Modify datasets & transforms script

rsna_datasets.py

@@ -15,7 +15,6 @@ class Segmentation3DDataset(IterableDataset):
         split: Literal["train", "test"],
         streaming: bool = True,
         volume_transforms: monai.transforms.Compose = None,
-        mask_transforms: monai.transforms.Compose = None,
         transform_configs: TypedDict(
             "",
             {
@@ -53,13 +52,6 @@ class Segmentation3DDataset(IterableDataset):
             streaming=streaming,
         )
 
-        self.mask_transforms = mask_transforms or rsna_transforms.mask_transforms(
-            crop_strategy=transform_configs["crop_strategy"],
-            voxel_spacing=transform_configs["voxel_spacing"],
-            volume_size=transform_configs["volume_size"],
-            axcodes=transform_configs["axcodes"],
-            streaming=streaming,
-        )
         self.yield_extra_info = True  # For debugging purposes
 
     def __iter__(self):
@@ -78,23 +70,20 @@ class Segmentation3DDataset(IterableDataset):
                 yield from self._process_one_sample(data, worker_id=worker_id)
 
     def _process_one_sample(self, data, worker_id):
-        img_data = self.volume_transforms(
-            {"img": data["img_path"], "metadata": data["metadata"]}
-        )
-        seg_data = self.mask_transforms({"seg": data["seg_path"]})
+        data["img"] = data.pop("img_path")
+        data["seg"] = data.pop("seg_path")
+        data = self.volume_transforms(data)
 
-        img_data = [img_data] if not isinstance(img_data, (list, tuple)) else img_data
-        seg_data = [seg_data] if not isinstance(seg_data, (list, tuple)) else seg_data
+        data = [data] if not isinstance(data, (list, tuple)) else data
 
-        for img, seg in zip(img_data, seg_data):
+        for crop in data:
             to_yield = {
-                "img": img["img"],
-                "seg": seg["seg"],
+                "img": crop["img"],
+                "seg": crop["seg"],
             }
             if self.yield_extra_info:
                 to_yield["worker_id"] = worker_id
-                to_yield["series_id"] = data["metadata"]["series_id"]
-
+                to_yield["series_id"] = data[0]["metadata"]["series_id"]
             yield to_yield
 
 
@@ -188,7 +177,6 @@ class MaskedClassification3DDataset(IterableDataset):
         split: Literal["train", "test"],
         streaming: bool = True,
         volume_transforms: monai.transforms.Compose = None,
-        mask_transforms: monai.transforms.Compose = None,
         transform_configs: TypedDict(
             "",
             {
@@ -225,13 +213,6 @@ class MaskedClassification3DDataset(IterableDataset):
             axcodes=transform_configs["axcodes"],
             streaming=streaming,
         )
-        self.mask_transforms = mask_transforms or rsna_transforms.mask_transforms(
-            crop_strategy=transform_configs["crop_strategy"],
-            voxel_spacing=transform_configs["voxel_spacing"],
-            volume_size=transform_configs["volume_size"],
-            axcodes=transform_configs["axcodes"],
-            streaming=streaming,
-        )
 
         self.yield_extra_info = True
 
@@ -251,17 +232,23 @@ class MaskedClassification3DDataset(IterableDataset):
                 yield from self._process_one_sample(data, worker_id=worker_id)
 
     def _process_one_sample(self, data, worker_id):
-        img_data = self.volume_transforms(
-            {"img": data["img_path"], "metadata": data["metadata"]}
+        img_seg_data = self.volume_transforms(
+            {
+                "img": data["img_path"],
+                "seg": data["seg_path"],
+                "metadata": data["metadata"],
+            }
+        )
+        img_seg_data = (
+            [img_seg_data]
+            if not isinstance(img_seg_data, (list, tuple))
+            else img_seg_data
         )
-        seg_data = self.mask_transforms({"seg": data["seg_path"]})
-        img_data = [img_data] if not isinstance(img_data, (list, tuple)) else img_data
-        seg_data = [seg_data] if not isinstance(seg_data, (list, tuple)) else seg_data
 
-        for img, seg in zip(img_data, seg_data):
+        for img_seg in img_seg_data:
             to_yield = {
-                "img": img["img"],
-                "seg": seg["seg"],
+                "img": img_seg["img"],
+                "seg": img_seg["seg"],
                 "bowel": data["bowel"],
                 "extravasation": data["extravasation"],
                 "kidney": data["kidney"],
@@ -283,7 +270,6 @@ class Segmentation2DDataset(IterableDataset):
         split: Literal["train", "test"],
         streaming: bool = True,
         volume_transforms: monai.transforms.Compose = None,
-        mask_transforms: monai.transforms.Compose = None,
         slice_transforms: torchvision.transforms.Compose = None,
         volume_transform_configs: TypedDict(
             "",
@@ -333,13 +319,6 @@ class Segmentation2DDataset(IterableDataset):
             axcodes=volume_transform_configs["axcodes"],
             streaming=streaming,
         )
-        self.mask_transforms = mask_transforms or rsna_transforms.mask_transforms(
-            crop_strategy=volume_transform_configs["crop_strategy"],
-            voxel_spacing=volume_transform_configs["voxel_spacing"],
-            volume_size=volume_transform_configs["volume_size"],
-            axcodes=volume_transform_configs["axcodes"],
-            streaming=streaming,
-        )
         self.slice_transforms = slice_transforms or rsna_transforms.slice_transforms(
             crop_strategy=slice_transform_configs["crop_strategy"],
             shorter_edge_length=slice_transform_configs["shorter_edge_length"],
@@ -363,26 +342,20 @@ class Segmentation2DDataset(IterableDataset):
                 yield from self._process_one_sample(data, worker_id=worker_id)
 
     def _process_one_sample(self, data, worker_id):
-        vol_img_data = self.volume_transforms(
-            {"img": data["img_path"], "metadata": data["metadata"]}
-        )
-        vol_seg_data = self.mask_transforms({"seg": data["seg_path"]})
-        vol_img_data = (
-            [vol_img_data]
-            if not isinstance(vol_img_data, (list, tuple))
-            else vol_img_data
-        )
-        vol_seg_data = (
-            [vol_seg_data]
-            if not isinstance(vol_seg_data, (list, tuple))
-            else vol_seg_data
+        vol_data = self.volume_transforms(
+            {
+                "img": data["img_path"],
+                "seg": data["seg_path"],
+                "metadata": data["metadata"],
+            }
         )
+        vol_data = [vol_data] if not isinstance(vol_data, (list, tuple)) else vol_data
 
-        for vol_img, vol_seg in zip(vol_img_data, vol_seg_data):
-            slice_len = vol_img["img"].size()[-1]
+        for vol in vol_data:
+            slice_len = vol["img"].size()[-1]
             for i in range(slice_len):
-                slice_img_data = self.slice_transforms(vol_img["img"][..., i])
-                slice_seg_data = self.slice_transforms(vol_seg["seg"][..., i])
+                slice_img_data = self.slice_transforms(vol["img"][..., i])
+                slice_seg_data = self.slice_transforms(vol["seg"][..., i])
 
                 slice_img_data = (
                     [slice_img_data]
@@ -528,7 +501,6 @@ class MaskedClassification2DDataset(IterableDataset):
         split: Literal["train", "test"],
         streaming: bool = True,
         volume_transforms: monai.transforms.Compose = None,
-        mask_transforms: monai.transforms.Compose = None,
         slice_transforms: torchvision.transforms.Compose = None,
         volume_transform_configs: TypedDict(
             "",
@@ -579,14 +551,6 @@ class MaskedClassification2DDataset(IterableDataset):
             streaming=streaming,
         )
 
-        self.mask_transforms = mask_transforms or rsna_transforms.mask_transforms(
-            crop_strategy=volume_transform_configs["crop_strategy"],
-            voxel_spacing=volume_transform_configs["voxel_spacing"],
-            volume_size=volume_transform_configs["volume_size"],
-            axcodes=volume_transform_configs["axcodes"],
-            streaming=streaming,
-        )
-
         self.slice_transforms = slice_transforms or rsna_transforms.slice_transforms(
             crop_strategy=slice_transform_configs["crop_strategy"],
             shorter_edge_length=slice_transform_configs["shorter_edge_length"],
@@ -610,26 +574,20 @@ class MaskedClassification2DDataset(IterableDataset):
                 yield from self._process_one_sample(data, worker_id=worker_id)
 
     def _process_one_sample(self, data, worker_id):
-        vol_img_data = self.volume_transforms(
-            {"img": data["img_path"], "metadata": data["metadata"]}
-        )
-        vol_seg_data = self.mask_transforms({"seg": data["seg_path"]})
-        vol_img_data = (
-            [vol_img_data]
-            if not isinstance(vol_img_data, (list, tuple))
-            else vol_img_data
-        )
-        vol_seg_data = (
-            [vol_seg_data]
-            if not isinstance(vol_seg_data, (list, tuple))
-            else vol_seg_data
+        vol_data = self.volume_transforms(
+            {
+                "img": data["img_path"],
+                "seg": data["seg_path"],
+                "metadata": data["metadata"],
+            }
         )
+        vol_data = [vol_data] if not isinstance(vol_data, (list, tuple)) else vol_data
 
-        for vol_img, vol_seg in zip(vol_img_data, vol_seg_data):
-            slice_len = vol_img["img"].size()[-1]
+        for vol in vol_data:
+            slice_len = vol["img"].size()[-1]
             for i in range(slice_len):
-                slice_img_data = self.slice_transforms(vol_img["img"][..., i])
-                slice_seg_data = self.slice_transforms(vol_seg["seg"][..., i])
+                slice_img_data = self.slice_transforms(vol["img"][..., i])
+                slice_seg_data = self.slice_transforms(vol["seg"][..., i])
 
                 slice_img_data = (
                     [slice_img_data]
@@ -658,4 +616,3 @@ class MaskedClassification2DDataset(IterableDataset):
                         to_yield["series_id"] = data["metadata"]["series_id"]
 
                     yield to_yield
-

rsna_transforms.py

@@ -1,4 +1,4 @@
-from typing import Optional, Literal
+from typing import Optional, Literal, Union
 
 from io import BytesIO
 import numpy as np
@@ -386,206 +386,79 @@ def volume_transforms(
     volume_size: tuple[int, int, int] = (96, 96, 96),
     axcodes: str = "RAS",
     streaming: bool = False,
-) -> monai.transforms.Compose:
-    if crop_strategy == "oversample":
-        return monai.transforms.Compose(
-            [
-                LoadNIfTIFromHFHubd(keys=["img"])
-                if streaming
-                else LoadNIfTIFromLocalCached(keys=["img"]),
-                monai.transforms.EnsureTyped(
-                    keys=["img"], data_type="tensor", dtype=torch.float32
-                ),
-                UnifyUnusualNIfTI(
-                    x_key="img",
-                    metadata_key="metadata",
-                    meta_pixel_representation_key="pixel_representation",
-                    meta_bits_allocated_key="bits_allocated",
-                    meta_bits_stored_key="bits_stored",
-                ),
-                monai.transforms.EnsureChannelFirstd(keys=["img"]),
-                monai.transforms.Orientationd(keys=["img"], axcodes=axcodes),
-                monai.transforms.Spacingd(
-                    keys=["img"], pixdim=voxel_spacing, mode=["bilinear"]
-                ),
-                monai.transforms.NormalizeIntensityd(keys=["img"], nonzero=False),
-                monai.transforms.ScaleIntensityd(keys=["img"], minv=-1.0, maxv=1.0),
-                monai.transforms.SpatialPadd(keys=["img"], spatial_size=volume_size),
-                monai.transforms.RandSpatialCropSamplesd(
-                    keys=["img"],
-                    roi_size=volume_size,
-                    num_samples=3,
-                    random_center=True,
-                    random_size=False,
-                ),
-            ]
-        )
+):
+    transform_steps = [
+        LoadNIfTIFromHFHubd(keys=["img", "seg"], allow_missing_keys=True)
+        if streaming
+        else LoadNIfTIFromLocalCached(keys=["img", "seg"], allow_missing_keys=True),
+        monai.transforms.EnsureTyped(
+            keys=["img", "seg"],
+            data_type="tensor",
+            dtype=torch.float32,
+            allow_missing_keys=True,
+        ),
+        UnifyUnusualNIfTI(
+            x_key="img",
+            metadata_key="metadata",
+            meta_pixel_representation_key="pixel_representation",
+            meta_bits_allocated_key="bits_allocated",
+            meta_bits_stored_key="bits_stored",
+        ),
+        monai.transforms.EnsureChannelFirstd(
+            keys=["img", "seg"], allow_missing_keys=True
+        ),
+        monai.transforms.Orientationd(
+            keys=["img", "seg"], axcodes=axcodes, allow_missing_keys=True
+        ),
+        monai.transforms.Spacingd(
+            keys=["img", "seg"],
+            pixdim=voxel_spacing,
+            mode=["bilinear", "nearest"],
+            allow_missing_keys=True,
+        ),
+        monai.transforms.NormalizeIntensityd(keys=["img"], nonzero=False),
+        monai.transforms.ScaleIntensityd(keys=["img"], minv=-1.0, maxv=1.0),
+        monai.transforms.SpatialPadd(
+            keys=["img", "seg"], spatial_size=volume_size, allow_missing_keys=True
+        ),
+    ]
 
-    elif crop_strategy == "center":
-        return monai.transforms.Compose(
-            [
-                LoadNIfTIFromHFHubd(keys=["img"])
-                if streaming
-                else LoadNIfTIFromLocalCached(keys=["img"]),
-                monai.transforms.EnsureTyped(
-                    keys=["img"], data_type="tensor", dtype=torch.float32
-                ),
-                UnifyUnusualNIfTI(
-                    x_key="img",
-                    metadata_key="metadata",
-                    meta_pixel_representation_key="pixel_representation",
-                    meta_bits_allocated_key="bits_allocated",
-                    meta_bits_stored_key="bits_stored",
-                ),
-                monai.transforms.EnsureChannelFirstd(keys=["img"]),
-                monai.transforms.Orientationd(keys=["img"], axcodes=axcodes),
-                monai.transforms.Spacingd(
-                    keys=["img"], pixdim=voxel_spacing, mode=["bilinear"]
-                ),
-                monai.transforms.NormalizeIntensityd(keys=["img"], nonzero=False),
-                monai.transforms.ScaleIntensityd(keys=["img"], minv=-1.0, maxv=1.0),
-                monai.transforms.SpatialPadd(keys=["img"], spatial_size=volume_size),
-                monai.transforms.CenterSpatialCropd(keys=["img"], roi_size=volume_size),
-            ]
+    if crop_strategy == "oversample":
+        transform_steps.append(
+            monai.transforms.RandSpatialCropSamplesd(
+                keys=["img", "seg"],
+                roi_size=volume_size,
+                num_samples=3,
+                random_center=True,
+                random_size=False,
+                allow_missing_keys=True,
+            )
         )
-
     elif crop_strategy == "random":
-        return monai.transforms.Compose(
-            [
-                LoadNIfTIFromHFHubd(keys=["img"])
-                if streaming
-                else LoadNIfTIFromLocalCached(keys=["img"]),
-                monai.transforms.EnsureTyped(
-                    keys=["img"], data_type="tensor", dtype=torch.float32
-                ),
-                UnifyUnusualNIfTI(
-                    x_key="img",
-                    metadata_key="metadata",
-                    meta_pixel_representation_key="pixel_representation",
-                    meta_bits_allocated_key="bits_allocated",
-                    meta_bits_stored_key="bits_stored",
-                ),
-                monai.transforms.EnsureChannelFirstd(keys=["img"]),
-                monai.transforms.Orientationd(keys=["img"], axcodes=axcodes),
-                monai.transforms.Spacingd(
-                    keys=["img"], pixdim=voxel_spacing, mode=["bilinear"]
-                ),
-                monai.transforms.NormalizeIntensityd(keys=["img"], nonzero=False),
-                monai.transforms.ScaleIntensityd(keys=["img"], minv=-1.0, maxv=1.0),
-                monai.transforms.SpatialPadd(keys=["img"], spatial_size=volume_size),
-                monai.transforms.RandSpatialCropd(
-                    keys=["img"],
-                    roi_size=volume_size,
-                    random_center=True,
-                    random_size=False,
-                ),
-            ]
-        )
-
-    elif crop_strategy == "none" or crop_strategy is None:
-        return monai.transforms.Compose(
-            [
-                LoadNIfTIFromHFHubd(keys=["img"])
-                if streaming
-                else LoadNIfTIFromLocalCached(keys=["img"]),
-                monai.transforms.EnsureTyped(
-                    keys=["img"], data_type="tensor", dtype=torch.float32
-                ),
-                UnifyUnusualNIfTI(
-                    x_key="img",
-                    metadata_key="metadata",
-                    meta_pixel_representation_key="pixel_representation",
-                    meta_bits_allocated_key="bits_allocated",
-                    meta_bits_stored_key="bits_stored",
-                ),
-                monai.transforms.EnsureChannelFirstd(keys=["img"]),
-                monai.transforms.Orientationd(keys=["img"], axcodes=axcodes),
-                monai.transforms.Spacingd(
-                    keys=["img"], pixdim=voxel_spacing, mode=["bilinear"]
-                ),
-                monai.transforms.NormalizeIntensityd(keys=["img"], nonzero=False),
-                monai.transforms.ScaleIntensityd(keys=["img"], minv=-1.0, maxv=1.0),
-            ]
-        )
-
-    else:
-        raise ValueError(
-            f"crop_strategy must be one of ['oversample', 'center', 'random', 'none'], got {crop_strategy}."
-        )
-
-
-def mask_transforms(
-    crop_strategy: Optional[Literal["oversample", "center", "none"]] = "oversample",
-    voxel_spacing: tuple[float, float, float] = (3.0, 3.0, 3.0),
-    volume_size: tuple[int, int, int] = (96, 96, 96),
-    axcodes: str = "RAS",
-    streaming: bool = False,
-) -> monai.transforms.Compose:
-    if crop_strategy == "oversample":
-        return monai.transforms.Compose(
-            [
-                LoadNIfTIFromHFHubd(keys=["seg"])
-                if streaming
-                else LoadNIfTIFromLocalCached(keys=["seg"]),
-                monai.transforms.EnsureTyped(
-                    keys=["seg"], data_type="tensor", dtype=torch.float32
-                ),
-                monai.transforms.EnsureChannelFirstd(keys=["seg"]),
-                monai.transforms.Orientationd(keys=["seg"], axcodes=axcodes),
-                monai.transforms.Spacingd(
-                    keys=["seg"], pixdim=voxel_spacing, mode=["nearest"]
-                ),
-                monai.transforms.SpatialPadd(keys=["seg"], spatial_size=volume_size),
-                monai.transforms.RandSpatialCropSamplesd(
-                    keys=["seg"],
-                    roi_size=volume_size,
-                    num_samples=3,
-                    random_center=True,
-                    random_size=False,
-                ),
-            ]
+        transform_steps.append(
+            monai.transforms.RandSpatialCropd(
+                keys=["img", "seg"],
+                roi_size=volume_size,
+                random_center=True,
+                random_size=False,
+                allow_missing_keys=True,
+            )
         )
-
     elif crop_strategy == "center":
-        return monai.transforms.Compose(
-            [
-                LoadNIfTIFromHFHubd(keys=["seg"])
-                if streaming
-                else LoadNIfTIFromLocalCached(keys=["seg"]),
-                monai.transforms.EnsureTyped(
-                    keys=["seg"], data_type="tensor", dtype=torch.float32
-                ),
-                monai.transforms.EnsureChannelFirstd(keys=["seg"]),
-                monai.transforms.Orientationd(keys=["seg"], axcodes=axcodes),
-                monai.transforms.Spacingd(
-                    keys=["seg"], pixdim=voxel_spacing, mode=["nearest"]
-                ),
-                monai.transforms.SpatialPadd(keys=["seg"], spatial_size=volume_size),
-                monai.transforms.CenterSpatialCropd(keys=["seg"], roi_size=volume_size),
-            ]
+        transform_steps.append(
+            monai.transforms.CenterSpatialCropd(
+                keys=["img", "seg"], roi_size=volume_size, allow_missing_keys=True
+            )
         )
-
     elif crop_strategy == "none" or crop_strategy is None:
-        return monai.transforms.Compose(
-            [
-                LoadNIfTIFromHFHubd(keys=["seg"])
-                if streaming
-                else LoadNIfTIFromLocalCached(keys=["seg"]),
-                monai.transforms.EnsureTyped(
-                    keys=["seg"], data_type="tensor", dtype=torch.float32
-                ),
-                monai.transforms.EnsureChannelFirstd(keys=["seg"]),
-                monai.transforms.Orientationd(keys=["seg"], axcodes=axcodes),
-                monai.transforms.Spacingd(
-                    keys=["seg"], pixdim=voxel_spacing, mode=["nearest"]
-                ),
-            ]
-        )
+        pass
     else:
         raise ValueError(
-            f"crop_strategy must be one of ['oversample', 'center', 'none'], got {crop_strategy}."
+            f"crop_strategy must be one of ['oversample', 'center', 'random', 'none'], got {crop_strategy}."
         )
 
+    return monai.transforms.Compose(transform_steps)
+
 
 def slice_transforms(
     crop_strategy: Literal["ten", "five", "center", "random"] = "ten",