feat: Introduce nnUNet inference pipeline with DICOM conversion, pretrained model installation, and configuration, updating gitignore to exclude archives.

.gitignore

@@ -162,3 +162,5 @@ cython_debug/
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 #.idea/
 
+*.pth
+*.zip

src/2508.yaml

@@ -0,0 +1,5 @@
+dataset_name_or_id:  2508                    # models trained on NTUH database
+# dataset_name_or_id:  3828                    # models trained on BraTS + NTUH
+nnunet_preprocessed: "./nnUNet/preprocessed" # directory for storing pre-processed data (optional)
+nnunet_raw:          "./nnUNet/raw/"          # directory for storing formated raw data (optional)
+nnunet_results:      "./nnUNet/results"      # directory for storing trained model checkpoints (optional)

src/2602.yaml

@@ -0,0 +1,4 @@
+dataset_name_or_id:  2602                    # models trained on BraTS + NTUH
+nnunet_preprocessed: "./nnUNet/preprocessed" # directory for storing pre-processed data (optional)
+nnunet_raw:          "./nnUNet/raw/"          # directory for storing formated raw data (optional)
+nnunet_results:      "./nnUNet/results"      # directory for storing trained model checkpoints (optional)

src/3828.yaml

@@ -0,0 +1,5 @@
+# dataset_name_or_id:  2508                    # models trained on NTUH database
+dataset_name_or_id:  3828                    # models trained on BraTS + NTUH
+nnunet_preprocessed: "./nnUNet/preprocessed" # directory for storing pre-processed data (optional)
+nnunet_raw:          "./nnUNet/raw/"          # directory for storing formated raw data (optional)
+nnunet_results:      "./nnUNet/results"      # directory for storing trained model checkpoints (optional)

src/check_dicom.py

@@ -0,0 +1,26 @@
+import pydicom
+import os
+import sys
+
+filepath = "/mnt/nextcloud/2026/計畫/TSHA_腦腫瘤驗證標記資料/Brain Tumor Label example/1.2.826.0.1.3680043.8.498.16118352232694493510220170548312112804.dcm"
+
+try:
+    ds = pydicom.dcmread(filepath)
+    print(f"Transfer Syntax: {ds.file_meta.TransferSyntaxUID}")
+    
+    try:
+        arr = ds.pixel_array
+        print("Successfully accessed pixel_array")
+        
+        # Try converting to uncompressed
+        ds.decompress()
+        print("Successfully decompressed")
+        
+    except Exception as e:
+        print(f"Error accessing pixel_array or decompressing: {e}")
+        # Check for handlers
+        import pydicom.uids
+        print(f"Available handlers for this syntax: {pydicom.config.pixel_data_handlers}")
+
+except Exception as e:
+    print(f"Error reading file: {e}")

src/inference.py

@@ -0,0 +1,110 @@
+import os
+import shutil
+
+from monai.apps.nnunet import nnUNetV2Runner
+
+import subprocess
+
+DATASET="Dataset2508_CKTarget"
+YAML   ="2508.yaml"
+
+DATASET="Dataset2602_BraTS-CK"
+YAML   ="2602.yaml"
+
+DATA_DIR='/mnt/nextcloud/2026/計畫/TSHA_腦腫瘤驗證標記資料/Brain Tumor Label example'
+
+imagesTs      = f'./nnUNet/raw/{DATASET}/imagesTs/'
+postprocessed = f'./nnUNet/results/{DATASET}/ensemble_predictions_postprocessed'
+
+OUTPUT_DIR = '/mnt/b4/Public/0'
+OUTPUT_DIR = '/mnt/nextcloud/2026/計畫/TSHA_腦腫瘤驗證標記資料/標記資料'
+
+def main():
+    if not os.path.exists(DATA_DIR):
+        print(f"Error: DATA_DIR {DATA_DIR} does not exist.")
+    else:
+        if os.path.exists(imagesTs):
+            shutil.rmtree(imagesTs)
+        os.makedirs(imagesTs)
+        
+        print(f"Converting DICOM from {DATA_DIR} to NIfTI in {imagesTs}...")
+        try:
+            # dicom2nifti.convert_directory(DATA_DIR, imagesTs, compression=True, reorient=True)
+            cmd = ["dcm2niix", "-o", imagesTs, "-f", "%n", "-z", "y", DATA_DIR]
+            print(' '.join(cmd))
+            subprocess.run(cmd, check=True)
+            print("Conversion completed successfully.")
+        except subprocess.CalledProcessError:
+            print("dcm2niix failed, likely due to compression. Attempting decompression with gdcmconv...")
+            
+            # Create a temporary directory for decompressed files
+            temp_dir = os.path.join(imagesTs, "temp_decompressed")
+            if os.path.exists(temp_dir):
+                shutil.rmtree(temp_dir)
+            os.makedirs(temp_dir)
+            
+            try:
+                # Decompress Dicom files
+                # Check for gdcmconv
+                if shutil.which("gdcmconv") is None:
+                     raise FileNotFoundError("gdcmconv not found. Please install gdcm-tools (e.g. conda install -c conda-forge gdcm).")
+
+                files_processed = 0
+                for fname in os.listdir(DATA_DIR):
+                    fpath = os.path.join(DATA_DIR, fname)
+                    if os.path.isfile(fpath):
+                        # Attempt validation/decompression
+                        subprocess.run(["gdcmconv", "-w", fpath, os.path.join(temp_dir, fname)], check=False)
+                        files_processed += 1
+                
+                if files_processed == 0:
+                     raise Exception("No files found to decompress.")
+
+                print(f"Decompressed {files_processed} files to {temp_dir}")
+
+                # Retry conversion
+                cmd = ["dcm2niix", "-o", imagesTs, "-f", "%n", "-z", "y", temp_dir]
+                print(' '.join(cmd))
+                subprocess.run(cmd, check=True)
+                print("Conversion completed successfully (after decompression).")
+            
+            except Exception as e:
+                print(f"Fallback conversion failed: {e}")
+            finally:
+                # Cleanup
+                if os.path.exists(temp_dir):
+                    shutil.rmtree(temp_dir)
+        except Exception as e:
+            print(f"Conversion failed: {e}")
+
+    INPUT_LIST = []
+
+    for f in sorted(os.scandir(imagesTs), key=lambda x: x.name):
+        if f.name.endswith(".nii.gz"):
+            if not f.name.endswith("_0000.nii.gz"):
+                new_path = f.path.replace(".nii.gz", "_0000.nii.gz")
+                INPUT_LIST.append(new_path)
+                os.rename(f.path, new_path)
+                print(f.path, new_path)
+
+    runner = nnUNetV2Runner(YAML)
+
+    runner.predict_ensemble_postprocessing()
+
+    for p in INPUT_LIST:
+        basename = os.path.basename(p).replace("_0000.nii.gz", ".nii.gz")
+        preprocessed = os.path.join(f'./nnUNet/results/{DATASET}/ensemble_predictions_postprocessed/', basename)
+        output_file = os.path.join(OUTPUT_DIR, basename)
+        shutil.move(preprocessed, output_file)
+        print(preprocessed, output_file)
+
+
+    # clean up
+    shutil.rmtree(f'./nnUNet/results/{DATASET}/pred_2d', ignore_errors=True)
+    shutil.rmtree(f'./nnUNet/results/{DATASET}/pred_3d_fullres', ignore_errors=True)
+    shutil.rmtree(f'./nnUNet/results/{DATASET}/pred_3d_lowres', ignore_errors=True)
+    shutil.rmtree(f'./nnUNet/results/{DATASET}/ensemble_predictions')
+    shutil.rmtree(f'./nnUNet/results/{DATASET}/ensemble_predictions_postprocessed')
+
+if __name__ == "__main__":
+    main()

src/input.yaml

@@ -0,0 +1,5 @@
+# dataset_name_or_id:  2508                    # models trained on NTUH database
+dataset_name_or_id:  3828                    # models trained on BraTS + NTUH
+nnunet_preprocessed: "./nnUNet/preprocessed" # directory for storing pre-processed data (optional)
+nnunet_raw:          "./nnUNet/raw"          # directory for storing formated raw data (optional)
+nnunet_results:      "./nnUNet/results"      # directory for storing trained model checkpoints (optional)

src/install_model.sh

@@ -0,0 +1,7 @@
+SCRIPT_PATH="$(dirname "$(readlink -f "$0")")"
+
+export nnUNet_preprocessed=./nnUNet/preprocessed
+export nnUNet_results=./nnUNet/results
+export nnUNet_raw=./nnUNet/rawy
+#nnUNetv2_install_pretrained_model_from_zip $SCRIPT_PATH/3828.zip
+nnUNetv2_install_pretrained_model_from_zip $SCRIPT_PATH/2602.zip

src/nnUNet

@@ -0,0 +1 @@
+/mnt/b4/Public/nnUNet/