Upload folder using huggingface_hub
Browse files
processed_code/preprocess_ctrate_train.py
ADDED
|
@@ -0,0 +1,148 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import nibabel as nib
|
| 3 |
+
import pandas as pd
|
| 4 |
+
import numpy as np
|
| 5 |
+
import torch
|
| 6 |
+
import monai
|
| 7 |
+
import torch.nn.functional as F
|
| 8 |
+
from multiprocessing import Pool
|
| 9 |
+
from tqdm import tqdm
|
| 10 |
+
|
| 11 |
+
def read_nii_files(directory):
|
| 12 |
+
"""
|
| 13 |
+
Retrieve paths of all NIfTI files in the given directory.
|
| 14 |
+
|
| 15 |
+
Args:
|
| 16 |
+
directory (str): Path to the directory containing NIfTI files.
|
| 17 |
+
|
| 18 |
+
Returns:
|
| 19 |
+
list: List of paths to NIfTI files.
|
| 20 |
+
"""
|
| 21 |
+
nii_files = []
|
| 22 |
+
for root, dirs, files in os.walk(directory):
|
| 23 |
+
for file in files:
|
| 24 |
+
if file.endswith('1.nii.gz'):
|
| 25 |
+
# /mnt/petrelfs/share_data/zhangxiaoman/DATA/CT-RATE/dataset/train_preprocessed
|
| 26 |
+
# preprocessed_file = file.replace('/mnt/petrelfs/share_data/zhangxiaoman/DATA/CT-RATE/dataset/train','/mnt/petrelfs/share_data/zhangxiaoman/DATA/CT-RATE/dataset/train_preprocessed')
|
| 27 |
+
nii_files.append(os.path.join(root, file))
|
| 28 |
+
return nii_files
|
| 29 |
+
|
| 30 |
+
def read_nii_data(file_path):
|
| 31 |
+
"""
|
| 32 |
+
Read NIfTI file data.
|
| 33 |
+
|
| 34 |
+
Args:
|
| 35 |
+
file_path (str): Path to the NIfTI file.
|
| 36 |
+
|
| 37 |
+
Returns:
|
| 38 |
+
np.ndarray: NIfTI file data.
|
| 39 |
+
"""
|
| 40 |
+
try:
|
| 41 |
+
nii_img = nib.load(file_path)
|
| 42 |
+
nii_data = nii_img.get_fdata()
|
| 43 |
+
return nii_data
|
| 44 |
+
except Exception as e:
|
| 45 |
+
print(f"Error reading file {file_path}: {e}")
|
| 46 |
+
return None
|
| 47 |
+
|
| 48 |
+
def resize_array(array, current_spacing, target_spacing):
|
| 49 |
+
"""
|
| 50 |
+
Resize the array to match the target spacing.
|
| 51 |
+
|
| 52 |
+
Args:
|
| 53 |
+
array (torch.Tensor): Input array to be resized.
|
| 54 |
+
current_spacing (tuple): Current voxel spacing (z_spacing, xy_spacing, xy_spacing).
|
| 55 |
+
target_spacing (tuple): Target voxel spacing (target_z_spacing, target_x_spacing, target_y_spacing).
|
| 56 |
+
|
| 57 |
+
Returns:
|
| 58 |
+
np.ndarray: Resized array.
|
| 59 |
+
"""
|
| 60 |
+
# Calculate new dimensions
|
| 61 |
+
original_shape = array.shape[2:]
|
| 62 |
+
scaling_factors = [
|
| 63 |
+
current_spacing[i] / target_spacing[i] for i in range(len(original_shape))
|
| 64 |
+
]
|
| 65 |
+
new_shape = [
|
| 66 |
+
int(original_shape[i] * scaling_factors[i]) for i in range(len(original_shape))
|
| 67 |
+
]
|
| 68 |
+
# Resize the array
|
| 69 |
+
resized_array = F.interpolate(array, size=new_shape, mode='trilinear', align_corners=False).cpu().numpy()
|
| 70 |
+
return resized_array
|
| 71 |
+
|
| 72 |
+
def process_file(file_path):
|
| 73 |
+
"""
|
| 74 |
+
Process a single NIfTI file.
|
| 75 |
+
|
| 76 |
+
Args:
|
| 77 |
+
file_path (str): Path to the NIfTI file.
|
| 78 |
+
|
| 79 |
+
Returns:
|
| 80 |
+
None
|
| 81 |
+
"""
|
| 82 |
+
monai_loader = monai.transforms.Compose(
|
| 83 |
+
[
|
| 84 |
+
monai.transforms.LoadImaged(keys=['image']),
|
| 85 |
+
monai.transforms.AddChanneld(keys=['image']),
|
| 86 |
+
monai.transforms.Orientationd(axcodes="LPS", keys=['image']), # zyx
|
| 87 |
+
# monai.transforms.Spacingd(keys=["image"], pixdim=(1, 1, 3), mode=("bilinear")),
|
| 88 |
+
monai.transforms.CropForegroundd(keys=["image"], source_key="image"),
|
| 89 |
+
monai.transforms.ToTensord(keys=["image"]),
|
| 90 |
+
]
|
| 91 |
+
)
|
| 92 |
+
|
| 93 |
+
dictionary = monai_loader({'image':file_path})
|
| 94 |
+
img_data = dictionary['image']
|
| 95 |
+
|
| 96 |
+
file_name = os.path.basename(file_path)
|
| 97 |
+
row = df[df['VolumeName'] == file_name]
|
| 98 |
+
slope = float(row["RescaleSlope"].iloc[0])
|
| 99 |
+
intercept = float(row["RescaleIntercept"].iloc[0])
|
| 100 |
+
xy_spacing = float(row["XYSpacing"].iloc[0][1:][:-2].split(",")[0])
|
| 101 |
+
z_spacing = float(row["ZSpacing"].iloc[0])
|
| 102 |
+
|
| 103 |
+
# Define the target spacing values for SAT segmentation
|
| 104 |
+
target_x_spacing = 1.0
|
| 105 |
+
target_y_spacing = 1.0
|
| 106 |
+
target_z_spacing = 3.0
|
| 107 |
+
|
| 108 |
+
current = (z_spacing, xy_spacing, xy_spacing)
|
| 109 |
+
target = (target_z_spacing, target_x_spacing, target_y_spacing)
|
| 110 |
+
img_data = slope * img_data + intercept
|
| 111 |
+
|
| 112 |
+
img_data = img_data[0].numpy()
|
| 113 |
+
img_data = img_data.transpose(2, 0, 1)
|
| 114 |
+
tensor = torch.tensor(img_data)
|
| 115 |
+
tensor = tensor.unsqueeze(0).unsqueeze(0)
|
| 116 |
+
|
| 117 |
+
resized_array = resize_array(tensor, current, target)
|
| 118 |
+
resized_array = resized_array[0][0]
|
| 119 |
+
resized_array = resized_array.transpose(1,2,0)
|
| 120 |
+
# print('resized:',resized_array.shape)
|
| 121 |
+
# resized: (231, 387, 387)
|
| 122 |
+
|
| 123 |
+
save_folder = "../upload_data/train_preprocessed/" #save folder for preprocessed
|
| 124 |
+
folder_path_new = os.path.join(save_folder, "train_" + file_name.split("_")[1], "train_" + file_name.split("_")[1] + file_name.split("_")[2]) #folder name for train or validation
|
| 125 |
+
os.makedirs(folder_path_new, exist_ok=True)
|
| 126 |
+
save_path = os.path.join(folder_path_new, file_name)
|
| 127 |
+
# np.savez(save_path, resized_array)
|
| 128 |
+
# Create an identity matrix
|
| 129 |
+
|
| 130 |
+
image_nifti = nib.Nifti1Image(resized_array,affine = np.eye(4))
|
| 131 |
+
nib.save(image_nifti, save_path)
|
| 132 |
+
|
| 133 |
+
|
| 134 |
+
|
| 135 |
+
|
| 136 |
+
# Example usage:
|
| 137 |
+
if __name__ == "__main__":
|
| 138 |
+
split_to_preprocess = '../src_data/train' #select the validation or test split
|
| 139 |
+
nii_files = read_nii_files(split_to_preprocess)
|
| 140 |
+
print(len(nii_files))
|
| 141 |
+
|
| 142 |
+
df = pd.read_csv("../src_data/metadata/train_metadata.csv") #select the metadata
|
| 143 |
+
|
| 144 |
+
num_workers = 18 # Number of worker processes
|
| 145 |
+
|
| 146 |
+
# # # Process files using multiprocessing with tqdm progress bar
|
| 147 |
+
with Pool(num_workers) as pool:
|
| 148 |
+
list(tqdm(pool.imap(process_file, nii_files), total=len(nii_files)))
|