Spaces:
Sleeping
Sleeping
Delete src/BrainIAC/app.py
Browse files- src/BrainIAC/app.py +0 -728
src/BrainIAC/app.py
DELETED
|
@@ -1,728 +0,0 @@
|
|
| 1 |
-
import os
|
| 2 |
-
import torch
|
| 3 |
-
import nibabel as nib
|
| 4 |
-
from flask import Flask, request, render_template, redirect, url_for, flash, jsonify
|
| 5 |
-
import tempfile
|
| 6 |
-
import yaml
|
| 7 |
-
import traceback # For detailed error printing
|
| 8 |
-
import zipfile
|
| 9 |
-
import dicom2nifti
|
| 10 |
-
import shutil
|
| 11 |
-
import subprocess # To run unzip command
|
| 12 |
-
import SimpleITK as sitk
|
| 13 |
-
import itk
|
| 14 |
-
import numpy as np
|
| 15 |
-
from scipy.signal import medfilt
|
| 16 |
-
import skimage.filters
|
| 17 |
-
import cv2 # For Gaussian Blur
|
| 18 |
-
import io # For saving plots to memory
|
| 19 |
-
import base64 # For encoding plots
|
| 20 |
-
import uuid # For unique IDs
|
| 21 |
-
|
| 22 |
-
# Configure Matplotlib for non-GUI backend *before* importing pyplot
|
| 23 |
-
import matplotlib
|
| 24 |
-
matplotlib.use('Agg')
|
| 25 |
-
import matplotlib.pyplot as plt
|
| 26 |
-
|
| 27 |
-
# --- Preprocessing Imports ---
|
| 28 |
-
try:
|
| 29 |
-
# Adjust import path based on Docker structure
|
| 30 |
-
# Assumes HD_BET is now at /app/BrainIAC/HD_BET
|
| 31 |
-
from HD_BET.run import run_hd_bet
|
| 32 |
-
# Import MONAI saliency visualizer
|
| 33 |
-
from monai.visualize.gradient_based import GuidedBackpropSmoothGrad
|
| 34 |
-
except ImportError as e:
|
| 35 |
-
print(f"Could not import HD_BET or MONAI visualize: {e}. Advanced features might fail.")
|
| 36 |
-
run_hd_bet = None
|
| 37 |
-
GuidedBackpropSmoothGrad = None
|
| 38 |
-
|
| 39 |
-
# Import necessary components from your existing modules
|
| 40 |
-
from model import Backbone, SingleScanModel, Classifier
|
| 41 |
-
# Removed: from dataset2 import NormalSynchronizedTransform3D
|
| 42 |
-
# Import specific MONAI transforms needed
|
| 43 |
-
from monai.transforms import Resized, ScaleIntensityd # Removed ToTensord, will handle manually
|
| 44 |
-
|
| 45 |
-
app = Flask(__name__)
|
| 46 |
-
app.secret_key = 'supersecretkey' # Needed for flashing messages
|
| 47 |
-
|
| 48 |
-
# --- Constants for Preprocessing ---
|
| 49 |
-
APP_DIR = os.path.dirname(__file__)
|
| 50 |
-
TEMPLATE_DIR = os.path.join(APP_DIR, "golden_image", "mni_templates")
|
| 51 |
-
PARAMS_RIGID_PATH = os.path.join(APP_DIR, "golden_image", "mni_templates", "Parameters_Rigid.txt")
|
| 52 |
-
DEFAULT_TEMPLATE_PATH = os.path.join(TEMPLATE_DIR, "nihpd_asym_13.0-18.5_t1w.nii") # Using adult template as default
|
| 53 |
-
HD_BET_CONFIG_PATH = os.path.join(APP_DIR, "HD_BET", "config.py")
|
| 54 |
-
HD_BET_MODEL_DIR = os.path.join(APP_DIR, "hdbet_model") # Path to copied models
|
| 55 |
-
|
| 56 |
-
# --- Configuration Loading ---
|
| 57 |
-
def load_config():
|
| 58 |
-
# Assuming config.yml is in the same directory as app.py
|
| 59 |
-
config_path = os.path.join(APP_DIR, 'config.yml')
|
| 60 |
-
try:
|
| 61 |
-
with open(config_path, 'r') as file:
|
| 62 |
-
config = yaml.safe_load(file)
|
| 63 |
-
# Add default image_size if not present in config
|
| 64 |
-
if 'data' not in config: config['data'] = {}
|
| 65 |
-
if 'image_size' not in config['data']: config['data']['image_size'] = [128, 128, 128]
|
| 66 |
-
|
| 67 |
-
except FileNotFoundError:
|
| 68 |
-
print(f"Error: Configuration file not found at {config_path}")
|
| 69 |
-
# Provide default config or handle error appropriately
|
| 70 |
-
config = {
|
| 71 |
-
'gpu': {'device': 'cpu'},
|
| 72 |
-
'infer': {'checkpoints': 'checkpoints/brainage_model_latest.pt'},
|
| 73 |
-
'data': {'image_size': [128, 128, 128]} # Default image size
|
| 74 |
-
}
|
| 75 |
-
return config
|
| 76 |
-
|
| 77 |
-
config = load_config()
|
| 78 |
-
# Ensure image_size is available, e.g., from config or a default
|
| 79 |
-
DEFAULT_IMAGE_SIZE = (128, 128, 128)
|
| 80 |
-
image_size_cfg = config.get('data', {}).get('image_size', DEFAULT_IMAGE_SIZE)
|
| 81 |
-
# Validate image_size format
|
| 82 |
-
if not isinstance(image_size_cfg, (list, tuple)) or len(image_size_cfg) != 3:
|
| 83 |
-
print(f"Warning: Invalid image_size in config ({image_size_cfg}). Using default {DEFAULT_IMAGE_SIZE}.")
|
| 84 |
-
image_size = DEFAULT_IMAGE_SIZE
|
| 85 |
-
else:
|
| 86 |
-
image_size = tuple(image_size_cfg) # Ensure it's a tuple for transforms
|
| 87 |
-
|
| 88 |
-
# --- Model Loading ---
|
| 89 |
-
def load_model(device, checkpoint_path):
|
| 90 |
-
backbone = Backbone()
|
| 91 |
-
classifier = Classifier(d_model=2048) # Make sure d_model matches your trained model
|
| 92 |
-
model = SingleScanModel(backbone, classifier)
|
| 93 |
-
|
| 94 |
-
try:
|
| 95 |
-
# Construct absolute path if checkpoint_path is relative
|
| 96 |
-
relative_path = config.get('infer', {}).get('checkpoints', 'checkpoints/brainage_model_latest.pt')
|
| 97 |
-
# Use path relative to app.py location
|
| 98 |
-
checkpoint_path_abs = os.path.join(APP_DIR, relative_path)
|
| 99 |
-
|
| 100 |
-
checkpoint = torch.load(checkpoint_path_abs, map_location=device)
|
| 101 |
-
# Adjust key if necessary based on how model was saved
|
| 102 |
-
if 'model_state_dict' in checkpoint:
|
| 103 |
-
model.load_state_dict(checkpoint['model_state_dict'])
|
| 104 |
-
else:
|
| 105 |
-
model.load_state_dict(checkpoint)
|
| 106 |
-
model.to(device)
|
| 107 |
-
model.eval()
|
| 108 |
-
print(f"Model loaded successfully from {checkpoint_path_abs} onto {device}.")
|
| 109 |
-
return model
|
| 110 |
-
except FileNotFoundError:
|
| 111 |
-
print(f"Error: Checkpoint file not found at {checkpoint_path_abs}")
|
| 112 |
-
return None
|
| 113 |
-
except Exception as e:
|
| 114 |
-
print(f"Error loading model checkpoint: {e}")
|
| 115 |
-
traceback.print_exc()
|
| 116 |
-
return None
|
| 117 |
-
|
| 118 |
-
device = torch.device(config.get('gpu', {}).get('device', 'cpu')) # Default to CPU
|
| 119 |
-
model = load_model(device, config) # Pass full config for path finding
|
| 120 |
-
|
| 121 |
-
# --- Preprocessing Functions from preprocess_utils.py ---
|
| 122 |
-
def bias_field_correction(img_array):
|
| 123 |
-
"""Performs N4 bias field correction using SimpleITK."""
|
| 124 |
-
image = sitk.GetImageFromArray(img_array)
|
| 125 |
-
# Ensure image is float32 for N4
|
| 126 |
-
if image.GetPixelID() != sitk.sitkFloat32:
|
| 127 |
-
image = sitk.Cast(image, sitk.sitkFloat32)
|
| 128 |
-
maskImage = sitk.OtsuThreshold(image, 0, 1, 200)
|
| 129 |
-
corrector = sitk.N4BiasFieldCorrectionImageFilter()
|
| 130 |
-
numberFittingLevels = 4
|
| 131 |
-
# Define iterations per level more robustly
|
| 132 |
-
max_iters = [min(50 * (2**i), 200) for i in range(numberFittingLevels)]
|
| 133 |
-
corrector.SetMaximumNumberOfIterations(max_iters)
|
| 134 |
-
# Set convergence threshold (optional, can speed up)
|
| 135 |
-
# corrector.SetConvergenceThreshold(1e-6)
|
| 136 |
-
print(" Running N4 Bias Field Correction...")
|
| 137 |
-
corrected_image = corrector.Execute(image, maskImage)
|
| 138 |
-
print(" N4 Correction finished.")
|
| 139 |
-
return sitk.GetArrayFromImage(corrected_image)
|
| 140 |
-
|
| 141 |
-
def denoise(volume, kernel_size=3):
|
| 142 |
-
"""Applies median filter for denoising."""
|
| 143 |
-
print(f" Applying median filter denoising (kernel={kernel_size})...")
|
| 144 |
-
return medfilt(volume, kernel_size)
|
| 145 |
-
|
| 146 |
-
def rescale_intensity(volume, percentils=[0.5, 99.5], bins_num=256):
|
| 147 |
-
"""Rescales intensity after removing background via Otsu."""
|
| 148 |
-
print(" Rescaling intensity...")
|
| 149 |
-
# Ensure input is float for Otsu and calculations
|
| 150 |
-
volume_float = volume.astype(np.float32)
|
| 151 |
-
try:
|
| 152 |
-
t = skimage.filters.threshold_otsu(volume_float, nbins=256)
|
| 153 |
-
print(f" Otsu threshold found: {t}")
|
| 154 |
-
volume_masked = np.copy(volume_float)
|
| 155 |
-
volume_masked[volume_masked < t] = 0 # Apply mask based on original values
|
| 156 |
-
obj_volume = volume_masked[np.where(volume_masked > 0)]
|
| 157 |
-
except ValueError: # Handle cases with near-uniform intensity
|
| 158 |
-
print(" Otsu failed (likely uniform image), skipping background mask.")
|
| 159 |
-
obj_volume = volume_float.flatten()
|
| 160 |
-
|
| 161 |
-
if obj_volume.size == 0:
|
| 162 |
-
print(" Warning: No foreground voxels found after Otsu. Scaling full volume.")
|
| 163 |
-
obj_volume = volume_float.flatten() # Fallback to full volume
|
| 164 |
-
min_value = np.min(obj_volume)
|
| 165 |
-
max_value = np.max(obj_volume)
|
| 166 |
-
else:
|
| 167 |
-
min_value = np.percentile(obj_volume, percentils[0])
|
| 168 |
-
max_value = np.percentile(obj_volume, percentils[1])
|
| 169 |
-
|
| 170 |
-
print(f" Intensity range used for scaling: [{min_value:.2f}, {max_value:.2f}]")
|
| 171 |
-
# Avoid division by zero if max == min
|
| 172 |
-
denominator = max_value - min_value
|
| 173 |
-
if denominator < 1e-6: denominator = 1e-6
|
| 174 |
-
|
| 175 |
-
# Create a copy to modify for output
|
| 176 |
-
output_volume = np.copy(volume_float)
|
| 177 |
-
# Apply scaling only to the object volume identified (or full volume as fallback)
|
| 178 |
-
if bins_num == 0:
|
| 179 |
-
# Scale to 0-1 (float)
|
| 180 |
-
output_volume = (volume_float - min_value) / denominator
|
| 181 |
-
output_volume = np.clip(output_volume, 0.0, 1.0) # Clip results to [0, 1]
|
| 182 |
-
else:
|
| 183 |
-
# Scale and bin
|
| 184 |
-
output_volume = np.round((volume_float - min_value) / denominator * (bins_num - 1))
|
| 185 |
-
output_volume = np.clip(output_volume, 0, bins_num - 1) # Ensure within bin range
|
| 186 |
-
|
| 187 |
-
# Ensure output is float32 for consistency
|
| 188 |
-
return output_volume.astype(np.float32)
|
| 189 |
-
|
| 190 |
-
def equalize_hist(volume, bins_num=256):
|
| 191 |
-
"""Performs histogram equalization on non-zero voxels."""
|
| 192 |
-
print(" Performing histogram equalization...")
|
| 193 |
-
# Create a mask of non-zero voxels
|
| 194 |
-
mask = volume > 1e-6 # Use a small epsilon for float comparison
|
| 195 |
-
obj_volume = volume[mask]
|
| 196 |
-
|
| 197 |
-
if obj_volume.size == 0:
|
| 198 |
-
print(" Warning: No non-zero voxels found for histogram equalization. Skipping.")
|
| 199 |
-
return volume # Return original volume if no foreground
|
| 200 |
-
|
| 201 |
-
# Compute histogram and CDF on the non-zero voxels
|
| 202 |
-
hist, bins = np.histogram(obj_volume, bins_num, range=(obj_volume.min(), obj_volume.max()))
|
| 203 |
-
cdf = hist.cumsum()
|
| 204 |
-
|
| 205 |
-
# Normalize CDF
|
| 206 |
-
cdf_normalized = (bins_num - 1) * cdf / float(cdf[-1])
|
| 207 |
-
|
| 208 |
-
# Interpolate new values for the object volume
|
| 209 |
-
equalized_obj_volume = np.interp(obj_volume, bins[:-1], cdf_normalized)
|
| 210 |
-
|
| 211 |
-
# Create a copy of the original volume to put the results back
|
| 212 |
-
equalized_volume = np.copy(volume)
|
| 213 |
-
equalized_volume[mask] = equalized_obj_volume
|
| 214 |
-
|
| 215 |
-
# Ensure output is float32
|
| 216 |
-
return equalized_volume.astype(np.float32)
|
| 217 |
-
|
| 218 |
-
def enhance(img_array, run_bias_correction=True, kernel_size=3, percentils=[0.5, 99.5], bins_num=256, run_equalize_hist=True):
|
| 219 |
-
"""Full enhancement pipeline from preprocess_utils."""
|
| 220 |
-
print("Starting enhancement pipeline...")
|
| 221 |
-
volume = img_array.astype(np.float32) # Ensure float input
|
| 222 |
-
try:
|
| 223 |
-
if run_bias_correction:
|
| 224 |
-
volume = bias_field_correction(volume)
|
| 225 |
-
volume = denoise(volume, kernel_size)
|
| 226 |
-
volume = rescale_intensity(volume, percentils, bins_num)
|
| 227 |
-
if run_equalize_hist:
|
| 228 |
-
volume = equalize_hist(volume, bins_num)
|
| 229 |
-
print("Enhancement pipeline finished.")
|
| 230 |
-
return volume
|
| 231 |
-
except Exception as e:
|
| 232 |
-
print(f"Error during enhancement: {e}")
|
| 233 |
-
traceback.print_exc()
|
| 234 |
-
raise RuntimeError(f"Failed enhancing image: {e}") # Re-raise to stop processing
|
| 235 |
-
|
| 236 |
-
# --- Registration Function (modified enhance call) ---
|
| 237 |
-
def register_image(input_nifti_path, output_nifti_path):
|
| 238 |
-
"""Registers input NIfTI to the default template using Elastix."""
|
| 239 |
-
print(f"Registering {input_nifti_path} to {DEFAULT_TEMPLATE_PATH}")
|
| 240 |
-
if not os.path.exists(PARAMS_RIGID_PATH):
|
| 241 |
-
raise FileNotFoundError(f"Elastix parameter file not found at {PARAMS_RIGID_PATH}")
|
| 242 |
-
if not os.path.exists(DEFAULT_TEMPLATE_PATH):
|
| 243 |
-
raise FileNotFoundError(f"Default template file not found at {DEFAULT_TEMPLATE_PATH}")
|
| 244 |
-
|
| 245 |
-
fixed_image = itk.imread(DEFAULT_TEMPLATE_PATH, itk.F)
|
| 246 |
-
moving_image = itk.imread(input_nifti_path, itk.F)
|
| 247 |
-
|
| 248 |
-
parameter_object = itk.ParameterObject.New()
|
| 249 |
-
parameter_object.AddParameterFile(PARAMS_RIGID_PATH)
|
| 250 |
-
|
| 251 |
-
result_image, _ = itk.elastix_registration_method(
|
| 252 |
-
fixed_image, moving_image,
|
| 253 |
-
parameter_object=parameter_object,
|
| 254 |
-
log_to_console=False # Keep console clean
|
| 255 |
-
)
|
| 256 |
-
itk.imwrite(result_image, output_nifti_path)
|
| 257 |
-
print(f"Registration output saved to {output_nifti_path}")
|
| 258 |
-
|
| 259 |
-
# --- Enhanced Image Function (calls actual enhance) ---
|
| 260 |
-
def run_enhance_on_file(input_nifti_path, output_nifti_path):
|
| 261 |
-
"""Reads NIfTI, runs enhance pipeline, saves NIfTI."""
|
| 262 |
-
print(f"Running full enhancement on {input_nifti_path}")
|
| 263 |
-
img_sitk = sitk.ReadImage(input_nifti_path)
|
| 264 |
-
img_array = sitk.GetArrayFromImage(img_sitk)
|
| 265 |
-
|
| 266 |
-
# Run the actual enhancement pipeline
|
| 267 |
-
enhanced_array = enhance(img_array, run_bias_correction=True) # Assuming N4 is desired
|
| 268 |
-
|
| 269 |
-
enhanced_img_sitk = sitk.GetImageFromArray(enhanced_array)
|
| 270 |
-
enhanced_img_sitk.CopyInformation(img_sitk) # Preserve metadata
|
| 271 |
-
sitk.WriteImage(enhanced_img_sitk, output_nifti_path)
|
| 272 |
-
print(f"Enhanced image saved to {output_nifti_path}")
|
| 273 |
-
|
| 274 |
-
# --- Skull Stripping Function (Set Environment Variable) ---
|
| 275 |
-
def run_skull_stripping(input_nifti_path, output_dir):
|
| 276 |
-
"""Runs HD-BET skull stripping."""
|
| 277 |
-
print(f"Running HD-BET skull stripping on {input_nifti_path}")
|
| 278 |
-
if run_hd_bet is None:
|
| 279 |
-
raise RuntimeError("HD-BET module could not be imported. Cannot perform skull stripping.")
|
| 280 |
-
|
| 281 |
-
# Removed environment variable setting as path is fixed in HD_BET/paths.py
|
| 282 |
-
# # Set environment variable *before* calling run_hd_bet
|
| 283 |
-
# # Ensure the target directory exists
|
| 284 |
-
# if not os.path.isdir(HD_BET_MODEL_DIR):
|
| 285 |
-
# raise FileNotFoundError(f"HD-BET model directory not found at specified path: {HD_BET_MODEL_DIR}")
|
| 286 |
-
# print(f"Setting HD_BET_MODELS environment variable to: {HD_BET_MODEL_DIR}")
|
| 287 |
-
# os.environ['HD_BET_MODELS'] = HD_BET_MODEL_DIR
|
| 288 |
-
|
| 289 |
-
# Check config path
|
| 290 |
-
if not os.path.exists(HD_BET_CONFIG_PATH):
|
| 291 |
-
alt_config_path = os.path.join(APP_DIR, "HD_BET", "HD_BET", "config.py")
|
| 292 |
-
if os.path.exists(alt_config_path):
|
| 293 |
-
print(f"Warning: Using alternative HD-BET config path: {alt_config_path}")
|
| 294 |
-
config_to_use = alt_config_path
|
| 295 |
-
else:
|
| 296 |
-
raise FileNotFoundError(f"HD-BET config file not found at {HD_BET_CONFIG_PATH} or {alt_config_path}")
|
| 297 |
-
else:
|
| 298 |
-
config_to_use = HD_BET_CONFIG_PATH
|
| 299 |
-
|
| 300 |
-
# Define output paths
|
| 301 |
-
base_name = os.path.basename(input_nifti_path).replace(".nii.gz", "").replace(".nii", "")
|
| 302 |
-
output_file_path = os.path.join(output_dir, f"{base_name}_bet.nii.gz")
|
| 303 |
-
output_mask_path = os.path.join(output_dir, f"{base_name}_bet_mask.nii.gz")
|
| 304 |
-
|
| 305 |
-
# Make sure output directory exists
|
| 306 |
-
os.makedirs(output_dir, exist_ok=True)
|
| 307 |
-
|
| 308 |
-
# Run HD-BET
|
| 309 |
-
run_hd_bet(input_nifti_path, output_file_path,
|
| 310 |
-
mode="fast",
|
| 311 |
-
device='cpu',
|
| 312 |
-
config_file=config_to_use,
|
| 313 |
-
postprocess=False,
|
| 314 |
-
do_tta=False,
|
| 315 |
-
keep_mask=True,
|
| 316 |
-
overwrite=True)
|
| 317 |
-
|
| 318 |
-
# Unset environment variable after use (optional, good practice)
|
| 319 |
-
# del os.environ['HD_BET_MODELS']
|
| 320 |
-
|
| 321 |
-
if not os.path.exists(output_file_path):
|
| 322 |
-
raise RuntimeError(f"HD-BET did not produce the expected output file: {output_file_path}")
|
| 323 |
-
|
| 324 |
-
print(f"Skull stripping output saved to {output_file_path}")
|
| 325 |
-
return output_file_path, output_mask_path
|
| 326 |
-
|
| 327 |
-
# --- Image Preprocessing ---
|
| 328 |
-
# Define necessary MONAI transforms directly
|
| 329 |
-
# Keys must match the dictionary keys we create later ('image')
|
| 330 |
-
resize_transform = Resized(keys=["image"], spatial_size=image_size)
|
| 331 |
-
scale_transform = ScaleIntensityd(keys=["image"], minv=0.0, maxv=1.0)
|
| 332 |
-
|
| 333 |
-
def preprocess_nifti(nifti_path):
|
| 334 |
-
"""Loads and preprocesses a NIfTI file, returning a 5D tensor."""
|
| 335 |
-
print(f"Preprocessing NIfTI: {nifti_path}")
|
| 336 |
-
scan_data = nib.load(nifti_path).get_fdata()
|
| 337 |
-
print(f" Loaded scan data shape: {scan_data.shape}")
|
| 338 |
-
scan_tensor = torch.tensor(scan_data, dtype=torch.float32).unsqueeze(0) # Add C dim
|
| 339 |
-
print(f" Shape after tensor+channel: {scan_tensor.shape}")
|
| 340 |
-
sample = {"image": scan_tensor}
|
| 341 |
-
sample_resized = resize_transform(sample)
|
| 342 |
-
print(f" Shape after resize: {sample_resized['image'].shape}")
|
| 343 |
-
sample_scaled = scale_transform(sample_resized)
|
| 344 |
-
print(f" Shape after scaling: {sample_scaled['image'].shape}")
|
| 345 |
-
input_tensor = sample_scaled["image"].unsqueeze(0).to(device) # Add B dim
|
| 346 |
-
print(f" Final shape for model: {input_tensor.shape}")
|
| 347 |
-
if input_tensor.dim() != 5:
|
| 348 |
-
raise ValueError(f"Preprocessing resulted in incorrect shape: {input_tensor.shape}. Expected 5D.")
|
| 349 |
-
return input_tensor
|
| 350 |
-
|
| 351 |
-
# --- Final NIfTI Preprocessing for Model ---
|
| 352 |
-
def preprocess_nifti_for_model(nifti_path):
|
| 353 |
-
"""Loads final NIfTI and prepares 5D tensor for the model."""
|
| 354 |
-
# ... (Same as previous preprocess_nifti function) ...
|
| 355 |
-
print(f"Preprocessing NIfTI for model: {nifti_path}")
|
| 356 |
-
scan_data = nib.load(nifti_path).get_fdata()
|
| 357 |
-
print(f" Loaded scan data shape: {scan_data.shape}")
|
| 358 |
-
scan_tensor = torch.tensor(scan_data, dtype=torch.float32).unsqueeze(0) # Add C dim
|
| 359 |
-
print(f" Shape after tensor+channel: {scan_tensor.shape}")
|
| 360 |
-
sample = {"image": scan_tensor}
|
| 361 |
-
sample_resized = resize_transform(sample)
|
| 362 |
-
print(f" Shape after resize: {sample_resized['image'].shape}")
|
| 363 |
-
sample_scaled = scale_transform(sample_resized)
|
| 364 |
-
print(f" Shape after scaling: {sample_scaled['image'].shape}")
|
| 365 |
-
input_tensor = sample_scaled["image"].unsqueeze(0).to(device) # Add B dim
|
| 366 |
-
print(f" Final shape for model: {input_tensor.shape}")
|
| 367 |
-
if input_tensor.dim() != 5:
|
| 368 |
-
raise ValueError(f"Preprocessing resulted in incorrect shape: {input_tensor.shape}. Expected 5D.")
|
| 369 |
-
return input_tensor
|
| 370 |
-
|
| 371 |
-
# --- Saliency Map Generation ---
|
| 372 |
-
def generate_saliency(model, input_tensor_5d):
|
| 373 |
-
"""Generates saliency map using GuidedBackpropSmoothGrad."""
|
| 374 |
-
if GuidedBackpropSmoothGrad is None:
|
| 375 |
-
raise ImportError("MONAI visualize components not imported. Cannot generate saliency map.")
|
| 376 |
-
if model is None:
|
| 377 |
-
raise ValueError("Model not loaded. Cannot generate saliency map.")
|
| 378 |
-
|
| 379 |
-
print("Generating saliency map...")
|
| 380 |
-
input_tensor_5d.requires_grad_(True)
|
| 381 |
-
# Use the backbone for saliency as in the original script
|
| 382 |
-
# Ensure model and backbone are on the correct device (CPU in this case)
|
| 383 |
-
visualizer = GuidedBackpropSmoothGrad(model=model.backbone.to(device),
|
| 384 |
-
stdev_spread=0.15,
|
| 385 |
-
n_samples=10,
|
| 386 |
-
magnitude=True)
|
| 387 |
-
|
| 388 |
-
try:
|
| 389 |
-
with torch.enable_grad():
|
| 390 |
-
saliency_map_5d = visualizer(input_tensor_5d.to(device))
|
| 391 |
-
print("Saliency map generated.")
|
| 392 |
-
|
| 393 |
-
# Detach, move to CPU, remove Batch and Channel dims for processing/plotting -> (D, H, W)
|
| 394 |
-
input_3d = input_tensor_5d.squeeze().cpu().detach().numpy()
|
| 395 |
-
saliency_3d = saliency_map_5d.squeeze().cpu().detach().numpy()
|
| 396 |
-
|
| 397 |
-
return input_3d, saliency_3d
|
| 398 |
-
|
| 399 |
-
except Exception as e:
|
| 400 |
-
print(f"Error during saliency map generation: {e}")
|
| 401 |
-
traceback.print_exc()
|
| 402 |
-
# Return None or empty arrays if generation fails
|
| 403 |
-
return None, None
|
| 404 |
-
finally:
|
| 405 |
-
# Ensure requires_grad is turned off if it was modified
|
| 406 |
-
input_tensor_5d.requires_grad_(False)
|
| 407 |
-
|
| 408 |
-
# --- Plotting Function for Single Slice ---
|
| 409 |
-
def create_plot_images_for_slice(mri_data_3d, saliency_data_3d, slice_index):
|
| 410 |
-
"""Creates base64 encoded PNGs for a specific axial slice index."""
|
| 411 |
-
print(f" Generating plots for slice index: {slice_index}")
|
| 412 |
-
if mri_data_3d is None or saliency_data_3d is None:
|
| 413 |
-
print(" Input or Saliency data is None, cannot generate plot.")
|
| 414 |
-
return None
|
| 415 |
-
if slice_index < 0 or slice_index >= mri_data_3d.shape[2]:
|
| 416 |
-
print(f" Error: Slice index {slice_index} out of bounds (0-{mri_data_3d.shape[2]-1}).")
|
| 417 |
-
return None
|
| 418 |
-
|
| 419 |
-
# Function to save plot to base64 string (copied from previous version)
|
| 420 |
-
def save_plot_to_base64(fig):
|
| 421 |
-
buf = io.BytesIO()
|
| 422 |
-
fig.savefig(buf, format='png', bbox_inches='tight', pad_inches=0, dpi=75)
|
| 423 |
-
plt.close(fig) # Close the figure immediately
|
| 424 |
-
buf.seek(0)
|
| 425 |
-
img_str = base64.b64encode(buf.read()).decode('utf-8')
|
| 426 |
-
buf.close()
|
| 427 |
-
return img_str
|
| 428 |
-
|
| 429 |
-
try:
|
| 430 |
-
mri_slice = mri_data_3d[:, :, slice_index]
|
| 431 |
-
saliency_slice_orig = saliency_data_3d[:, :, slice_index]
|
| 432 |
-
|
| 433 |
-
# --- Normalize MRI Slice (using volume stats if available, otherwise slice stats) ---
|
| 434 |
-
# For consistency, ideally pass volume stats, but recalculating per slice is fallback
|
| 435 |
-
p1_vol, p99_vol = np.percentile(mri_data_3d, (1, 99))
|
| 436 |
-
mri_norm_denom = p99_vol - p1_vol
|
| 437 |
-
if mri_norm_denom < 1e-6: mri_norm_denom = 1e-6
|
| 438 |
-
mri_slice_norm = np.clip(mri_slice, p1_vol, p99_vol)
|
| 439 |
-
mri_slice_norm = (mri_slice_norm - p1_vol) / mri_norm_denom
|
| 440 |
-
|
| 441 |
-
# --- Process Saliency Slice ---
|
| 442 |
-
saliency_slice = np.copy(saliency_slice_orig)
|
| 443 |
-
saliency_slice[saliency_slice < 0] = 0 # Ensure non-negative
|
| 444 |
-
saliency_slice_blurred = cv2.GaussianBlur(saliency_slice, (15, 15), 0)
|
| 445 |
-
# Use volume max for normalization if possible, fallback to slice max
|
| 446 |
-
s_max_vol = np.max(saliency_data_3d[saliency_data_3d >= 0]) # Max of non-negative values in volume
|
| 447 |
-
if s_max_vol < 1e-6: s_max_vol = 1e-6
|
| 448 |
-
# --- Add logging for the calculated global max ---
|
| 449 |
-
print(f" Calculated Global Max Saliency (s_max_vol) for normalization: {s_max_vol:.4f}")
|
| 450 |
-
# --------------------------------------------------
|
| 451 |
-
saliency_slice_norm = saliency_slice_blurred / s_max_vol
|
| 452 |
-
threshold_value = 0.0
|
| 453 |
-
saliency_slice_thresholded = np.where(saliency_slice_norm > threshold_value, saliency_slice_norm, 0)
|
| 454 |
-
|
| 455 |
-
# --- Generate Plots ---
|
| 456 |
-
slice_plots = {}
|
| 457 |
-
|
| 458 |
-
# Plot 1: Input Slice
|
| 459 |
-
fig1, ax1 = plt.subplots(figsize=(3, 3))
|
| 460 |
-
ax1.imshow(mri_slice_norm, cmap='gray', interpolation='none', origin='lower')
|
| 461 |
-
ax1.axis('off')
|
| 462 |
-
slice_plots['input_slice'] = save_plot_to_base64(fig1)
|
| 463 |
-
|
| 464 |
-
# Plot 2: Saliency Heatmap
|
| 465 |
-
fig2, ax2 = plt.subplots(figsize=(3, 3))
|
| 466 |
-
ax2.imshow(saliency_slice_thresholded, cmap='magma', interpolation='none', origin='lower')
|
| 467 |
-
ax2.axis('off')
|
| 468 |
-
slice_plots['heatmap_slice'] = save_plot_to_base64(fig2)
|
| 469 |
-
|
| 470 |
-
# Plot 3: Overlay
|
| 471 |
-
fig3, ax3 = plt.subplots(figsize=(3, 3))
|
| 472 |
-
ax3.imshow(mri_slice_norm, cmap='gray', interpolation='none', origin='lower')
|
| 473 |
-
if np.max(saliency_slice_thresholded) > 0:
|
| 474 |
-
# Remove fixed levels to let contour auto-determine levels based on slice data
|
| 475 |
-
ax3.contour(saliency_slice_thresholded, cmap='magma', origin='lower', linewidths=1.0)
|
| 476 |
-
ax3.axis('off')
|
| 477 |
-
slice_plots['overlay_slice'] = save_plot_to_base64(fig3)
|
| 478 |
-
|
| 479 |
-
print(f" Generated plots successfully for slice {slice_index}.")
|
| 480 |
-
return slice_plots
|
| 481 |
-
|
| 482 |
-
except Exception as e:
|
| 483 |
-
print(f"Error generating plots for slice {slice_index}: {e}")
|
| 484 |
-
traceback.print_exc()
|
| 485 |
-
return None
|
| 486 |
-
|
| 487 |
-
# --- Flask Routes ---
|
| 488 |
-
@app.route('/', methods=['GET'])
|
| 489 |
-
def index():
|
| 490 |
-
return render_template('index.html')
|
| 491 |
-
|
| 492 |
-
@app.route('/predict', methods=['POST'])
|
| 493 |
-
def predict():
|
| 494 |
-
if model is None:
|
| 495 |
-
flash('Model not loaded. Cannot perform prediction.', 'error')
|
| 496 |
-
return redirect(url_for('index'))
|
| 497 |
-
|
| 498 |
-
# Get form data
|
| 499 |
-
file_type = request.form.get('file_type')
|
| 500 |
-
run_preprocess_flag = request.form.get('preprocess') == 'yes'
|
| 501 |
-
generate_saliency_flag = request.form.get('generate_saliency') == 'yes' # Get saliency flag
|
| 502 |
-
file = request.files.get('scan_file')
|
| 503 |
-
|
| 504 |
-
# --- Basic Input Validation ---
|
| 505 |
-
if not file_type:
|
| 506 |
-
flash('Please select a file type (NIfTI or DICOM).', 'error')
|
| 507 |
-
return redirect(url_for('index'))
|
| 508 |
-
if not file or file.filename == '':
|
| 509 |
-
flash('No scan file selected', 'error')
|
| 510 |
-
return redirect(url_for('index'))
|
| 511 |
-
|
| 512 |
-
print(f"Received upload: type='{file_type}', filename='{file.filename}', preprocess={run_preprocess_flag}, saliency={generate_saliency_flag}")
|
| 513 |
-
|
| 514 |
-
# --- Setup Temporary Directory ---
|
| 515 |
-
# temp_dir_obj = tempfile.TemporaryDirectory() # <--- PROBLEM: Cleans up automatically
|
| 516 |
-
# Use mkdtemp to create a persistent temporary directory
|
| 517 |
-
# NOTE: Requires a manual cleanup strategy later!
|
| 518 |
-
try:
|
| 519 |
-
temp_dir = tempfile.mkdtemp()
|
| 520 |
-
except Exception as e:
|
| 521 |
-
print(f"Error creating temporary directory: {e}")
|
| 522 |
-
flash("Server error: Could not create temporary directory.", "error")
|
| 523 |
-
return redirect(url_for('index'))
|
| 524 |
-
|
| 525 |
-
# Generate a unique ID based on the temp directory name
|
| 526 |
-
unique_id = os.path.basename(temp_dir)
|
| 527 |
-
print(f"Created persistent temp directory: {temp_dir} (ID: {unique_id})")
|
| 528 |
-
nifti_for_preprocessing_path = None # Path to the NIfTI before optional preprocessing
|
| 529 |
-
|
| 530 |
-
try:
|
| 531 |
-
# --- Handle Upload and DICOM Conversion ---
|
| 532 |
-
# --- Handle NIfTI Upload ---
|
| 533 |
-
if file_type == 'nifti':
|
| 534 |
-
if not file.filename.endswith('.nii.gz'):
|
| 535 |
-
flash('Invalid file type for NIfTI selection. Please upload .nii.gz', 'error')
|
| 536 |
-
# temp_dir_obj.cleanup() # No object to cleanup, need manual rmtree
|
| 537 |
-
shutil.rmtree(temp_dir, ignore_errors=True)
|
| 538 |
-
return redirect(url_for('index'))
|
| 539 |
-
uploaded_file_path = os.path.join(temp_dir, "uploaded_scan.nii.gz")
|
| 540 |
-
file.save(uploaded_file_path)
|
| 541 |
-
print(f"Saved uploaded NIfTI file to: {uploaded_file_path}")
|
| 542 |
-
nifti_for_preprocessing_path = uploaded_file_path
|
| 543 |
-
|
| 544 |
-
# --- Handle DICOM Upload ---
|
| 545 |
-
elif file_type == 'dicom':
|
| 546 |
-
if not file.filename.endswith('.zip'):
|
| 547 |
-
flash('Invalid file type for DICOM selection. Please upload a .zip file.', 'error')
|
| 548 |
-
# temp_dir_obj.cleanup()
|
| 549 |
-
shutil.rmtree(temp_dir, ignore_errors=True)
|
| 550 |
-
return redirect(url_for('index'))
|
| 551 |
-
uploaded_zip_path = os.path.join(temp_dir, "dicom_files.zip")
|
| 552 |
-
file.save(uploaded_zip_path)
|
| 553 |
-
print(f"Saved uploaded DICOM zip to: {uploaded_zip_path}")
|
| 554 |
-
dicom_input_dir = os.path.join(temp_dir, "dicom_input")
|
| 555 |
-
nifti_output_dir = os.path.join(temp_dir, "nifti_output")
|
| 556 |
-
os.makedirs(dicom_input_dir, exist_ok=True)
|
| 557 |
-
os.makedirs(nifti_output_dir, exist_ok=True)
|
| 558 |
-
try:
|
| 559 |
-
# Use shutil.unpack_archive for better cross-platform compatibility potentially
|
| 560 |
-
shutil.unpack_archive(uploaded_zip_path, dicom_input_dir)
|
| 561 |
-
print(f"Unzip successful.")
|
| 562 |
-
except Exception as e:
|
| 563 |
-
print(f"Unzip failed: {e}")
|
| 564 |
-
flash(f'Error unzipping DICOM file: {e}', 'error')
|
| 565 |
-
# temp_dir_obj.cleanup()
|
| 566 |
-
shutil.rmtree(temp_dir, ignore_errors=True)
|
| 567 |
-
return redirect(url_for('index'))
|
| 568 |
-
try:
|
| 569 |
-
dicom2nifti.convert_directory(dicom_input_dir, nifti_output_dir, compression=True, reorient=True)
|
| 570 |
-
nifti_files = [f for f in os.listdir(nifti_output_dir) if f.endswith('.nii.gz')]
|
| 571 |
-
if not nifti_files:
|
| 572 |
-
raise RuntimeError("dicom2nifti did not produce a .nii.gz file.")
|
| 573 |
-
nifti_for_preprocessing_path = os.path.join(nifti_output_dir, nifti_files[0])
|
| 574 |
-
print(f"DICOM conversion successful. NIfTI file: {nifti_for_preprocessing_path}")
|
| 575 |
-
except Exception as e:
|
| 576 |
-
print(f"DICOM to NIfTI conversion failed: {e}")
|
| 577 |
-
flash(f'Error converting DICOM to NIfTI: {e}', 'error')
|
| 578 |
-
# temp_dir_obj.cleanup()
|
| 579 |
-
shutil.rmtree(temp_dir, ignore_errors=True)
|
| 580 |
-
return redirect(url_for('index'))
|
| 581 |
-
else:
|
| 582 |
-
flash('Invalid file type selected.', 'error')
|
| 583 |
-
# temp_dir_obj.cleanup()
|
| 584 |
-
shutil.rmtree(temp_dir, ignore_errors=True)
|
| 585 |
-
return redirect(url_for('index'))
|
| 586 |
-
|
| 587 |
-
if not nifti_for_preprocessing_path or not os.path.exists(nifti_for_preprocessing_path):
|
| 588 |
-
flash('Error: Could not find the NIfTI file for processing.', 'error')
|
| 589 |
-
# temp_dir_obj.cleanup()
|
| 590 |
-
shutil.rmtree(temp_dir, ignore_errors=True)
|
| 591 |
-
return redirect(url_for('index'))
|
| 592 |
-
|
| 593 |
-
# --- Optional Preprocessing Steps ---
|
| 594 |
-
nifti_to_predict_path = nifti_for_preprocessing_path
|
| 595 |
-
if run_preprocess_flag:
|
| 596 |
-
print("--- Running Optional Preprocessing Pipeline ---")
|
| 597 |
-
try:
|
| 598 |
-
registered_path = os.path.join(temp_dir, "registered.nii.gz")
|
| 599 |
-
register_image(nifti_for_preprocessing_path, registered_path)
|
| 600 |
-
enhanced_path = os.path.join(temp_dir, "enhanced.nii.gz")
|
| 601 |
-
run_enhance_on_file(registered_path, enhanced_path)
|
| 602 |
-
skullstrip_output_dir = os.path.join(temp_dir, "skullstripped")
|
| 603 |
-
skullstripped_path, _ = run_skull_stripping(enhanced_path, skullstrip_output_dir)
|
| 604 |
-
nifti_to_predict_path = skullstripped_path
|
| 605 |
-
print("--- Optional Preprocessing Pipeline Complete ---")
|
| 606 |
-
except Exception as e:
|
| 607 |
-
print(f"Error during optional preprocessing pipeline: {e}")
|
| 608 |
-
traceback.print_exc()
|
| 609 |
-
flash(f'Error during preprocessing: {e}', 'error')
|
| 610 |
-
# temp_dir_obj.cleanup()
|
| 611 |
-
shutil.rmtree(temp_dir, ignore_errors=True)
|
| 612 |
-
return redirect(url_for('index'))
|
| 613 |
-
else:
|
| 614 |
-
print("--- Skipping Optional Preprocessing Pipeline ---")
|
| 615 |
-
|
| 616 |
-
# --- Final Preprocessing for Model & Prediction ---
|
| 617 |
-
input_tensor_5d = preprocess_nifti_for_model(nifti_to_predict_path)
|
| 618 |
-
print("Performing prediction...")
|
| 619 |
-
with torch.no_grad():
|
| 620 |
-
output = model(input_tensor_5d)
|
| 621 |
-
predicted_age = output.item()
|
| 622 |
-
predicted_age_years = predicted_age / 12 # Adjust if needed
|
| 623 |
-
print(f"Prediction successful: {predicted_age_years:.2f} years")
|
| 624 |
-
|
| 625 |
-
# --- Saliency Data Handling (Generate, Save, Get Initial Plot) ---
|
| 626 |
-
saliency_output_for_template = None # Initialize
|
| 627 |
-
if generate_saliency_flag:
|
| 628 |
-
print("--- Generating & Saving Saliency Data ---")
|
| 629 |
-
try:
|
| 630 |
-
input_3d_for_plot, saliency_3d = generate_saliency(model, input_tensor_5d)
|
| 631 |
-
|
| 632 |
-
if input_3d_for_plot is not None and saliency_3d is not None:
|
| 633 |
-
num_slices = input_3d_for_plot.shape[2]
|
| 634 |
-
center_slice_index = num_slices // 2
|
| 635 |
-
|
| 636 |
-
# Save the numpy arrays for the dynamic route
|
| 637 |
-
input_array_path = os.path.join(temp_dir, f"{unique_id}_input.npy")
|
| 638 |
-
saliency_array_path = os.path.join(temp_dir, f"{unique_id}_saliency.npy")
|
| 639 |
-
np.save(input_array_path, input_3d_for_plot)
|
| 640 |
-
np.save(saliency_array_path, saliency_3d)
|
| 641 |
-
print(f"Saved input array to {input_array_path}")
|
| 642 |
-
print(f"Saved saliency array to {saliency_array_path}")
|
| 643 |
-
|
| 644 |
-
# Generate ONLY the center slice plots for the initial view
|
| 645 |
-
center_slice_plots = create_plot_images_for_slice(input_3d_for_plot, saliency_3d, center_slice_index)
|
| 646 |
-
|
| 647 |
-
if center_slice_plots:
|
| 648 |
-
# Prepare data structure for the template
|
| 649 |
-
saliency_output_for_template = {
|
| 650 |
-
'center_slice_plots': center_slice_plots,
|
| 651 |
-
'num_slices': num_slices,
|
| 652 |
-
'center_slice_index': center_slice_index,
|
| 653 |
-
'unique_id': unique_id, # Pass the ID for filenames
|
| 654 |
-
'temp_dir_path': temp_dir # Pass the full path for lookup
|
| 655 |
-
}
|
| 656 |
-
print("--- Saliency Data Saved & Initial Plot Generated ---")
|
| 657 |
-
else:
|
| 658 |
-
print("--- Center Slice Plotting Failed ---")
|
| 659 |
-
flash('Failed to generate initial saliency plot.', 'warning')
|
| 660 |
-
else:
|
| 661 |
-
print("--- Saliency Generation Failed --- ")
|
| 662 |
-
flash('Saliency map generation failed.', 'warning')
|
| 663 |
-
|
| 664 |
-
except Exception as e:
|
| 665 |
-
print(f"Error during saliency processing/saving: {e}")
|
| 666 |
-
traceback.print_exc()
|
| 667 |
-
flash('Could not generate or save saliency maps due to an error.', 'error')
|
| 668 |
-
|
| 669 |
-
# Render result, passing prediction and potentially the NEW saliency structure
|
| 670 |
-
return render_template('index.html',
|
| 671 |
-
prediction=f"{predicted_age_years:.2f} years",
|
| 672 |
-
saliency_info=saliency_output_for_template) # Pass the new dict
|
| 673 |
-
|
| 674 |
-
except Exception as e:
|
| 675 |
-
flash(f'Error processing file: {e}', 'error')
|
| 676 |
-
print(f"Caught Exception during prediction process: {e}")
|
| 677 |
-
traceback.print_exc()
|
| 678 |
-
# Ensure cleanup happens even if exception occurs mid-process
|
| 679 |
-
# temp_dir_obj.cleanup()
|
| 680 |
-
if temp_dir and os.path.exists(temp_dir):
|
| 681 |
-
shutil.rmtree(temp_dir, ignore_errors=True) # Manual cleanup on general error
|
| 682 |
-
return redirect(url_for('index'))
|
| 683 |
-
|
| 684 |
-
# NOTE: Temporary directory created with mkdtemp is NOT automatically cleaned.
|
| 685 |
-
# Need a separate mechanism (e.g., cron job, background task) to remove old directories
|
| 686 |
-
# from the system's temporary location (e.g., /tmp) based on age.
|
| 687 |
-
# Leaving the directory here so /get_slice can access the files.
|
| 688 |
-
|
| 689 |
-
# --- New Route for Dynamic Slice Loading ---
|
| 690 |
-
@app.route('/get_slice/<unique_id>/<int:slice_index>')
|
| 691 |
-
def get_slice(unique_id, slice_index):
|
| 692 |
-
# Get the actual temporary directory path from query parameter
|
| 693 |
-
temp_dir_path = request.args.get('path')
|
| 694 |
-
if not temp_dir_path:
|
| 695 |
-
print("Error: 'path' query parameter missing in /get_slice request")
|
| 696 |
-
return jsonify({"error": "Required path information missing."}), 400
|
| 697 |
-
|
| 698 |
-
# Construct paths using the provided directory path and unique ID
|
| 699 |
-
input_array_path = os.path.join(temp_dir_path, f"{unique_id}_input.npy")
|
| 700 |
-
saliency_array_path = os.path.join(temp_dir_path, f"{unique_id}_saliency.npy")
|
| 701 |
-
print(f"Attempting to load slice {slice_index} for ID {unique_id} from actual path: {temp_dir_path}")
|
| 702 |
-
|
| 703 |
-
try:
|
| 704 |
-
# Check using the exact paths constructed above
|
| 705 |
-
if not os.path.exists(input_array_path) or not os.path.exists(saliency_array_path):
|
| 706 |
-
print(f"Error: .npy files not found for ID {unique_id} at {temp_dir_path}")
|
| 707 |
-
return jsonify({"error": "Saliency data not found. It might have expired or failed to save."}), 404
|
| 708 |
-
|
| 709 |
-
input_3d = np.load(input_array_path)
|
| 710 |
-
saliency_3d = np.load(saliency_array_path)
|
| 711 |
-
print(f"Loaded arrays for ID {unique_id}. Input shape: {input_3d.shape}, Saliency shape: {saliency_3d.shape}")
|
| 712 |
-
|
| 713 |
-
# Generate plots for the requested slice using the helper function
|
| 714 |
-
slice_plots = create_plot_images_for_slice(input_3d, saliency_3d, slice_index)
|
| 715 |
-
|
| 716 |
-
if slice_plots:
|
| 717 |
-
return jsonify(slice_plots) # Return plot data as JSON
|
| 718 |
-
else:
|
| 719 |
-
return jsonify({"error": f"Failed to generate plots for slice {slice_index}."}), 500
|
| 720 |
-
|
| 721 |
-
except Exception as e:
|
| 722 |
-
print(f"Error in /get_slice for ID {unique_id}, slice {slice_index}: {e}")
|
| 723 |
-
traceback.print_exc()
|
| 724 |
-
return jsonify({"error": "An internal error occurred while fetching the slice data."}), 500
|
| 725 |
-
|
| 726 |
-
if __name__ == '__main__':
|
| 727 |
-
# Use '0.0.0.0' to make it accessible outside the container
|
| 728 |
-
app.run(host='0.0.0.0', port=5000, debug=False) # Turn off debug for production/docker
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|