dir = os.path.join(data_dir, "imagesTr")
labels_dir = os.path.join(data_dir, "labelsTr")

target_spacing = [1.5, 1.5, 2.0]

def load_nifti_file(filepath):
"""Load a NIfTI file and return the image data and header info."""
nifti_img = nib.load(filepath)
data = nifti_img.get_fdata()
spacing = nifti_img.header.get_zooms()
return data, spacing

def resample_volume(volume, original_spacing, target_spacing):

resize_factor = [o / t for o, t in zip(original_spacing, target_spacing)]
new_shape = [int(s * r) for s, r in zip(volume.shape, resize_factor)]
resampled_volume = zoom(volume, resize_factor, order=3)  # Sử dụng nội suy bậc 1
return resampled_volume

def normalize_volume(volume):

mean = np.mean(volume)
std = np.std(volume)
normalized_volume = (volume - mean) / std
return normalized_volume

def preprocess_and_save(image_path, label_path, output_path):

image, image_spacing = load_nifti_file(image_path)
label, label_spacing = load_nifti_file(label_path)


resampled_image = resample_volume(image, image_spacing, target_spacing)
resampled_label = resample_volume(label, label_spacing, target_spacing)


normalized_image = normalize_volume(resampled_image)

# Save as .h5
with h5py.File(output_path, "w") as f:
    f.create_dataset("image", data=normalized_image, compression="gzip")
    f.create_dataset("label", data=resampled_label, compression="gzip")

sample_image = os.path.join(images_dir, "img0001.nii.gz")
sample_label = os.path.join(labels_dir, "label0001.nii.gz")
output_file = "0001.h5"

preprocess_and_save(sample_image, sample_label, output_file)

with h5py.File(output_file, "r") as f:
image = f["image"][:]
label = f["label"][:]
print("Image shape:", image.shape)
print("Label shape:", label.shape)
print("Image mean:", np.mean(image))
print("Image std:", np.std(image))