Add SIIM ACR Pneumothorax dataset

scripts/datasets/check_siim_acr.py

@@ -0,0 +1,23 @@
+from torch_em.util.debug import check_loader
+from torch_em.data import MinInstanceSampler
+from torch_em.data.datasets.medical import get_siim_acr_loader
+
+
+ROOT = "/media/anwai/ANWAI/data/siim_acr"
+
+
+def check_siim_acr():
+    loader = get_siim_acr_loader(
+        path=ROOT,
+        split="train",
+        patch_shape=(512, 512),
+        batch_size=2,
+        download=True,
+        resize_inputs=True,
+        sampler=MinInstanceSampler()
+    )
+    check_loader(loader, 8)
+
+
+if __name__ == "__main__":
+    check_siim_acr()

torch_em/data/datasets/medical/__init__.py

@@ -1,2 +1,3 @@
 from .autopet import get_autopet_loader
 from .btcv import get_btcv_dataset, get_btcv_loader
+from .siim_acr import get_siim_acr_dataset, get_siim_acr_loader

torch_em/data/datasets/medical/siim_acr.py

@@ -0,0 +1,99 @@
+import os
+from glob import glob
+from typing import Union, Tuple
+
+import torch_em
+from torch_em.transform.generic import ResizeInputs
+
+from .. import util
+from ... import ImageCollectionDataset
+
+
+KAGGLE_DATASET_NAME = "vbookshelf/pneumothorax-chest-xray-images-and-masks"
+CHECKSUM = "1ade68d31adb996c531bb686fb9d02fe11876ddf6f25594ab725e18c69d81538"
+
+
+def get_siim_acr_data(path, download):
+    os.makedirs(path, exist_ok=True)
+
+    data_dir = os.path.join(path, "siim-acr-pneumothorax")
+    if os.path.exists(data_dir):
+        return data_dir
+
+    util.download_source_kaggle(path=path, dataset_name=KAGGLE_DATASET_NAME, download=download)
+
+    zip_path = os.path.join(path, "pneumothorax-chest-xray-images-and-masks.zip")
+    util._check_checksum(path=zip_path, checksum=CHECKSUM)
+    util.unzip(zip_path=zip_path, dst=path)
+
+    return data_dir
+
+
+def _get_siim_acr_paths(path, split, download):
+    data_dir = get_siim_acr_data(path=path, download=download)
+
+    assert split in ["train", "test"], f"'{split}' is not a valid split."
+
+    image_paths = sorted(glob(os.path.join(data_dir, "png_images", f"*_{split}_*.png")))
+    gt_paths = sorted(glob(os.path.join(data_dir, "png_masks", f"*_{split}_*.png")))
+
+    return image_paths, gt_paths
+
+
+def get_siim_acr_dataset(
+    path: Union[os.PathLike, str],
+    split: str,
+    patch_shape: Tuple[int, int],
+    download: bool = False,
+    resize_inputs: bool = False,
+    **kwargs
+):
+    """Dataset for pneumothorax segmentation in CXR.
+
+    The database is located at https://www.kaggle.com/datasets/vbookshelf/pneumothorax-chest-xray-images-and-masks/data
+
+    This dataset is from the "SIIM-ACR Pneumothorax Segmentation" competition:
+    https://kaggle.com/competitions/siim-acr-pneumothorax-segmentation
+
+    Please cite it if you use this dataset for a publication.
+    """
+    image_paths, gt_paths = _get_siim_acr_paths(path=path, split=split, download=download)
+
+    if resize_inputs:
+        raw_trafo = ResizeInputs(target_shape=patch_shape, is_label=False)
+        label_trafo = ResizeInputs(target_shape=patch_shape, is_label=True)
+        patch_shape = None
+    else:
+        patch_shape = patch_shape
+        raw_trafo, label_trafo = None, None
+
+    dataset = ImageCollectionDataset(
+        raw_image_paths=image_paths,
+        label_image_paths=gt_paths,
+        patch_shape=patch_shape,
+        raw_transform=raw_trafo,
+        label_transform=label_trafo,
+        **kwargs
+    )
+    dataset.max_sampling_attempts = 5000
+
+    return dataset
+
+
+def get_siim_acr_loader(
+    path: Union[os.PathLike, str],
+    split: str,
+    patch_shape: Tuple[int, int],
+    batch_size: int,
+    download: bool = False,
+    resize_inputs: bool = False,
+    **kwargs
+):
+    """Dataloader for pneumothorax segmentation in CXR. See `get_siim_acr_dataset` for details.
+    """
+    ds_kwargs, loader_kwargs = util.split_kwargs(torch_em.default_segmentation_dataset, **kwargs)
+    dataset = get_siim_acr_dataset(
+        path=path, split=split, patch_shape=patch_shape, download=download, resize_inputs=resize_inputs, **ds_kwargs
+    )
+    loader = torch_em.get_data_loader(dataset=dataset, batch_size=batch_size, **loader_kwargs)
+    return loader

torch_em/data/datasets/util.py

@@ -146,6 +146,23 @@ def download_source_empiar(path, access_id, download):
     return download_path
 
 
+def download_source_kaggle(path, dataset_name, download):
+    if not download:
+        raise RuntimeError(f"Cannot fine the data at {path}, but download was set to False.")
+
+    try:
+        from kaggle.api.kaggle_api_extended import KaggleApi
+    except ModuleNotFoundError:
+        msg = "Please install the Kaggle API. You can do this using 'pip install kaggle'. "
+        msg += "After you have installed kaggle, you would need an API token. "
+        msg += "Follow the instructions at https://www.kaggle.com/docs/api."
+        raise ModuleNotFoundError(msg)
+
+    api = KaggleApi()
+    api.authenticate()
+    api.dataset_download_files(dataset=dataset_name, path=path, quiet=False)
+
+
 def update_kwargs(kwargs, key, value, msg=None):
     if key in kwargs:
         msg = f"{key} will be over-ridden in loader kwargs." if msg is None else msg

torch_em/data/image_collection_dataset.py

@@ -68,7 +68,8 @@ def __init__(
         self.label_images = label_image_paths
         self._ndim = 2
 
-        assert len(patch_shape) == self._ndim
+        if patch_shape is not None:
+            assert len(patch_shape) == self._ndim
         self.patch_shape = patch_shape
 
         self.raw_transform = raw_transform
@@ -95,11 +96,16 @@ def ndim(self):
         return self._ndim
 
     def _sample_bounding_box(self, shape):
+        if self.patch_shape is None:
+            patch_shape_for_bb = shape
+        else:
+            patch_shape_for_bb = self.patch_shape
+
         bb_start = [
             np.random.randint(0, sh - psh) if sh - psh > 0 else 0
-            for sh, psh in zip(shape, self.patch_shape)
+            for sh, psh in zip(shape, patch_shape_for_bb)
         ]
-        return tuple(slice(start, start + psh) for start, psh in zip(bb_start, self.patch_shape))
+        return tuple(slice(start, start + psh) for start, psh in zip(bb_start, patch_shape_for_bb))
 
     def _ensure_patch_shape(self, raw, labels, have_raw_channels, have_label_channels, channel_first):
         shape = raw.shape
@@ -137,7 +143,8 @@ def _load_data(self, raw_path, label_path):
         if have_raw_channels:
             channel_first = raw.shape[-1] > 16
 
-        raw, label = self._ensure_patch_shape(raw, label, have_raw_channels, have_label_channels, channel_first)
+        if self.patch_shape is not None:
+            raw, label = self._ensure_patch_shape(raw, label, have_raw_channels, have_label_channels, channel_first)
         shape = raw.shape
 
         prefix_box = tuple()

torch_em/transform/generic.py

@@ -3,7 +3,7 @@
 import numpy as np
 import torch
 
-from skimage.transform import rescale
+from skimage.transform import rescale, resize
 
 
 class Tile(torch.nn.Module):
@@ -72,6 +72,28 @@ def __call__(self, *inputs):
         return outputs
 
 
+class ResizeInputs:
+    def __init__(self, target_shape, is_label=False):
+        self.target_shape = target_shape
+        self.is_label = is_label
+
+    def __call__(self, inputs):
+        if self.is_label:
+            anti_aliasing = True
+        else:
+            anti_aliasing = False
+
+        inputs = resize(
+            image=inputs,
+            output_shape=self.target_shape,
+            order=3,
+            anti_aliasing=anti_aliasing,
+            preserve_range=True,
+        )
+
+        return inputs
+
+
 class PadIfNecessary:
     def __init__(self, shape):
         self.shape = tuple(shape)