File size: 7,338 Bytes
345e387
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
"""
Prepare LIDC-IDRI data for deterministic baselines.
Creates flat directories with majority-vote merged masks.
Also prepares nnU-Net format dataset.
"""
import os
import sys
import glob
import argparse
import numpy as np
from PIL import Image
from tqdm import tqdm
import shutil


def majority_vote_mask(mask_paths):
    """Create majority vote mask from multiple annotator masks (>=2/4 agree)."""
    masks = []
    for p in mask_paths:
        m = np.array(Image.open(p).convert("L"))
        m = (m > 127).astype(np.uint8)  # Binarize
        masks.append(m)
    
    # Stack and sum: pixel = 1 if >= 2 annotators agree
    stacked = np.stack(masks, axis=0)
    vote = (np.sum(stacked, axis=0) >= 2).astype(np.uint8)
    return vote * 255  # Save as 0/255 PNG


def process_split(data_dir, output_dir, split_name):
    """Process a train or test split."""
    images_dir = os.path.join(output_dir, "images")
    masks_dir = os.path.join(output_dir, "masks")
    os.makedirs(images_dir, exist_ok=True)
    os.makedirs(masks_dir, exist_ok=True)
    
    # Find all patient directories
    patient_dirs = sorted(glob.glob(os.path.join(data_dir, "LIDC-IDRI-*")))
    
    count = 0
    skipped = 0
    for patient_dir in tqdm(patient_dirs, desc=f"Processing {split_name}"):
        patient_id = os.path.basename(patient_dir)
        nodule_dirs = sorted(glob.glob(os.path.join(patient_dir, "nodule-*")))
        
        for nodule_dir in nodule_dirs:
            nodule_id = os.path.basename(nodule_dir)
            image_files = sorted(glob.glob(os.path.join(nodule_dir, "images", "slice-*.png")))
            
            for img_path in image_files:
                slice_name = os.path.basename(img_path)  # e.g., slice-0.png
                slice_id = slice_name.replace(".png", "")  # e.g., slice-0
                
                # Find all annotator masks for this slice
                mask_paths = []
                for mask_dir in sorted(glob.glob(os.path.join(nodule_dir, "mask-*"))):
                    mask_path = os.path.join(mask_dir, slice_name)
                    if os.path.exists(mask_path):
                        mask_paths.append(mask_path)
                
                if len(mask_paths) < 2:
                    skipped += 1
                    continue
                
                # Create output filename: LIDC-IDRI-0001_nodule-0_slice-0
                out_name = f"{patient_id}_{nodule_id}_{slice_id}.png"
                
                # Copy image
                shutil.copy2(img_path, os.path.join(images_dir, out_name))
                
                # Create and save majority vote mask
                mv_mask = majority_vote_mask(mask_paths)
                Image.fromarray(mv_mask).save(os.path.join(masks_dir, out_name))
                
                count += 1
    
    print(f"{split_name}: Processed {count} slices, skipped {skipped}")
    return count


def prepare_nnunet_format(flat_train_dir, flat_test_dir, nnunet_raw_dir):
    """Convert flat dataset to nnU-Net v2 format."""
    dataset_dir = os.path.join(nnunet_raw_dir, "Dataset001_LIDC")
    
    imagesTr = os.path.join(dataset_dir, "imagesTr")
    labelsTr = os.path.join(dataset_dir, "labelsTr")
    imagesTs = os.path.join(dataset_dir, "imagesTs")
    labelsTs = os.path.join(dataset_dir, "labelsTs")
    
    for d in [imagesTr, labelsTr, imagesTs, labelsTs]:
        os.makedirs(d, exist_ok=True)
    
    # nnU-Net expects: case_XXXX_0000.png for images, case_XXXX.png for labels
    # Channel suffix _0000 for single-channel
    
    print("Converting to nnU-Net format...")
    
    # Training
    train_images = sorted(glob.glob(os.path.join(flat_train_dir, "images", "*.png")))
    for i, img_path in enumerate(tqdm(train_images, desc="nnU-Net train")):
        basename = os.path.splitext(os.path.basename(img_path))[0]
        case_id = f"LIDC_{i:05d}"
        
        # Copy image with _0000 suffix
        shutil.copy2(img_path, os.path.join(imagesTr, f"{case_id}_0000.png"))
        
        # Copy mask (convert 0/255 to 0/1 for nnU-Net)
        mask_path = os.path.join(flat_train_dir, "masks", os.path.basename(img_path))
        mask = np.array(Image.open(mask_path).convert("L"))
        mask = (mask > 127).astype(np.uint8)
        Image.fromarray(mask).save(os.path.join(labelsTr, f"{case_id}.png"))
    
    # Testing
    test_images = sorted(glob.glob(os.path.join(flat_test_dir, "images", "*.png")))
    for i, img_path in enumerate(tqdm(test_images, desc="nnU-Net test")):
        basename = os.path.splitext(os.path.basename(img_path))[0]
        case_id = f"LIDC_{i:05d}"
        
        shutil.copy2(img_path, os.path.join(imagesTs, f"{case_id}_0000.png"))
        
        mask_path = os.path.join(flat_test_dir, "masks", os.path.basename(img_path))
        mask = np.array(Image.open(mask_path).convert("L"))
        mask = (mask > 127).astype(np.uint8)
        Image.fromarray(mask).save(os.path.join(labelsTs, f"{case_id}.png"))
    
    # Create dataset.json
    import json
    dataset_json = {
        "channel_names": {"0": "CT"},
        "labels": {"background": 0, "nodule": 1},
        "numTraining": len(train_images),
        "file_ending": ".png",
        "name": "Dataset001_LIDC",
        "description": "LIDC-IDRI Lung Nodule Segmentation (majority vote GT)",
        "reference": "LIDC-IDRI",
        "licence": "CC BY 3.0",
        "release": "1.0"
    }
    with open(os.path.join(dataset_dir, "dataset.json"), "w") as f:
        json.dump(dataset_json, f, indent=2)
    
    # Save mapping from nnU-Net case IDs to original names (for prediction conversion)
    mapping = {}
    for i, img_path in enumerate(sorted(glob.glob(os.path.join(flat_test_dir, "images", "*.png")))):
        case_id = f"LIDC_{i:05d}"
        original_name = os.path.splitext(os.path.basename(img_path))[0]
        mapping[case_id] = original_name
    
    with open(os.path.join(dataset_dir, "test_case_mapping.json"), "w") as f:
        json.dump(mapping, f, indent=2)
    
    print(f"nnU-Net dataset created at {dataset_dir}")
    print(f"  Training: {len(train_images)} cases")
    print(f"  Testing: {len(test_images)} cases")


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--data_root", type=str, default="data", help="Root data directory")
    parser.add_argument("--skip_nnunet", action="store_true", help="Skip nnU-Net format conversion")
    args = parser.parse_args()
    
    train_dir = os.path.join(args.data_root, "training")
    test_dir = os.path.join(args.data_root, "testing")
    
    flat_train = os.path.join(args.data_root, "flat_train")
    flat_test = os.path.join(args.data_root, "flat_test")
    
    print("=" * 60)
    print("Preparing flat dataset with majority-vote masks")
    print("=" * 60)
    
    n_train = process_split(train_dir, flat_train, "Training")
    n_test = process_split(test_dir, flat_test, "Testing")
    
    print(f"\nTotal: {n_train} train, {n_test} test slices")
    
    if not args.skip_nnunet:
        print("\n" + "=" * 60)
        print("Preparing nnU-Net format dataset")
        print("=" * 60)
        nnunet_raw = os.path.join(args.data_root, "nnUNet_raw")
        prepare_nnunet_format(flat_train, flat_test, nnunet_raw)
    
    print("\nDone!")


if __name__ == "__main__":
    main()