FIX: Avoid diverting dtype from original

Author: mgxd

niworkflows/interfaces/cifti.py

@@ -330,6 +330,7 @@ def _create_cifti_image(
         BOLD data saved as CIFTI dtseries
     """
     bold_img = nb.load(bold_file)
+    bold_dtype = bold_img.get_data_dtype()
     label_img = nb.load(label_file)
     if label_img.shape != bold_img.shape[:3]:
         warnings.warn("Resampling bold volume to match label dimensions")
@@ -339,7 +340,7 @@ def _create_cifti_image(
     bold_img = _reorient_image(bold_img, orientation="LAS")
     label_img = _reorient_image(label_img, orientation="LAS")
 
-    bold_data = bold_img.get_fdata(dtype="float32")
+    bold_data = bold_img.get_fdata()
     timepoints = bold_img.shape[3]
     label_data = np.asanyarray(label_img.dataobj).astype("int16")
 
@@ -355,7 +356,7 @@ def _create_cifti_image(
             hemi = structure.split("_")[-1]
             # currently only supports L/R cortex
             surf = nb.load(bold_surfs[hemi == "RIGHT"])
-            surf_verts = len(surf.darrays[0].data)
+            surf_verts = len(surf.agg_data()[0])
             if annotation_files[0].endswith(".annot"):
                 annot = nb.freesurfer.read_annot(annotation_files[hemi == "RIGHT"])
                 # remove medial wall
@@ -391,7 +392,7 @@ def _create_cifti_image(
                     else np.concatenate((ts, bold_data[ijk]))
                 )
                 vox += [
-                    [ijk[0][ix], ijk[1][ix], ijk[2][ix]] for ix, row in enumerate(ts)
+                    [ijk[0][idx], ijk[1][idx], ijk[2][idx]] for idx in range(len(ts))
                 ]
 
             vox = ci.Cifti2VoxelIndicesIJK(vox)
@@ -403,7 +404,7 @@ def _create_cifti_image(
                 voxel_indices_ijk=vox,
             )
             idx_offset += len(vox)
-            bm_ts = np.column_stack((bm_ts, ts.T))
+            bm_ts = np.column_stack((bm_ts, ts.T)).astype(bold_dtype)
         # add each brain structure to list
         brainmodels.append(bm)
 
@@ -439,7 +440,7 @@ def _create_cifti_image(
     matrix.append(geometry_map)
     matrix.metadata = ci.Cifti2MetaData(meta)
     hdr = ci.Cifti2Header(matrix)
-    img = ci.Cifti2Image(bm_ts, hdr)
+    img = ci.Cifti2Image(dataobj=bm_ts, header=hdr, nifti_header=bold_img.header)
     img.nifti_header.set_intent("NIFTI_INTENT_CONNECTIVITY_DENSE_SERIES")
 
     out_file = "{}.dtseries.nii".format(split_filename(bold_file)[1])