[experiment] Add iostar database

pyproject.toml

@@ -430,6 +430,13 @@ chasedb1-2nd = "mednet.libs.segmentation.config.data.chasedb1.second_annotator"
 # drive dataset - retinography
 drive = "mednet.libs.segmentation.config.data.drive.default"
 
+# iostar - retinography
+iostar-vessel = "mednet.libs.segmentation.config.data.iostar.vessel"
+iostar-disc = "mednet.libs.segmentation.config.data.iostar.optic_disc"
+
+# montgomery county - cxr
+# montgomery = "mednet.libs.segmentation.config.data.montgomery.default"
+
 # stare dataset - retinography
 stare = "mednet.libs.segmentation.config.data.stare.ah"
 stare-2nd = "mednet.libs.segmentation.config.data.stare.vk"

src/mednet/libs/segmentation/config/data/iostar/datamodule.py

+# SPDX-FileCopyrightText: Copyright © 2024 Idiap Research Institute <contact@idiap.ch>
+#
+# SPDX-License-Identifier: GPL-3.0-or-later
+"""IOSTAR (training set) for Vessel and Optic-Disc Segmentation."""
+
+import importlib.resources
+import os
+from pathlib import Path
+
+import PIL.Image
+from mednet.libs.common.data.datamodule import CachingDataModule
+from mednet.libs.common.data.split import JSONDatabaseSplit
+from mednet.libs.common.data.typing import DatabaseSplit, Sample
+from mednet.libs.segmentation.data.typing import (
+    SegmentationRawDataLoader as _SegmentationRawDataLoader,
+)
+from torchvision import tv_tensors
+from torchvision.transforms.functional import to_tensor
+
+from ....utils.rc import load_rc
+
+CONFIGURATION_KEY_DATADIR = "datadir." + (__name__.rsplit(".", 2)[-2])
+"""Key to search for in the configuration file for the root directory of this
+database."""
+
+
+class SegmentationRawDataLoader(_SegmentationRawDataLoader):
+    """A specialized raw-data-loader for the iostar dataset."""
+
+    datadir: str
+    """This variable contains the base directory where the database raw data is
+    stored."""
+
+    def __init__(self):
+        self.datadir = load_rc().get(
+            CONFIGURATION_KEY_DATADIR, os.path.realpath(os.curdir)
+        )
+
+    def sample(self, sample: tuple[str, str, str]) -> Sample:
+        """Load a single image sample from the disk.
+
+        Parameters
+        ----------
+        sample
+            A tuple containing the path suffix, within the dataset root folder,
+            where to find the image to be loaded, and an integer, representing the
+            sample label.
+
+        Returns
+        -------
+            The sample representation.
+        """
+
+        image = PIL.Image.open(Path(self.datadir) / str(sample[0])).convert(
+            mode="RGB"
+        )
+        tensor = tv_tensors.Image(to_tensor(image))
+        target = tv_tensors.Image(
+            to_tensor(
+                PIL.Image.open(Path(self.datadir) / str(sample[1])).convert(
+                    mode="1", dither=None
+                )
+            )
+        )
+        mask = tv_tensors.Mask(
+            to_tensor(
+                PIL.Image.open(Path(self.datadir) / str(sample[2])).convert(
+                    mode="1", dither=None
+                )
+            )
+        )
+
+        return tensor, dict(target=target, mask=mask, name=sample[0])  # type: ignore[arg-type]
+
+
+def make_split(basename: str) -> DatabaseSplit:
+    """Return a database split for the iostar database.
+
+    Parameters
+    ----------
+    basename
+        Name of the .json file containing the split to load.
+
+    Returns
+    -------
+        An instance of DatabaseSplit.
+    """
+
+    return JSONDatabaseSplit(
+        importlib.resources.files(__name__.rsplit(".", 1)[0]).joinpath(basename)
+    )
+
+
+class DataModule(CachingDataModule):
+    """IOSTAR (training set) for Vessel and Optic-Disc Segmentation.
+
+    The IOSTAR vessel segmentation dataset includes 30 images with a resolution of
+    1024 × 1024 pixels. All the vessels in this dataset are annotated by a group of
+    experts working in the field of retinal image analysis. Additionally the
+    dataset includes annotations for the optic disc and the artery/vein ratio.
+
+    * Reference: [IOSTAR-2016]_
+    * Original resolution (height x width): 1024 x 1024
+    * Split reference: [MEYER-2017]_
+    * Protocol ``vessel``:
+
+    * Training samples: 20 (including labels and masks)
+    * Test samples: 10 (including labels and masks)
+
+    * Protocol ``optic-disc``:
+
+    * Training samples: 20 (including labels and masks)
+    * Test samples: 10 (including labels and masks)
+
+    Parameters
+    ----------
+    split_filename
+        Name of the .json file containing the split to load.
+    """
+
+    def __init__(self, split_filename: str):
+        super().__init__(
+            database_split=make_split(split_filename),
+            raw_data_loader=SegmentationRawDataLoader(),
+        )

src/mednet/libs/segmentation/config/data/iostar/optic-disc.json

+{
+ "train": [
+  [
+   "image/STAR 01_OSC.jpg",
+   "mask_OD/STAR 01_OSC_ODMask.tif",
+   "mask/STAR 01_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 02_ODC.jpg",
+   "mask_OD/STAR 02_ODC_ODMask.tif",
+   "mask/STAR 02_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 03_OSN.jpg",
+   "mask_OD/STAR 03_OSN_ODMask.tif",
+   "mask/STAR 03_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 05_ODC.jpg",
+   "mask_OD/STAR 05_ODC_ODMask.tif",
+   "mask/STAR 05_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 06_ODN.jpg",
+   "mask_OD/STAR 06_ODN_ODMask.tif",
+   "mask/STAR 06_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 08_OSN.jpg",
+   "mask_OD/STAR 08_OSN_ODMask.tif",
+   "mask/STAR 08_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 09_OSN.jpg",
+   "mask_OD/STAR 09_OSN_ODMask.tif",
+   "mask/STAR 09_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 10_OSN.jpg",
+   "mask_OD/STAR 10_OSN_ODMask.tif",
+   "mask/STAR 10_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 13_OSN.jpg",
+   "mask_OD/STAR 13_OSN_ODMask.tif",
+   "mask/STAR 13_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 15_OSN.jpg",
+   "mask_OD/STAR 15_OSN_ODMask.tif",
+   "mask/STAR 15_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 16_OSN.jpg",
+   "mask_OD/STAR 16_OSN_ODMask.tif",
+   "mask/STAR 16_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 17_ODN.jpg",
+   "mask_OD/STAR 17_ODN_ODMask.tif",
+   "mask/STAR 17_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 20_ODC.jpg",
+   "mask_OD/STAR 20_ODC_ODMask.tif",
+   "mask/STAR 20_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 21_OSC.jpg",
+   "mask_OD/STAR 21_OSC_ODMask.tif",
+   "mask/STAR 21_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 24_OSC.jpg",
+   "mask_OD/STAR 24_OSC_ODMask.tif",
+   "mask/STAR 24_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 26_ODC.jpg",
+   "mask_OD/STAR 26_ODC_ODMask.tif",
+   "mask/STAR 26_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 28_ODN.jpg",
+   "mask_OD/STAR 28_ODN_ODMask.tif",
+   "mask/STAR 28_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 30_ODC.jpg",
+   "mask_OD/STAR 30_ODC_ODMask.tif",
+   "mask/STAR 30_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 31_ODN.jpg",
+   "mask_OD/STAR 31_ODN_ODMask.tif",
+   "mask/STAR 31_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 32_ODC.jpg",
+   "mask_OD/STAR 32_ODC_ODMask.tif",
+   "mask/STAR 32_ODC_Mask.tif"
+  ]
+ ],
+ "test": [
+  [
+   "image/STAR 34_ODC.jpg",
+   "mask_OD/STAR 34_ODC_ODMask.tif",
+   "mask/STAR 34_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 36_OSC.jpg",
+   "mask_OD/STAR 36_OSC_ODMask.tif",
+   "mask/STAR 36_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 37_ODN.jpg",
+   "mask_OD/STAR 37_ODN_ODMask.tif",
+   "mask/STAR 37_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 38_ODC.jpg",
+   "mask_OD/STAR 38_ODC_ODMask.tif",
+   "mask/STAR 38_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 39_ODC.jpg",
+   "mask_OD/STAR 39_ODC_ODMask.tif",
+   "mask/STAR 39_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 40_OSC.jpg",
+   "mask_OD/STAR 40_OSC_ODMask.tif",
+   "mask/STAR 40_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 43_OSC.jpg",
+   "mask_OD/STAR 43_OSC_ODMask.tif",
+   "mask/STAR 43_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 44_OSN.jpg",
+   "mask_OD/STAR 44_OSN_ODMask.tif",
+   "mask/STAR 44_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 45_ODC.jpg",
+   "mask_OD/STAR 45_ODC_ODMask.tif",
+   "mask/STAR 45_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 48_OSN.jpg",
+   "mask_OD/STAR 48_OSN_ODMask.tif",
+   "mask/STAR 48_OSN_Mask.tif"
+  ]
+ ]
+}

src/mednet/libs/segmentation/config/data/iostar/optic_disc.py

+# SPDX-FileCopyrightText: Copyright © 2024 Idiap Research Institute <contact@idiap.ch>
+#
+# SPDX-License-Identifier: GPL-3.0-or-later
+from mednet.libs.segmentation.config.data.iostar.datamodule import DataModule
+
+datamodule = DataModule("optic-disc.json")

src/mednet/libs/segmentation/config/data/iostar/vessel.json

+{
+ "train": [
+  [
+   "image/STAR 01_OSC.jpg",
+   "GT/STAR 01_OSC_GT.tif",
+   "mask/STAR 01_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 02_ODC.jpg",
+   "GT/STAR 02_ODC_GT.tif",
+   "mask/STAR 02_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 03_OSN.jpg",
+   "GT/STAR 03_OSN_GT.tif",
+   "mask/STAR 03_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 05_ODC.jpg",
+   "GT/STAR 05_ODC_GT.tif",
+   "mask/STAR 05_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 06_ODN.jpg",
+   "GT/STAR 06_ODN_GT.tif",
+   "mask/STAR 06_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 08_OSN.jpg",
+   "GT/STAR 08_OSN_GT.tif",
+   "mask/STAR 08_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 09_OSN.jpg",
+   "GT/STAR 09_OSN_GT.tif",
+   "mask/STAR 09_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 10_OSN.jpg",
+   "GT/STAR 10_OSN_GT.tif",
+   "mask/STAR 10_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 13_OSN.jpg",
+   "GT/STAR 13_OSN_GT.tif",
+   "mask/STAR 13_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 15_OSN.jpg",
+   "GT/STAR 15_OSN_GT.tif",
+   "mask/STAR 15_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 16_OSN.jpg",
+   "GT/STAR 16_OSN_GT.tif",
+   "mask/STAR 16_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 17_ODN.jpg",
+   "GT/STAR 17_ODN_GT.tif",
+   "mask/STAR 17_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 20_ODC.jpg",
+   "GT/STAR 20_ODC_GT.tif",
+   "mask/STAR 20_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 21_OSC.jpg",
+   "GT/STAR 21_OSC_GT.tif",
+   "mask/STAR 21_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 24_OSC.jpg",
+   "GT/STAR 24_OSC_GT.tif",
+   "mask/STAR 24_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 26_ODC.jpg",
+   "GT/STAR 26_ODC_GT.tif",
+   "mask/STAR 26_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 28_ODN.jpg",
+   "GT/STAR 28_ODN_GT.tif",
+   "mask/STAR 28_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 30_ODC.jpg",
+   "GT/STAR 30_ODC_GT.tif",
+   "mask/STAR 30_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 31_ODN.jpg",
+   "GT/STAR 31_ODN_GT.tif",
+   "mask/STAR 31_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 32_ODC.jpg",
+   "GT/STAR 32_ODC_GT.tif",
+   "mask/STAR 32_ODC_Mask.tif"
+  ]
+ ],
+ "test": [
+  [
+   "image/STAR 34_ODC.jpg",
+   "GT/STAR 34_ODC_GT.tif",
+   "mask/STAR 34_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 36_OSC.jpg",
+   "GT/STAR 36_OSC_GT.tif",
+   "mask/STAR 36_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 37_ODN.jpg",
+   "GT/STAR 37_ODN_GT.tif",
+   "mask/STAR 37_ODN_Mask.tif"
+  ],
+  [
+   "image/STAR 38_ODC.jpg",
+   "GT/STAR 38_ODC_GT.tif",
+   "mask/STAR 38_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 39_ODC.jpg",
+   "GT/STAR 39_ODC_GT.tif",
+   "mask/STAR 39_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 40_OSC.jpg",
+   "GT/STAR 40_OSC_GT.tif",
+   "mask/STAR 40_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 43_OSC.jpg",
+   "GT/STAR 43_OSC_GT.tif",
+   "mask/STAR 43_OSC_Mask.tif"
+  ],
+  [
+   "image/STAR 44_OSN.jpg",
+   "GT/STAR 44_OSN_GT.tif",
+   "mask/STAR 44_OSN_Mask.tif"
+  ],
+  [
+   "image/STAR 45_ODC.jpg",
+   "GT/STAR 45_ODC_GT.tif",
+   "mask/STAR 45_ODC_Mask.tif"
+  ],
+  [
+   "image/STAR 48_OSN.jpg",
+   "GT/STAR 48_OSN_GT.tif",
+   "mask/STAR 48_OSN_Mask.tif"
+  ]
+ ]
+}

src/mednet/libs/segmentation/config/data/iostar/vessel.py

+# SPDX-FileCopyrightText: Copyright © 2024 Idiap Research Institute <contact@idiap.ch>
+#
+# SPDX-License-Identifier: GPL-3.0-or-later
+from mednet.libs.segmentation.config.data.iostar.datamodule import DataModule
+
+datamodule = DataModule("vessel.json")

src/mednet/libs/segmentation/tests/test_iostar.py

+# SPDX-FileCopyrightText: Copyright © 2024 Idiap Research Institute <contact@idiap.ch>
+#
+# SPDX-License-Identifier: GPL-3.0-or-later
+"""Tests for iostar dataset."""
+
+import importlib
+
+import pytest
+from click.testing import CliRunner
+
+
+def id_function(val):
+    if isinstance(val, dict):
+        return str(val)
+    return repr(val)
+
+
+@pytest.mark.parametrize(
+    "split,lengths",
+    [
+        ("vessel", dict(train=20, test=10)),
+        ("optic-disc", dict(train=20, test=10)),
+    ],
+    ids=id_function,  # just changes how pytest prints it
+)
+def test_protocol_consistency(
+    database_checkers,
+    split: str,
+    lengths: dict[str, int],
+):
+    from mednet.libs.segmentation.config.data.iostar.datamodule import (
+        make_split,
+    )
+
+    database_checkers.check_split(
+        make_split(f"{split}.json"),
+        lengths=lengths,
+    )
+
+
+@pytest.mark.skip_if_rc_var_not_set("datadir.iostar")
+def test_database_check():
+    from mednet.libs.segmentation.scripts.database import check
+
+    runner = CliRunner()
+    result = runner.invoke(check, ["iostar-vessel"])
+    assert (
+        result.exit_code == 0
+    ), f"Exit code {result.exit_code} != 0 -- Output:\n{result.output}"
+
+
+@pytest.mark.skip_if_rc_var_not_set("datadir.iostar")
+@pytest.mark.parametrize(
+    "dataset",
+    [
+        "train",
+        "test",
+    ],
+)
+@pytest.mark.parametrize(
+    "name",
+    ["vessel", "optic_disc"],
+)
+def test_loading(database_checkers, name: str, dataset: str):
+    datamodule = importlib.import_module(
+        f".{name}",
+        "mednet.libs.segmentation.config.data.iostar",
+    ).datamodule
+
+    datamodule.model_transforms = []  # should be done before setup()
+    datamodule.setup("predict")  # sets up all datasets
+
+    loader = datamodule.predict_dataloader()[dataset]
+
+    limit = 3  # limit load checking
+    for batch in loader:
+        if limit == 0:
+            break
+        database_checkers.check_loaded_batch(
+            batch,
+            batch_size=1,
+            color_planes=3,
+            expected_num_targets=1,
+        )
+        limit -= 1
+
+
+@pytest.mark.skip_if_rc_var_not_set("datadir.iostar")
+def test_raw_transforms_image_quality(database_checkers, datadir):
+    reference_histogram_file = str(
+        datadir / "histograms/raw_data/histograms_iostar_vessel.json",
+    )
+
+    datamodule = importlib.import_module(
+        ".vessel",
+        "mednet.libs.segmentation.config.data.iostar",
+    ).datamodule
+
+    datamodule.model_transforms = []
+    datamodule.setup("predict")
+
+    database_checkers.check_image_quality(datamodule, reference_histogram_file)
+
+
+@pytest.mark.skip_if_rc_var_not_set("datadir.iostar")
+@pytest.mark.parametrize(
+    "model_name",
+    ["lwnet"],
+)
+def test_model_transforms_image_quality(database_checkers, datadir, model_name):
+    reference_histogram_file = str(
+        datadir
+        / f"histograms/models/histograms_{model_name}_iostar_vessel.json",
+    )
+
+    datamodule = importlib.import_module(
+        ".vessel",
+        "mednet.libs.segmentation.config.data.iostar",
+    ).datamodule
+
+    model = importlib.import_module(
+        f".{model_name}",
+        "mednet.libs.segmentation.config.models",
+    ).model
+
+    datamodule.model_transforms = model.model_transforms
+    datamodule.setup("predict")
+
+    database_checkers.check_image_quality(
+        datamodule,
+        reference_histogram_file,
+        compare_type="statistical",
+        pearson_coeff_threshold=0.005,
+    )