Fixing Tests, Fixing MPP issue, Adding tolerance to in memory dataset

.gitignore

@@ -20,3 +20,4 @@ pathopatch.egg-info/
 dist
 build
 push_build.yml
+debug

README.md

@@ -331,3 +331,30 @@ TBD
 <p xmlns:cc="http://creativecommons.org/ns#" xmlns:dct="http://purl.org/dc/terms/"><a property="dct:title" rel="cc:attributionURL" href="https://github.com/TIO-IKIM/PathoPatcher">PathoPatcher</a> by <a rel="cc:attributionURL dct:creator" property="cc:attributionName" href="https://github.com/FabianHoerst">Fabian Hörst, University Hospital Essen,</a> is licensed under <a href="http://creativecommons.org/licenses/by-nc-sa/4.0/?ref=chooser-v1" target="_blank" rel="license noopener noreferrer" style="display:inline-block;">CC BY-NC-SA 4.0
 
   <img style="height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/cc.svg?ref=chooser-v1"><img style="height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/by.svg?ref=chooser-v1"><img style="height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/nc.svg?ref=chooser-v1"><img style="height:22px!important;margin-left:3px;vertical-align:text-bottom;" src="https://mirrors.creativecommons.org/presskit/icons/sa.svg?ref=chooser-v1"></a></p>
+
+## Citation
+```latex
+@InProceedings{10.1007/978-3-658-44037-4_91,
+    author="H{\"o}rst, Fabian
+            and Schaheer, Sajad H.
+            and Baldini, Giulia
+            and Bahnsen, Fin H.
+            and Egger, Jan
+            and Kleesiek, Jens",
+    editor="Maier, Andreas
+            and Deserno, Thomas M.
+            and Handels, Heinz
+            and Maier-Hein, Klaus
+            and Palm, Christoph
+            and Tolxdorff, Thomas",
+    title="Accelerating Artificial Intelligence-based Whole Slide Image Analysis with an Optimized Preprocessing Pipeline",
+    booktitle="Bildverarbeitung f{\"u}r die Medizin 2024",
+    year="2024",
+    publisher="Springer Fachmedien Wiesbaden",
+    address="Wiesbaden",
+    pages="356--361",,
+    isbn="978-3-658-44037-4"
+}
+
+
+```

pathopatch/patch_extraction/dataset.py

@@ -4,7 +4,11 @@
 # @ Fabian Hörst, [email protected]
 # Institute for Artifical Intelligence in Medicine,
 # University Medicine Essen
+import sys
 
+from click import Option
+
+sys.path.append("/Users/fhoerst/Fabian-Projekte/Preprocessing/PathoPatcher")
 
 import logging
 import os
@@ -22,8 +26,7 @@
 from shapely.geometry import Polygon
 from torch.utils.data import Dataset
 from torchvision.transforms.v2 import ToTensor
-
-from pathopatch import logger
+from PIL import Image
 from pathopatch.utils.exceptions import WrongParameterException
 from pathopatch.utils.patch_util import (
     DeepZoomGeneratorOS,
@@ -43,11 +46,11 @@
 warnings.filterwarnings("ignore", category=UserWarning)
 
 
-class PreProcessingDatasetConfig(BaseModel):
+class LivePatchWSIConfig(BaseModel):
     """Storing the configuration for the PatchWSIDataset
 
     Args:
-        wsipath (str): Path to the WSI
+        wsi_path (str): Path to the WSI
         wsi_properties (dict, optional): Dictionary with manual WSI metadata, but just applies if metadata cannot be derived from OpenSlide (e.g., for .tiff files). Supported keys are slide_mpp and magnification
         patch_size (int, optional): The size of the patches in pixel that will be retrieved from the WSI, e.g. 256 for 256px. Defaults to 256.
         patch_overlap (float, optional): The percentage amount pixels that should overlap between two different patches.
@@ -63,6 +66,7 @@ class PreProcessingDatasetConfig(BaseModel):
             expresses which kind of downsampling should be used with
             respect to the highest possible resolution. Defaults to 0.
         level (int, optional): The tile level for sampling, alternative to downsample. Defaults to None.
+        target_mpp_tolerance(float, optional): Tolerance for the target_mpp. If wsi mpp is within a range target_mpp +/- tolarance, no rescaling is performed. Defaults to 0.0.
         annotation_path (str, optional): Path to the .json file with the annotations. Defaults to None.
         label_map_file (str, optional): Path to the .json file with the label map. Defaults to None.
         label_map (dict, optional): Dictionary with the label map. Defaults to None.
@@ -96,6 +100,7 @@ class PreProcessingDatasetConfig(BaseModel):
     target_mpp: Optional[float]
     target_mag: Optional[float]
     level: Optional[int]
+    target_mpp_tolerance: Optional[float] = 0.0
 
     # annotation specific settings
     annotation_path: Optional[str]
@@ -174,7 +179,7 @@ def __post_init_post_parse__(self) -> None:
 class LivePatchWSIDataset(Dataset):
     def __init__(
         self,
-        slide_processor_config: PreProcessingDatasetConfig,
+        slide_processor_config: LivePatchWSIConfig,
         logger: logging.Logger = None,
         transforms: Callable = ToTensor(),
     ) -> None:
@@ -184,7 +189,7 @@ def __init__(
         functionality for loading and processing WSIs.
 
         Args:
-            slide_processor_config (PreProcessingDatasetConfig): Configuration for preprocessing the dataset.
+            slide_processor_config (LivePatchWSIConfig): Configuration for preprocessing the dataset.
             logger (logging.Logger, optional): Logger for logging events. Defaults to None.
             transforms (Callable, optional): Transforms to apply to the patches. Defaults to ToTensor().
 
@@ -195,7 +200,7 @@ def __init__(
             wsi_metadata (dict): Metadata of the WSI
             deepzoomgenerator (Union[DeepZoomGeneratorOS, Any]): Class for tile extraction, deepzoom-interface
             tile_extractor (Union[DeepZoomGeneratorOS, Any]): Instance of self.deepzoomgenerator
-            config (PreProcessingDatasetConfig): Configuration for preprocessing the dataset
+            config (LivePatchWSIConfig): Configuration for preprocessing the dataset
             logger (logging.Logger): Logger for logging events
             rescaling_factor (int): Rescaling factor for the slide
             interesting_coords (List[Tuple[int, int, float]]): List of interesting coordinates (patches -> row, col, ratio)
@@ -296,9 +301,6 @@ def _set_tissue_detector(self) -> None:
 
         Raises:
             ImportError: If torch or torchvision cannot be imported.
-
-        Returns:
-            None
         """
         try:
             import torch.nn as nn
@@ -318,7 +320,7 @@ def _set_tissue_detector(self) -> None:
             "cuda:0" if torch.cuda.is_available() else "cpu"
         )
         if self.detector_device == "cpu":
-            logger.warning(
+            self.logger.warning(
                 "No CUDA device detected - Speed may be very slow. Please consider performing extraction on CUDA device or disable tissue detector!"
             )
         model = mobilenet_v3_small().to(device=self.detector_device)
@@ -383,7 +385,7 @@ def _prepare_slide(
                 # Extract the float value
                 if match:
                     slide_mpp = float(match.group(1))
-                    logger.warning(
+                    self.logger.warning(
                         f"MPP {slide_mpp:.4f} was extracted from the comment of the WSI (Tiff-Metadata comment string) - Please check for correctness!"
                     )
                 else:
@@ -414,15 +416,25 @@ def _prepare_slide(
         resulting_mpp = None
 
         if self.config.target_mpp is not None:
-            self.config.downsample, self.rescaling_factor = target_mpp_to_downsample(
-                slide_properties["mpp"],
-                self.config.target_mpp,
-            )
+            if (
+                not slide_properties["mpp"] - self.config.target_mpp_tolerance
+                <= self.config.target_mpp
+                <= slide_properties["mpp"] + self.config.target_mpp_tolerance
+            ):
+                (
+                    self.config.downsample,
+                    self.rescaling_factor,
+                ) = target_mpp_to_downsample(
+                    slide_properties["mpp"],
+                    self.config.target_mpp,
+                )
+            else:
+                self.config.downsample = 1
+                self.rescaling_factor = 1.0
             if self.rescaling_factor != 1.0:
                 resulting_mpp = (
                     slide_properties["mpp"]
                     * self.rescaling_factor
-                    / 2
                     * self.config.downsample
                 )
             else:
@@ -519,7 +531,7 @@ def _prepare_slide(
             )
         self.logger.debug(f"Number of patches sampled: {len(interesting_coords)}")
         if len(interesting_coords) == 0:
-            logger.warning(f"No patches sampled from {self.config.wsi_path}")
+            self.logger.warning(f"No patches sampled from {self.config.wsi_path}")
 
         self.wsi_metadata = {
             "orig_n_tiles_cols": n_cols,
@@ -539,7 +551,7 @@ def _prepare_slide(
 
         return list(interesting_coords), level, polygons_downsampled, region_labels
 
-    def _get_wsi_annotations(self, downsample: int):
+    def _get_wsi_annotations(self, downsample: int):  # TODO: docstring
         region_labels: List[str] = []
         polygons: List[Polygon] = []
         polygons_downsampled: List[Polygon] = []
@@ -610,7 +622,7 @@ def __getitem__(self, index: int) -> Tuple[np.ndarray, dict, np.ndarray]:
             ratio = {}
             patch_mask = np.zeros(
                 (self.res_tile_size, self.res_tile_size), dtype=np.uint8
-            )  # TODO:
+            )
         else:
             intersected_labels, ratio, patch_mask = get_intersected_labels(
                 tile_size=self.res_tile_size,
@@ -634,6 +646,19 @@ def __getitem__(self, index: int) -> Tuple[np.ndarray, dict, np.ndarray]:
                 normalization_vector_path=self.config.normalization_vector_json,
             )
 
+        if self.res_tile_size != self.config.patch_size:
+            image_tile = Image.fromarray(image_tile)
+            if self.res_tile_size > self.config.patch_size:
+                image_tile.thumbnail(
+                    (self.config.patch_size, self.config.patch_size),
+                    getattr(Image, "Resampling", Image).LANCZOS,
+                )
+            else:
+                image_tile = image_tile.resize(
+                    (self.config.patch_size, self.config.patch_size),
+                    getattr(Image, "Resampling", Image).LANCZOS,
+                )
+            image_tile = np.array(image_tile)
         try:
             image_tile = self.transforms(image_tile)
         except TypeError:
@@ -651,24 +676,24 @@ def __getitem__(self, index: int) -> Tuple[np.ndarray, dict, np.ndarray]:
         return image_tile, patch_metadata, patch_mask
 
 
-class PatchWSIDataloader:
-    """Dataloader for PatchWSIDataset
+class LivePatchWSIDataloader:
+    """Dataloader for LivePatchWSIDataset
 
     Args:
-        dataset (PatchWSIDataset): Dataset to load patches from.
+        dataset (LivePatchWSIDataset): Dataset to load patches from.
         batch_size (int): Batch size for the dataloader.
         shuffle (bool, optional): To shuffle iterations. Defaults to False.
         seed (int, optional): Seed for shuffle. Defaults to 42.
     """
 
     def __init__(
         self,
-        dataset: PatchWSIDataset,
+        dataset: LivePatchWSIDataset,
         batch_size: int,
         shuffle: bool = False,
         seed: int = 42,
     ) -> None:
-        assert isinstance(dataset, PatchWSIDataset)
+        assert isinstance(dataset, LivePatchWSIDataset)
         assert isinstance(batch_size, int)
         assert isinstance(shuffle, bool)
         assert isinstance(seed, int)
@@ -682,6 +707,7 @@ def __init__(
         if self.shuffle:
             grtr = np.random.default_rng(seed)
             self.element_list = grtr.permutation(self.element_list)
+        self.i = 0
 
     def __iter__(self):
         self.i = 0
@@ -732,4 +758,21 @@ def __next__(self) -> Tuple[torch.Tensor, List[dict], List[np.ndarray]]:
             raise StopIteration
 
     def __len__(self):
-        return int(np.ceil(len(self.dataset) / self.batch_size))
+        return int(np.ceil((len(self.dataset) - self.discard_count) / self.batch_size))
+
+
+if __name__ == "__main__":
+    """Just for testing purposes"""
+    logger = logging.getLogger(__name__)
+    logger.setLevel(logging.DEBUG)
+    logger.info("Test")
+    config = LivePatchWSIConfig(
+        wsi_path="/Users/fhoerst/Fabian-Projekte/Selocan/RicardoScans/266819.svs",
+        patch_size=256,
+        patch_overlap=0,
+        target_mpp=0.3,
+        target_mpp_tolerance=0.1,
+    )
+    ps_dataset = LivePatchWSIDataset(config, logger)
+    ps_dataloader = LivePatchWSIDataloader(ps_dataset, batch_size=8)
+    ps_dataloader.__next__()

pathopatch/patch_extraction/patch_extraction.py

@@ -837,9 +837,8 @@ def _prepare_wsi(
                 resulting_mpp = (
                     slide_properties["mpp"]
                     * self.rescaling_factor
-                    / 2
                     * self.config.downsample
-                )
+                )  # TODO: should it be divided by 2 or not?
             else:
                 resulting_mpp = slide_properties["mpp"] * self.config.downsample
         # target mag has precedence before downsample!

pathopatch/utils/patch_util.py

@@ -173,7 +173,7 @@ def target_mpp_to_downsample(
                 "We perform rescaling, but this may not be accurate and is very slow!"
             )
             downsample = int(np.floor(target_mpp / base_mpp))
-            rescaling_factor = target_mpp / base_mpp
+            rescaling_factor = target_mpp / (base_mpp * downsample)
     else:
         logger.warning(
             f"Requested mpp resolution ({target_mpp}) is not a power of the base resultion {base_mpp}. "

setup.py

@@ -1,6 +1,6 @@
 from setuptools import find_packages, setup
 
-VERSION = "0.9.3a"
+VERSION = "0.5.4b"
 DESCRIPTION = "PathoPatch - Accelerating Artificial Intelligence Based Whole Slide Image Analysis with an Optimized Preprocessing Pipeline"
 with open("docs/README_pypi.md", "r") as fh:
     LONG_DESCRIPTION = fh.read()

tests/static_test_files/preprocessing/target_mpp_macenko/results/CMU-1-Small-Region/metadata.yaml

@@ -1,5 +1,5 @@
-orig_n_tiles_cols: 2
-orig_n_tiles_rows: 3
+orig_n_tiles_cols: 4
+orig_n_tiles_rows: 5
 base_magnification: 20.0
 downsampling: 2
 label_map:
@@ -9,7 +9,7 @@ patch_size: 256
 base_mpp: 0.499
 target_patch_mpp: 1.2
 stain_normalization: true
-magnification: 4.158333333333334
+magnification: 8.316666666666668
 level: 11
 patch_distribution:
   0: 0

tests/test_core_modules/test_target_mpp_macenko.py

@@ -85,18 +85,18 @@ def test_resulting_patches_wsi(self) -> None:
 
         self.assertEqual(yaml_config, test_file)
 
-    # def test_macenko_patch(self) -> None:
-    #     """Test if Macenko worked correctly"""
-    #     gt_path = (
-    #         self.gt_folder / self.wsi_name / "patches" / "CMU-1-Small-Region_1_1.png"
-    #     )
-    #     gt_image = np.array(Image.open(gt_path.resolve()))
+    def test_macenko_patch(self) -> None:
+        """Test if Macenko worked correctly"""
+        gt_path = (
+            self.gt_folder / self.wsi_name / "patches" / "CMU-1-Small-Region_1_1.png"
+        )
+        gt_image = np.array(Image.open(gt_path.resolve()))
 
-    #     test_path = (
-    #         self.slide_processor.config.output_path
-    #         / self.wsi_name
-    #         / "patches"
-    #         / "CMU-1-Small-Region_1_1.png"
-    #     )
-    #     test_image = np.array(Image.open(test_path.resolve()))
-    #     assert_almost_equal(test_image, gt_image)
+        test_path = (
+            self.slide_processor.config.output_path
+            / self.wsi_name
+            / "patches"
+            / "CMU-1-Small-Region_1_1.png"
+        )
+        test_image = np.array(Image.open(test_path.resolve()))
+        assert_almost_equal(test_image, gt_image)