Fix: cleanup utils import/export and robust channel-last handling in DiceHelper (refs #8366)

Signed-off-by: 林旻佑 <[email protected]>

Author: 林旻佑

monai/inferers/utils.py

@@ -38,9 +38,6 @@
 
 __all__ = ["sliding_window_inference"]
 
-dfrom typing import Optional, Tuple
-import torch
-
 def ensure_channel_first(
     x: torch.Tensor,
     spatial_ndim: Optional[int] = None,
@@ -105,6 +102,8 @@ def ensure_channel_first(
     raise ValueError(
         f"cannot infer channel dim for shape={tuple(x.shape)}; expected [N,C,spatial...] or [N,spatial...,C]"
     )
+    
+
 
 def sliding_window_inference(
     inputs: torch.Tensor | MetaTensor,