Update Fastmri reader for inference and add support of Tensor for switch_endianness (#5548)

Fixes #5552 .

### Description

A few sentences describing the changes proposed in this pull request.
Update Fastmri reader for inference

### Types of changes
<!--- Put an `x` in all the boxes that apply, and remove the not
applicable items -->
- [x] Non-breaking change (fix or new feature that would not break
existing functionality).
- [ ] Breaking change (fix or new feature that would cause existing
functionality to change).
- [ ] New tests added to cover the changes.
- [ ] Integration tests passed locally by running `./runtests.sh -f -u
--net --coverage`.
- [ ] Quick tests passed locally by running `./runtests.sh --quick
--unittests --disttests`.
- [ ] In-line docstrings updated.
- [ ] Documentation updated, tested `make html` command in the `docs/`
folder.

Signed-off-by: Can Zhao <canz@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: Can-Zhao

monai/apps/reconstruction/fastmri_reader.py

@@ -19,7 +19,6 @@
 from monai.data.image_reader import ImageReader
 from monai.data.utils import is_supported_format
 from monai.utils import FastMRIKeys, optional_import, require_pkg
-from monai.utils.type_conversion import convert_to_tensor
 
 h5py, has_h5py = optional_import("h5py")
 
@@ -85,7 +84,7 @@ def get_data(self, dat: Dict) -> Tuple[ndarray, dict]:
         header = self._get_meta_dict(dat)
         data: ndarray = np.array(dat[FastMRIKeys.KSPACE])
         header[FastMRIKeys.MASK] = (
-            convert_to_tensor(np.array(dat[FastMRIKeys.MASK])).unsqueeze(0)[None, ..., None]
+            np.expand_dims(np.array(dat[FastMRIKeys.MASK]), 0)[None, ..., None]
             if FastMRIKeys.MASK in dat.keys()
             else np.zeros(data.shape)
         )

monai/transforms/io/array.py

@@ -68,7 +68,15 @@ def switch_endianness(data, new="<"):
         data: input to be converted.
         new: the target endianness, currently support "<" or ">".
     """
-    if isinstance(data, np.ndarray):
+    if isinstance(data, torch.Tensor):
+        device = data.device
+        requires_grad: bool = data.requires_grad
+        data = (
+            torch.from_numpy(switch_endianness(data.cpu().detach().numpy(), new))
+            .to(device)
+            .requires_grad_(requires_grad=requires_grad)
+        )
+    elif isinstance(data, np.ndarray):
         # default to system endian
         sys_native = "<" if (sys.byteorder == "little") else ">"
         current_ = sys_native if data.dtype.byteorder not in ("<", ">") else data.dtype.byteorder