changed input to 1 channel(pet)

Author: vaibhavbahel

src/baseline/v2/datasets/transforms.py

@@ -5,10 +5,9 @@
     NormalizeIntensityd,
     ScaleIntensityRanged,
     ConcatItemsd,
-    RandCropByPosNegLabeld,
+    RandSpatialCropd,
     RandFlipd,
     EnsureTyped,
-    Lambdad
 )
 
 
@@ -17,21 +16,16 @@ def get_train_transforms(patch_size, spacing):
     transforms = Compose(
         [
 
-            LoadImaged(keys=["pet", "topogram", "mri_in", "mri_out", "ct"]),
+            LoadImaged(keys=["pet", "ct"]),
 
-            EnsureChannelFirstd(keys=["pet", "topogram", "mri_in", "mri_out", "ct"]),
+            EnsureChannelFirstd(keys=["pet", "ct"]),
 
-            # expand topogram depth
-            Lambdad(keys=["topogram"], func=lambda x: x.repeat(1,1,1,531)),
-
-            # normalize PET and MRI
             NormalizeIntensityd(
-                keys=["pet","mri_in","mri_out"],
+                keys=["pet"],
                 nonzero=True,
                 channel_wise=True
             ),
 
-            # normalize CT
             ScaleIntensityRanged(
                 keys=["ct"],
                 a_min=-1000,
@@ -41,20 +35,16 @@ def get_train_transforms(patch_size, spacing):
                 clip=True,
             ),
 
-            # combine modalities
+            # now input is just PET
             ConcatItemsd(
-                keys=["pet","topogram","mri_in","mri_out"],
+                keys=["pet"],
                 name="input"
             ),
 
-            # patch sampling
-            RandCropByPosNegLabeld(
+            RandSpatialCropd(
                 keys=["input","ct"],
-                label_key="ct",
-                spatial_size=patch_size,
-                pos=1,
-                neg=1,
-                num_samples=1
+                roi_size=patch_size,
+                random_size=False
             ),
 
             RandFlipd(

src/baseline/v2/inference/inference.py

@@ -50,20 +50,18 @@
 
 transforms = Compose(
 [
-    LoadImaged(keys=["pet","topogram","mri_in","mri_out"]),
+    LoadImaged(keys=["pet"]),
 
-    EnsureChannelFirstd(keys=["pet","topogram","mri_in","mri_out"]),
-
-    Lambdad(keys=["topogram"], func=lambda x: x.repeat(1,1,1,531)),
+    EnsureChannelFirstd(keys=["pet"]),
 
     NormalizeIntensityd(
-        keys=["pet","mri_in","mri_out"],
+        keys=["pet"],
         nonzero=True,
         channel_wise=True
     ),
 
     ConcatItemsd(
-        keys=["pet","topogram","mri_in","mri_out"],
+        keys=["pet"],
         name="input"
     ),
 
@@ -99,9 +97,6 @@
 
     data = {
         "pet": sub/"features/nacpet.nii.gz",
-        "topogram": sub/"features/topogram.nii.gz",
-        "mri_in": sub/"features/mri_combined_in_phase.nii.gz",
-        "mri_out": sub/"features/mri_combined_out_phase.nii.gz",
     }
 
     data = transforms(data)

src/baseline/v2/models/unet.py

@@ -147,6 +147,6 @@ def forward(self, x):
 def build_model():
 
     return UNet3D(
-        in_channels=4,
+        in_channels=1,  # only PET as input
         out_channels=1
     )