GSOC

coca_project/src/COCA_metadata_compiler.py

@@ -0,0 +1,94 @@
+import json
+import pandas as pd
+import SimpleITK as sitk
+from pathlib import Path
+from tqdm import tqdm
+import numpy as np
+
+class COCAMetadataCompiler:
+    def __init__(self, project_root: str):
+        self.project_root = Path(project_root)
+        self.resampled_dir = self.project_root / "data_resampled"
+        self.canonical_dir = self.project_root / "data_canonical" / "images"
+        self.output_path = self.project_root / "data_canonical" / "tables" / "metadata_summary.parquet"
+
+        # Ensure table directory exists
+        self.output_path.parent.mkdir(parents=True, exist_ok=True)
+
+    def run(self):
+        all_meta = []
+
+        # We iterate through the resampled folder because these are the files 
+        # your model will actually use for training.
+        scan_folders = [f for f in self.resampled_dir.iterdir() if f.is_dir()]
+        print(f"Compiling metadata for {len(scan_folders)} resampled scans...")
+
+        for scan_folder in tqdm(scan_folders):
+            scan_id = scan_folder.name
+
+            # 1. Define Paths
+            img_path = scan_folder / f"{scan_id}_img.nii.gz"
+            seg_path = scan_folder / f"{scan_id}_seg.nii.gz"
+            json_path = self.canonical_dir / scan_id / f"{scan_id}_meta.json"
+
+            if not img_path.exists() or not seg_path.exists():
+                continue
+
+            try:
+                # 2. Load the actual resampled mask to get accurate counts
+                # This fixes the 'Voxel sum mismatch' error
+                res_seg = sitk.ReadImage(str(seg_path))
+                res_seg_array = sitk.GetArrayFromImage(res_seg)
+
+                # Calculate counts based on the CURRENT file state
+                current_voxels = int(np.sum(res_seg_array > 0))
+                current_slices = np.where(np.any(res_seg_array > 0, axis=(1, 2)))[0].tolist()
+
+                # Get geometry info
+                spacing = res_seg.GetSpacing()
+                shape = res_seg.GetSize()
+
+                # 3. Pull patient info from the original JSON if it exists
+                patient_id = "Unknown"
+                if json_path.exists():
+                    with open(json_path, 'r') as f:
+                        meta_data = json.load(f)
+                        patient_id = meta_data.get("patient_id", "Unknown")
+
+                # 4. Build the Row
+                row = {
+                    "scan_id": scan_id,
+                    "patient_id": patient_id,
+                    "image_path": str(img_path.resolve()),
+                    "mask_path": str(seg_path.resolve()),
+                    "has_pos": 1 if current_voxels > 0 else 0,
+                    "n_pos_voxels": current_voxels,
+                    "n_pos_slices": len(current_slices),
+                    "spacing_x": spacing[0],
+                    "spacing_y": spacing[1],
+                    "spacing_z": spacing[2],
+                    "shape_x": shape[0],
+                    "shape_y": shape[1],
+                    "shape_z": shape[2]
+                }
+                all_meta.append(row)
+
+            except Exception as e:
+                print(f"  [ERROR] Processing {scan_id}: {e}")
+
+        if not all_meta:
+            print("No data found to compile.")
+            return
+
+        # Create DataFrame and Export
+        df = pd.DataFrame(all_meta)
+        df.to_parquet(self.output_path, index=False, compression='snappy')
+
+        print("-" * 50)
+        print(f"SUCCESS: Metadata compiled at {self.output_path}")
+        print(f"Positive cases: {df['has_pos'].sum()} / {len(df)}")
+        print("-" * 50)
+
+if __name__ == "__main__":
+    compiler = COCAMetadataCompiler(r"C:\coca_project")
+    compiler.run()