added two tutorials

fireants/registration/greedy.py

@@ -26,7 +26,7 @@ def __init__(self, scales: List[int], iterations: List[float],
                 fixed_images: BatchedImages, moving_images: BatchedImages,
                 loss_type: str = "cc",
                 deformation_type: str = 'geodesic',
-                optimizer: str = 'SGD', optimizer_params: dict = {},
+                optimizer: str = 'Adam', optimizer_params: dict = {},
                 optimizer_lr: float = 0.1, 
                 integrator_n: Union[str, int] = 6,
                 mi_kernel_type: str = 'b-spline', cc_kernel_type: str = 'rectangular',

fireants/scripts/tune_empire10.py

@@ -46,10 +46,10 @@ def forward(self, moved_seg, fixed_seg, *args, **kwargs):
         den = moved_flat.sum(1) + fixed_flat.sum(1)
         return 1 - (num/den).mean()
 
-def get_image_paths(brain_id, masks=False):
+def get_image_paths(lung_id, masks=False):
     PATH = SCANS_PATH if not masks else MASKS_PATH
-    fixed = osp.join(PATH, "{:02d}_Fixed.mhd".format(brain_id))
-    moving = osp.join(PATH, "{:02d}_Moving.mhd".format(brain_id))
+    fixed = osp.join(PATH, "{:02d}_Fixed.mhd".format(lung_id))
+    moving = osp.join(PATH, "{:02d}_Moving.mhd".format(lung_id))
     return fixed, moving
 
 def augment_image_with_mask(image, mask):
@@ -82,11 +82,11 @@ def register_dataset(config):
     }
     count = 0
 
-    for brain_id in range(1, 31):
+    for lung_id in range(1, 31):
         torch.cuda.empty_cache()
         time.sleep(2)
-        fixed_img_path, moving_img_path = get_image_paths(brain_id)
-        fixed_seg_path, moving_seg_path = get_image_paths(brain_id, masks=True)
+        fixed_img_path, moving_img_path = get_image_paths(lung_id)
+        fixed_seg_path, moving_seg_path = get_image_paths(lung_id, masks=True)
         # load fixed images
         fixed_image, fixed_seg = Image.load_file(fixed_img_path), Image.load_file(fixed_seg_path, is_segmentation=True)
         fixed_image = augment_image_with_mask(fixed_image, fixed_seg)
@@ -128,7 +128,7 @@ def register_dataset(config):
         deformable.optimize(save_transformed=False)
         def_end = time.time()
 
-        print(f"For scan pair: {brain_id}")
+        print(f"For scan pair: {lung_id}")
         print(f"Affine time: {aff_end - aff_start}")
         print(f"Deformable time: {def_end - def_start}\n")
         print(f"Total time: {def_end + aff_end - def_start - aff_start}\n")
@@ -168,8 +168,8 @@ def register_dataset(config):
                 moved_lung = moved_lung.detach().cpu().numpy().astype(np.float32)[0, 0]
                 moved_lung = sitk.GetImageFromArray(moved_lung)
                 moved_lung.CopyInformation(fixed_image.images[0].itk_image)
-                sitk.WriteImage(moved_lung, osp.join(out_root_dir, "{:02d}.nii.gz".format(brain_id)))
-                print("Written to {}".format(osp.join(out_root_dir, "{:02d}.nii.gz".format(brain_id))))
+                sitk.WriteImage(moved_lung, osp.join(out_root_dir, "{:02d}.nii.gz".format(lung_id)))
+                print("Written to {}".format(osp.join(out_root_dir, "{:02d}.nii.gz".format(lung_id))))
 
             ## we dont need the segmentations
             with grad_context():
@@ -203,12 +203,12 @@ def register_dataset(config):
                 defX.CopyInformation(fixed_image.itk_image)
                 defY.CopyInformation(fixed_image.itk_image)
                 defZ.CopyInformation(fixed_image.itk_image)
-                brain_dir = osp.join(out_root_dir, "{:02d}".format(brain_id))
-                os.makedirs(brain_dir, exist_ok=True)
-                sitk.WriteImage(defX, osp.join(brain_dir, "defX.mhd"))
-                sitk.WriteImage(defY, osp.join(brain_dir, "defY.mhd"))
-                sitk.WriteImage(defZ, osp.join(brain_dir, "defZ.mhd"))
-                print("Written to {}".format(brain_dir))
+                lung_dir = osp.join(out_root_dir, "{:02d}".format(lung_id))
+                os.makedirs(lung_dir, exist_ok=True)
+                sitk.WriteImage(defX, osp.join(lung_dir, "defX.mhd"))
+                sitk.WriteImage(defY, osp.join(lung_dir, "defY.mhd"))
+                sitk.WriteImage(defZ, osp.join(lung_dir, "defZ.mhd"))
+                print("Written to {}".format(lung_dir))
                 del fixed_image
 
 
@@ -247,7 +247,7 @@ def register_dataset(config):
         print("Running...")
         results = tuner.fit()
         ray.shutdown()
-    # create evaluation brains and deformation fields
+    # create evaluation lung and deformation fields
     elif mode == 'eval':
         register_dataset(config)
     else:

tutorials/[Tutorial 1] Basic Usage.ipynb

@@ -319,7 +319,11 @@
    "id": "73600c28",
    "metadata": {},
    "source": [
-    "# Let's run deformable registration"
+    "# Let's run deformable registration\n",
+    "\n",
+    "The affine transform aligned the moving image to overlap more with the fixed image, but there wasn't much to align in the first place. Therefore, only the anterior portion of the skull seems a little deformed.\n",
+    "\n",
+    "Let's run deformable registration and see the difference."
    ]
   },
   {
@@ -441,7 +445,9 @@
    "id": "79cc7f31",
    "metadata": {},
    "source": [
-    "### Note the moved image looks a lot more like the fixed image now. Let's see if the diffeomorphic constraint is satisfied."
+    "Note the moved image looks a lot more like the fixed image now. Let's see if the diffeomorphic constraint is satisfied.\n",
+    "\n",
+    "## Check diffeomorphism"
    ]
   },
   {