diff --git a/setup.py b/setup.py
index 976c88682..9b0e90f34 100644
--- a/setup.py
+++ b/setup.py
@@ -21,7 +21,7 @@
             'p_tqdm',
             'xvfbwrapper',
             'fury',
-            'nnunetv2==2.1',
+            'nnunetv2>=2.2.1',
             'requests==2.27.1;python_version<"3.10"',
             'requests;python_version>="3.10"',
             'rt_utils',
diff --git a/tests/test_locally.py b/tests/test_locally.py
index 8f1b34bf5..e27f812c6 100755
--- a/tests/test_locally.py
+++ b/tests/test_locally.py
@@ -1,5 +1,7 @@
 import sys
 from pathlib import Path
+p_dir = str(Path(__file__).absolute().parents[1])
+if p_dir not in sys.path: sys.path.insert(0, p_dir)
 
 import pytest
 import os
@@ -7,16 +9,18 @@
 import glob
 import shutil
 import subprocess
-from tqdm import tqdm
-import pandas as pd
-import numpy as np
 from collections import defaultdict
-import psutil
 import time
 import threading
-import torch
 import platform
 
+import psutil
+from tqdm import tqdm
+import pandas as pd
+import numpy as np
+import torch
+import nnunetv2
+
 from totalsegmentator.python_api import totalsegmentator
 from totalsegmentator.map_to_binary import class_map
 from totalsegmentator.excel import set_xlsx_column_width_to_content
@@ -222,7 +226,8 @@ def are_logs_similar(last_log, new_log, cols, tolerance_percent=0.04):
             "memory_gpu_15mm", "memory_gpu_3mm",
             "cpu_utilization_15mm", "cpu_utilization_3mm",
             "gpu_utilization_15mm", "gpu_utilization_3mm",
-            "python_version", "torch_version", "cuda_version", "cudnn_version",
+            "python_version", "torch_version", "nnunet_version", 
+            "cuda_version", "cudnn_version",
             "gpu_name"]
     overview_file = Path(f"{base_dir}/overview.xlsx")
     if overview_file.exists():
@@ -240,6 +245,7 @@ def are_logs_similar(last_log, new_log, cols, tolerance_percent=0.04):
                cpu_utilization["15mm"], cpu_utilization["3mm"],
                gpu_utilization["15mm"], gpu_utilization["3mm"],
                platform.python_version(), torch.__version__,
+               nnunetv2.__version__,
                float(torch.version.cuda), int(torch.backends.cudnn.version()),
                torch.cuda.get_device_name(0)]
 
diff --git a/totalsegmentator/nnunet.py b/totalsegmentator/nnunet.py
index 72d2a4d0d..1f2ad0790 100644
--- a/totalsegmentator/nnunet.py
+++ b/totalsegmentator/nnunet.py
@@ -22,10 +22,10 @@
 from totalsegmentator.libs import nostdout
 
 # nnUNet 2.1
-with nostdout():
-    from nnunetv2.inference.predict_from_raw_data import predict_from_raw_data
+# with nostdout():
+#     from nnunetv2.inference.predict_from_raw_data import predict_from_raw_data
 # nnUNet 2.2
-# from nnunetv2.inference.predict_from_raw_data import nnUNetPredictor
+from nnunetv2.inference.predict_from_raw_data import nnUNetPredictor
 
 from nnunetv2.utilities.file_path_utilities import get_output_folder
 
@@ -178,46 +178,46 @@ def nnUNetv2_predict(dir_in, dir_out, task_id, model="3d_fullres", folds=None,
     allow_tqdm = not quiet
 
     # nnUNet 2.1
-    predict_from_raw_data(dir_in,
-                          dir_out,
-                          model_folder,
-                          folds,
-                          step_size,
-                          use_gaussian=True,
-                          use_mirroring=not disable_tta,
-                          perform_everything_on_gpu=True,
-                          verbose=verbose,
-                          save_probabilities=save_probabilities,
-                          overwrite=not continue_prediction,
-                          checkpoint_name=chk,
-                          num_processes_preprocessing=npp,
-                          num_processes_segmentation_export=nps,
-                          folder_with_segs_from_prev_stage=prev_stage_predictions,
-                          num_parts=num_parts,
-                          part_id=part_id,
-                          device=device)
+    # predict_from_raw_data(dir_in,
+    #                       dir_out,
+    #                       model_folder,
+    #                       folds,
+    #                       step_size,
+    #                       use_gaussian=True,
+    #                       use_mirroring=not disable_tta,
+    #                       perform_everything_on_gpu=True,
+    #                       verbose=verbose,
+    #                       save_probabilities=save_probabilities,
+    #                       overwrite=not continue_prediction,
+    #                       checkpoint_name=chk,
+    #                       num_processes_preprocessing=npp,
+    #                       num_processes_segmentation_export=nps,
+    #                       folder_with_segs_from_prev_stage=prev_stage_predictions,
+    #                       num_parts=num_parts,
+    #                       part_id=part_id,
+    #                       device=device)
 
     # nnUNet 2.2
-    # predictor = nnUNetPredictor(
-    #     tile_step_size=step_size,
-    #     use_gaussian=True,
-    #     use_mirroring=not disable_tta,
-    #     perform_everything_on_gpu=True,
-    #     device=device,
-    #     verbose=verbose,
-    #     verbose_preprocessing=verbose,
-    #     allow_tqdm=allow_tqdm
-    # )
-    # predictor.initialize_from_trained_model_folder(
-    #     model_folder,
-    #     use_folds=folds,
-    #     checkpoint_name=chk,
-    # )
-    # predictor.predict_from_files(dir_in, dir_out,
-    #                              save_probabilities=save_probabilities, overwrite=not continue_prediction,
-    #                              num_processes_preprocessing=npp, num_processes_segmentation_export=nps,
-    #                              folder_with_segs_from_prev_stage=prev_stage_predictions,
-    #                              num_parts=num_parts, part_id=part_id)
+    predictor = nnUNetPredictor(
+        tile_step_size=step_size,
+        use_gaussian=True,
+        use_mirroring=not disable_tta,
+        perform_everything_on_gpu=True,
+        device=device,
+        verbose=verbose,
+        verbose_preprocessing=verbose,
+        allow_tqdm=allow_tqdm
+    )
+    predictor.initialize_from_trained_model_folder(
+        model_folder,
+        use_folds=folds,
+        checkpoint_name=chk,
+    )
+    predictor.predict_from_files(dir_in, dir_out,
+                                 save_probabilities=save_probabilities, overwrite=not continue_prediction,
+                                 num_processes_preprocessing=npp, num_processes_segmentation_export=nps,
+                                 folder_with_segs_from_prev_stage=prev_stage_predictions,
+                                 num_parts=num_parts, part_id=part_id)
 
 
 def save_segmentation_nifti(class_map_item, tmp_dir=None, file_out=None, nora_tag=None, header=None, task_name=None, quiet=None):