File size: 8,352 Bytes
a7e0222
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
"""
Sample Data Loader for NeuroSAM3.

Provides curated demo images (brain tumors + healthy) from an HF Dataset repo.
Images are loaded on-demand via HuggingFace Hub API — no bundling in the Space repo.

Dataset: mmrech/neurosam3-samples (private HF Dataset)
Source images from:
  - Figshare: brain_tumor_dataset (3064 T1-weighted MRIs, 3 tumor types)
  - Kaggle: brain-mri-scans-for-brain-tumor-classification
"""

from typing import Optional, List, Dict, Any, Tuple
import os
import tempfile
from pathlib import Path
from logger_config import logger
from config import HF_TOKEN

# Dataset configuration
SAMPLE_DATASET_REPO = "mmrech/neurosam3-samples"
SAMPLE_CACHE_DIR = os.path.join(tempfile.gettempdir(), "neurosam3_samples")

# Sample categories available in the dataset
SAMPLE_CATEGORIES = {
    "glioma": {
        "description": "Glioma tumors (T1-weighted MRI)",
        "count": 8,
        "modality": "MRI",
        "pathology": True,
    },
    "meningioma": {
        "description": "Meningioma tumors (T1-weighted MRI)",
        "count": 6,
        "modality": "MRI",
        "pathology": True,
    },
    "pituitary": {
        "description": "Pituitary tumors (T1-weighted MRI)",
        "count": 6,
        "modality": "MRI",
        "pathology": True,
    },
    "healthy": {
        "description": "Normal brain (T1/T2 MRI)",
        "count": 5,
        "modality": "MRI",
        "pathology": False,
    },
    "ct_normal": {
        "description": "Normal brain CT scans",
        "count": 3,
        "modality": "CT",
        "pathology": False,
    },
    "ct_hemorrhage": {
        "description": "CT with intracranial hemorrhage",
        "count": 2,
        "modality": "CT",
        "pathology": True,
    },
}


def get_sample_categories() -> Dict[str, Dict[str, Any]]:
    """Get available sample categories and their metadata."""
    return SAMPLE_CATEGORIES


def list_samples(category: Optional[str] = None) -> List[Dict[str, str]]:
    """
    List available sample images, optionally filtered by category.

    Returns:
        List of dicts with 'filename', 'category', 'modality', 'description'
    """
    samples = []

    categories = [category] if category else list(SAMPLE_CATEGORIES.keys())

    for cat in categories:
        if cat not in SAMPLE_CATEGORIES:
            continue
        info = SAMPLE_CATEGORIES[cat]
        for i in range(1, info["count"] + 1):
            samples.append({
                "filename": f"{cat}/{cat}_{i:03d}.png",
                "category": cat,
                "modality": info["modality"],
                "description": f"{info['description']} (sample {i})",
                "has_pathology": info["pathology"],
            })

    return samples


def load_sample_image(
    category: str,
    index: int = 1,
) -> Optional[str]:
    """
    Load a sample image from the HF Dataset repo.

    Args:
        category: One of SAMPLE_CATEGORIES keys
        index: Image index (1-based)

    Returns:
        Local file path to the downloaded image, or None on failure
    """
    if category not in SAMPLE_CATEGORIES:
        logger.error(f"Unknown category: {category}. Available: {list(SAMPLE_CATEGORIES.keys())}")
        return None

    info = SAMPLE_CATEGORIES[category]
    if index < 1 or index > info["count"]:
        logger.error(f"Index {index} out of range for {category} (1-{info['count']})")
        return None

    filename = f"{category}/{category}_{index:03d}.png"

    # Check cache first
    cache_path = os.path.join(SAMPLE_CACHE_DIR, filename)
    if os.path.exists(cache_path):
        return cache_path

    # Download from HF Hub
    try:
        from huggingface_hub import hf_hub_download

        local_path = hf_hub_download(
            repo_id=SAMPLE_DATASET_REPO,
            filename=filename,
            repo_type="dataset",
            token=HF_TOKEN,
            cache_dir=SAMPLE_CACHE_DIR,
        )
        logger.info(f"Loaded sample: {filename}")
        return local_path

    except Exception as e:
        logger.warning(f"Could not load sample from HF Hub: {e}")
        # Fallback: try to generate a synthetic sample
        return _generate_synthetic_sample(category, index)


def load_random_sample(
    modality: Optional[str] = None,
    pathology: Optional[bool] = None,
) -> Optional[Tuple[str, Dict[str, Any]]]:
    """
    Load a random sample image matching criteria.

    Args:
        modality: Filter by "CT" or "MRI" (None for any)
        pathology: Filter by pathology presence (None for any)

    Returns:
        Tuple of (file_path, metadata) or None
    """
    import random

    candidates = []
    for cat, info in SAMPLE_CATEGORIES.items():
        if modality and info["modality"] != modality:
            continue
        if pathology is not None and info["pathology"] != pathology:
            continue
        candidates.append(cat)

    if not candidates:
        return None

    category = random.choice(candidates)
    info = SAMPLE_CATEGORIES[category]
    index = random.randint(1, info["count"])

    path = load_sample_image(category, index)
    if path:
        return path, {
            "category": category,
            "index": index,
            "modality": info["modality"],
            "pathology": info["pathology"],
            "description": info["description"],
        }
    return None


def load_category_batch(category: str) -> List[str]:
    """
    Load all images from a category (for research pipeline demo).

    Args:
        category: Category name

    Returns:
        List of file paths
    """
    if category not in SAMPLE_CATEGORIES:
        return []

    paths = []
    info = SAMPLE_CATEGORIES[category]

    for i in range(1, info["count"] + 1):
        path = load_sample_image(category, i)
        if path:
            paths.append(path)

    return paths


def _generate_synthetic_sample(category: str, index: int) -> Optional[str]:
    """
    Generate a synthetic sample image as fallback when HF Dataset is unavailable.
    Creates a simple grayscale image with simulated structures.
    """
    try:
        import numpy as np
        from PIL import Image

        # Create synthetic brain-like image
        size = 256
        img = np.zeros((size, size), dtype=np.uint8)

        # Background with some texture
        y, x = np.ogrid[:size, :size]
        center = size // 2

        # Skull (outer ellipse)
        skull_mask = ((x - center)**2 / (110**2) + (y - center)**2 / (120**2)) <= 1
        img[skull_mask] = 30

        # Brain (inner ellipse)
        brain_mask = ((x - center)**2 / (90**2) + (y - center)**2 / (100**2)) <= 1
        img[brain_mask] = 80 + np.random.randint(0, 20, size=img[brain_mask].shape).astype(np.uint8)

        # Add pathology if category indicates it
        info = SAMPLE_CATEGORIES.get(category, {})
        if info.get("pathology", False):
            # Add a "tumor" blob
            tumor_x = center + np.random.randint(-30, 30)
            tumor_y = center + np.random.randint(-30, 30)
            tumor_r = np.random.randint(15, 35)
            tumor_mask = ((x - tumor_x)**2 + (y - tumor_y)**2) <= tumor_r**2
            img[tumor_mask & brain_mask] = 160 + np.random.randint(0, 40, size=img[tumor_mask & brain_mask].shape).astype(np.uint8)

        # Save
        os.makedirs(os.path.join(SAMPLE_CACHE_DIR, category), exist_ok=True)
        save_path = os.path.join(SAMPLE_CACHE_DIR, f"{category}/{category}_{index:03d}.png")
        Image.fromarray(img).save(save_path)

        logger.info(f"Generated synthetic sample: {save_path}")
        return save_path

    except Exception as e:
        logger.error(f"Failed to generate synthetic sample: {e}")
        return None


def get_dataset_info() -> Dict[str, Any]:
    """Get information about the sample dataset."""
    total = sum(info["count"] for info in SAMPLE_CATEGORIES.values())
    return {
        "repo": SAMPLE_DATASET_REPO,
        "total_samples": total,
        "categories": list(SAMPLE_CATEGORIES.keys()),
        "modalities": ["CT", "MRI"],
        "sources": [
            "Figshare: brain_tumor_dataset (Cheng, 2017) — 3064 T1-weighted MRIs",
            "Kaggle: brain-mri-scans-for-brain-tumor-classification",
        ],
        "note": "Images loaded on-demand from HF Hub. Synthetic fallback if unavailable.",
    }