[REF] use custom patching implementation and use Apache 2.0 license (#…

…153)

* rm clam-based patch implementation

This implementation is licensed GPL3 and is incompatible with Apache 2.0
license. We would like to use Apache 2.0 so this commit removes the GPL3
licensed code.

* add histolab for patching impl

* replace histolab with custom impl

* add scikit-image >= 0.20.0

* add custom segment and patch impl

* replace cli with new patch impl

* use new patching impl

* use Apache 2.0 license

We have removed code from CLAM and are no longer bound by GPL. As of
this commit, wsinfer is licensed under Apache 2.0.

* when thresholding, use arr>=thresh

* use binary thresholding in lieu of otsu

* use new cli args in patching

* fix x,y coord creation + sort coords after query

Before this commit, we were missing the bottom and right edge of patches
because the stop of the coord range was set slightly too low. This
commit fixes that, so now all patches are represented.

This commit also sorts the coordinate indexes after shapely tree query,
because it seems that the order query returns is arbitrary. Sorting
that ensures that when we index the coordinates that are contained in
the polygon, those coordinates are in ascending order (with y changing
most rapidly).

* run isort

* mypy ignore skimage.morphology

* update expected number of patches in test-package

* add test patching test of 100px @ 0.5 mpp

* add binary_threshold argument

* remove patches that extend out of the border of the slide

Our patching algorithm ignores patches that go outside of the slide
boundary.

.github/workflows/ci.yml

@@ -69,7 +69,7 @@ jobs:
           test -f results/run_metadata_*.json
           test -f results/patches/JP2K-33003-1.h5
           test -f results/model-outputs/JP2K-33003-1.csv
-          test $(wc -l < results/model-outputs/JP2K-33003-1.csv) -eq 653
+          test $(wc -l < results/model-outputs/JP2K-33003-1.csv) -eq 675
 
   # This is run on multiple operating systems.
   test-package:
@@ -105,7 +105,7 @@ jobs:
           test -f results/run_metadata_*.json
           test -f results/patches/JP2K-33003-1.h5
           test -f results/model-outputs/JP2K-33003-1.csv
-          test $(wc -l < results/model-outputs/JP2K-33003-1.csv) -eq 653
+          test $(wc -l < results/model-outputs/JP2K-33003-1.csv) -eq 675
       # FIXME: tissue segmentation has different outputs on Windows. The patch sizes
       # are the same but the coordinates found are different.
       - name: Run 'wsinfer run' on Windows
@@ -119,7 +119,7 @@ jobs:
           Test-Path -Path results/run_metadata_*.json -PathType Leaf
           Test-Path -Path results/patches/JP2K-33003-1.h5 -PathType Leaf
           Test-Path -Path results/model-outputs/JP2K-33003-1.csv -PathType Leaf
-          # test $(python -c "print(sum(1 for _ in open('results/model-outputs/JP2K-33003-1.csv')))") -eq 653
+          # test $(python -c "print(sum(1 for _ in open('results/model-outputs/JP2K-33003-1.csv')))") -eq 675
 
   style-and-types:
     runs-on: ubuntu-latest

setup.cfg

@@ -6,15 +6,14 @@ author_email = jakub.kaczmarzyk@stonybrookmedicine.edu
 description = Run patch-based classification on pathology whole slide images.
 long_description = file: README.md
 long_description_content_type = text/markdown
-license = GNU General Public License v3 (GPLv3)
 license_file = LICENSE
 classifiers =
     Development Status :: 4 - Beta
     Environment :: Console
     Intended Audience :: Developers
     Intended Audience :: Healthcare Industry
     Intended Audience :: Science/Research
-    License :: OSI Approved :: GNU General Public License v3 (GPLv3)
+    License :: OSI Approved :: Apache Software License
     Operating System :: OS Independent
     Programming Language :: Python :: 3
     Programming Language :: Python :: 3 :: Only
@@ -38,6 +37,7 @@ install_requires =
     pandas
     pillow
     pyyaml
+    scikit-image>=0.20.0
     shapely
     tifffile
     tiffslide
@@ -101,6 +101,8 @@ ignore_missing_imports = True
 ignore_missing_imports = True
 [mypy-shapely.*]
 ignore_missing_imports = True
+[mypy-skimage.morphology]
+ignore_missing_imports = True
 [mypy-tifffile]
 ignore_missing_imports = True
 [mypy-zarr.storage]

tests/reference/pancancer-lymphocytes-inceptionv4.tcga/purple.csv

@@ -19,7 +19,6 @@ minx,miny,width,height,prob_notils,prob_tils
 0,3400,200,200,1.0,3.427372535086404e-12
 0,3600,200,200,1.0,3.427372535086404e-12
 0,3800,200,200,1.0,3.427372535086404e-12
-0,4000,200,200,0.9999136924743652,8.626562339486554e-05
 200,0,200,200,1.0,3.427372535086404e-12
 200,200,200,200,1.0,3.427372535086404e-12
 200,400,200,200,1.0,3.427372535086404e-12
@@ -40,7 +39,6 @@ minx,miny,width,height,prob_notils,prob_tils
 200,3400,200,200,1.0,3.427372535086404e-12
 200,3600,200,200,1.0,3.427372535086404e-12
 200,3800,200,200,1.0,3.427372535086404e-12
-200,4000,200,200,0.9999136924743652,8.626562339486554e-05
 400,0,200,200,1.0,3.427372535086404e-12
 400,200,200,200,1.0,3.427372535086404e-12
 400,400,200,200,1.0,3.427372535086404e-12
@@ -61,7 +59,6 @@ minx,miny,width,height,prob_notils,prob_tils
 400,3400,200,200,1.0,3.427372535086404e-12
 400,3600,200,200,1.0,3.427372535086404e-12
 400,3800,200,200,1.0,3.427372535086404e-12
-400,4000,200,200,0.9999136924743652,8.626562339486554e-05
 600,0,200,200,1.0,3.427372535086404e-12
 600,200,200,200,1.0,3.427372535086404e-12
 600,400,200,200,1.0,3.427372535086404e-12
@@ -82,7 +79,6 @@ minx,miny,width,height,prob_notils,prob_tils
 600,3400,200,200,1.0,3.427372535086404e-12
 600,3600,200,200,1.0,3.427372535086404e-12
 600,3800,200,200,1.0,3.427372535086404e-12
-600,4000,200,200,0.9999136924743652,8.626562339486554e-05
 800,0,200,200,1.0,3.427372535086404e-12
 800,200,200,200,1.0,3.427372535086404e-12
 800,400,200,200,1.0,3.427372535086404e-12
@@ -103,7 +99,6 @@ minx,miny,width,height,prob_notils,prob_tils
 800,3400,200,200,1.0,3.427372535086404e-12
 800,3600,200,200,1.0,3.427372535086404e-12
 800,3800,200,200,1.0,3.427372535086404e-12
-800,4000,200,200,0.9999136924743652,8.626562339486554e-05
 1000,0,200,200,1.0,3.427372535086404e-12
 1000,200,200,200,1.0,3.427372535086404e-12
 1000,400,200,200,1.0,3.427372535086404e-12
@@ -124,7 +119,6 @@ minx,miny,width,height,prob_notils,prob_tils
 1000,3400,200,200,1.0,3.427372535086404e-12
 1000,3600,200,200,1.0,3.427372535086404e-12
 1000,3800,200,200,1.0,3.427372535086404e-12
-1000,4000,200,200,0.9999136924743652,8.626562339486554e-05
 1200,0,200,200,1.0,3.427372535086404e-12
 1200,200,200,200,1.0,3.427372535086404e-12
 1200,400,200,200,1.0,3.427372535086404e-12
@@ -145,7 +139,6 @@ minx,miny,width,height,prob_notils,prob_tils
 1200,3400,200,200,1.0,3.427372535086404e-12
 1200,3600,200,200,1.0,3.427372535086404e-12
 1200,3800,200,200,1.0,3.427372535086404e-12
-1200,4000,200,200,0.9999136924743652,8.626562339486554e-05
 1400,0,200,200,1.0,3.427372535086404e-12
 1400,200,200,200,1.0,3.427372535086404e-12
 1400,400,200,200,1.0,3.427372535086404e-12
@@ -166,7 +159,6 @@ minx,miny,width,height,prob_notils,prob_tils
 1400,3400,200,200,1.0,3.427372535086404e-12
 1400,3600,200,200,1.0,3.427372535086404e-12
 1400,3800,200,200,1.0,3.427372535086404e-12
-1400,4000,200,200,0.9999136924743652,8.626562339486554e-05
 1600,0,200,200,1.0,3.427372535086404e-12
 1600,200,200,200,1.0,3.427372535086404e-12
 1600,400,200,200,1.0,3.427372535086404e-12
@@ -187,7 +179,6 @@ minx,miny,width,height,prob_notils,prob_tils
 1600,3400,200,200,1.0,3.427372535086404e-12
 1600,3600,200,200,1.0,3.427372535086404e-12
 1600,3800,200,200,1.0,3.427372535086404e-12
-1600,4000,200,200,0.9999136924743652,8.626562339486554e-05
 1800,0,200,200,1.0,3.427372535086404e-12
 1800,200,200,200,1.0,3.427372535086404e-12
 1800,400,200,200,1.0,3.427372535086404e-12
@@ -208,7 +199,6 @@ minx,miny,width,height,prob_notils,prob_tils
 1800,3400,200,200,1.0,3.427372535086404e-12
 1800,3600,200,200,1.0,3.427372535086404e-12
 1800,3800,200,200,1.0,3.427372535086404e-12
-1800,4000,200,200,0.9999136924743652,8.626562339486554e-05
 2000,0,200,200,1.0,3.427372535086404e-12
 2000,200,200,200,1.0,3.427372535086404e-12
 2000,400,200,200,1.0,3.427372535086404e-12
@@ -229,7 +219,6 @@ minx,miny,width,height,prob_notils,prob_tils
 2000,3400,200,200,1.0,3.427372535086404e-12
 2000,3600,200,200,1.0,3.427372535086404e-12
 2000,3800,200,200,1.0,3.427372535086404e-12
-2000,4000,200,200,0.9999136924743652,8.626562339486554e-05
 2200,0,200,200,1.0,3.427372535086404e-12
 2200,200,200,200,1.0,3.427372535086404e-12
 2200,400,200,200,1.0,3.427372535086404e-12
@@ -250,7 +239,6 @@ minx,miny,width,height,prob_notils,prob_tils
 2200,3400,200,200,1.0,3.427372535086404e-12
 2200,3600,200,200,1.0,3.427372535086404e-12
 2200,3800,200,200,1.0,3.427372535086404e-12
-2200,4000,200,200,0.9999136924743652,8.626562339486554e-05
 2400,0,200,200,1.0,3.427372535086404e-12
 2400,200,200,200,1.0,3.427372535086404e-12
 2400,400,200,200,1.0,3.427372535086404e-12
@@ -271,7 +259,6 @@ minx,miny,width,height,prob_notils,prob_tils
 2400,3400,200,200,1.0,3.427372535086404e-12
 2400,3600,200,200,1.0,3.427372535086404e-12
 2400,3800,200,200,1.0,3.427372535086404e-12
-2400,4000,200,200,0.9999136924743652,8.626562339486554e-05
 2600,0,200,200,1.0,3.427372535086404e-12
 2600,200,200,200,1.0,3.427372535086404e-12
 2600,400,200,200,1.0,3.427372535086404e-12
@@ -292,7 +279,6 @@ minx,miny,width,height,prob_notils,prob_tils
 2600,3400,200,200,1.0,3.427372535086404e-12
 2600,3600,200,200,1.0,3.427372535086404e-12
 2600,3800,200,200,1.0,3.427372535086404e-12
-2600,4000,200,200,0.9999136924743652,8.626562339486554e-05
 2800,0,200,200,1.0,3.427372535086404e-12
 2800,200,200,200,1.0,3.427372535086404e-12
 2800,400,200,200,1.0,3.427372535086404e-12
@@ -313,7 +299,6 @@ minx,miny,width,height,prob_notils,prob_tils
 2800,3400,200,200,1.0,3.427372535086404e-12
 2800,3600,200,200,1.0,3.427372535086404e-12
 2800,3800,200,200,1.0,3.427372535086404e-12
-2800,4000,200,200,0.9999136924743652,8.626562339486554e-05
 3000,0,200,200,1.0,3.427372535086404e-12
 3000,200,200,200,1.0,3.427372535086404e-12
 3000,400,200,200,1.0,3.427372535086404e-12
@@ -334,7 +319,6 @@ minx,miny,width,height,prob_notils,prob_tils
 3000,3400,200,200,1.0,3.427372535086404e-12
 3000,3600,200,200,1.0,3.427372535086404e-12
 3000,3800,200,200,1.0,3.427372535086404e-12
-3000,4000,200,200,0.9999136924743652,8.626562339486554e-05
 3200,0,200,200,1.0,3.427372535086404e-12
 3200,200,200,200,1.0,3.427372535086404e-12
 3200,400,200,200,1.0,3.427372535086404e-12
@@ -355,7 +339,6 @@ minx,miny,width,height,prob_notils,prob_tils
 3200,3400,200,200,1.0,3.427372535086404e-12
 3200,3600,200,200,1.0,3.427372535086404e-12
 3200,3800,200,200,1.0,3.427372535086404e-12
-3200,4000,200,200,0.9999136924743652,8.626562339486554e-05
 3400,0,200,200,1.0,3.427372535086404e-12
 3400,200,200,200,1.0,3.427372535086404e-12
 3400,400,200,200,1.0,3.427372535086404e-12
@@ -376,7 +359,6 @@ minx,miny,width,height,prob_notils,prob_tils
 3400,3400,200,200,1.0,3.427372535086404e-12
 3400,3600,200,200,1.0,3.427372535086404e-12
 3400,3800,200,200,1.0,3.427372535086404e-12
-3400,4000,200,200,0.9999136924743652,8.626562339486554e-05
 3600,0,200,200,1.0,3.427372535086404e-12
 3600,200,200,200,1.0,3.427372535086404e-12
 3600,400,200,200,1.0,3.427372535086404e-12
@@ -397,7 +379,6 @@ minx,miny,width,height,prob_notils,prob_tils
 3600,3400,200,200,1.0,3.427372535086404e-12
 3600,3600,200,200,1.0,3.427372535086404e-12
 3600,3800,200,200,1.0,3.427372535086404e-12
-3600,4000,200,200,0.9999136924743652,8.626562339486554e-05
 3800,0,200,200,1.0,3.427372535086404e-12
 3800,200,200,200,1.0,3.427372535086404e-12
 3800,400,200,200,1.0,3.427372535086404e-12
@@ -418,25 +399,3 @@ minx,miny,width,height,prob_notils,prob_tils
 3800,3400,200,200,1.0,3.427372535086404e-12
 3800,3600,200,200,1.0,3.427372535086404e-12
 3800,3800,200,200,1.0,3.427372535086404e-12
-3800,4000,200,200,0.9999136924743652,8.626562339486554e-05
-4000,0,200,200,0.9996557235717772,0.0003442850720603
-4000,200,200,200,0.9996557235717772,0.0003442850720603
-4000,400,200,200,0.9996557235717772,0.0003442850720603
-4000,600,200,200,0.9996557235717772,0.0003442850720603
-4000,800,200,200,0.9996557235717772,0.0003442850720603
-4000,1000,200,200,0.9996557235717772,0.0003442850720603
-4000,1200,200,200,0.9996557235717772,0.0003442850720603
-4000,1400,200,200,0.9996557235717772,0.0003442850720603
-4000,1600,200,200,0.9996557235717772,0.0003442850720603
-4000,1800,200,200,0.9996557235717772,0.0003442850720603
-4000,2000,200,200,0.9996557235717772,0.0003442850720603
-4000,2200,200,200,0.9996557235717772,0.0003442850720603
-4000,2400,200,200,0.9996557235717772,0.0003442850720603
-4000,2600,200,200,0.9996557235717772,0.0003442850720603
-4000,2800,200,200,0.9996557235717772,0.0003442850720603
-4000,3000,200,200,0.9996557235717772,0.0003442850720603
-4000,3200,200,200,0.9996557235717772,0.0003442850720603
-4000,3400,200,200,0.9996557235717772,0.0003442850720603
-4000,3600,200,200,0.9996557235717772,0.0003442850720603
-4000,3800,200,200,0.9996557235717772,0.0003442850720603
-4000,4000,200,200,0.9894591569900512,0.0105408066883683

tests/test_all.py

@@ -291,7 +291,7 @@ def test_convert_to_sbu():
 
 @pytest.mark.parametrize(
     ["patch_size", "patch_spacing"],
-    [(256, 0.25), (256, 0.50), (350, 0.25), (100, 0.3)],
+    [(256, 0.25), (256, 0.50), (350, 0.25), (100, 0.3), (100, 0.5)],
 )
 def test_patch_cli(
     patch_size: int, patch_spacing: float, tmp_path: Path, tiff_image: Path
@@ -306,22 +306,20 @@ def test_patch_cli(
         cli,
         [
             "patch",
-            "--source",
+            "--wsi-dir",
             str(tiff_image.parent),
-            "--save-dir",
+            "--results-dir",
             str(savedir),
-            "--patch-size",
+            "--patch-size-px",
             str(patch_size),
-            "--patch-spacing",
+            "--patch-spacing-um-px",
             str(patch_spacing),
         ],
     )
     assert result.exit_code == 0
     stem = tiff_image.stem
     assert (savedir / "masks" / f"{stem}.jpg").exists()
     assert (savedir / "patches" / f"{stem}.h5").exists()
-    assert (savedir / "process_list_autogen.csv").exists()
-    assert (savedir / "stitches" / f"{stem}.jpg").exists()
 
     expected_patch_size = round(patch_size * patch_spacing / orig_slide_spacing)
     sqrt_expected_num_patches = round(orig_slide_size / expected_patch_size)

wsinfer/cli/infer.py

@@ -24,7 +24,7 @@
 
 from ..modellib import models
 from ..modellib.run_inference import run_inference
-from ..patchlib.create_patches_fp import create_patches
+from ..patchlib import segment_and_patch_directory_of_slides
 
 
 def _num_cpus() -> int:
@@ -350,18 +350,20 @@ def run(
 
     click.secho("\nFinding patch coordinates...\n", fg="green")
 
-    create_patches(
-        source=str(wsi_dir),
-        save_dir=str(results_dir),
-        patch_size=model_obj.config.patch_size_pixels,
-        patch_spacing=model_obj.config.spacing_um_px,
-        seg=True,
-        patch=True,
-        # Stitching is a bottleneck when using tiffslide.
-        # TODO: figure out why this is...
-        stitch=False,
-        # FIXME: allow customization of this preset
-        preset="tcga.csv",
+    # FIXME: add presets for different tissue types?
+
+    segment_and_patch_directory_of_slides(
+        wsi_dir=wsi_dir,
+        save_dir=results_dir,
+        patch_size_px=model_obj.config.patch_size_pixels,
+        patch_spacing_um_px=model_obj.config.spacing_um_px,
+        thumbsize=(2048, 2048),
+        # TODO: these can be made arguments to the CLI.
+        median_filter_size=7,
+        binary_threshold=7,
+        closing_kernel_size=6,
+        min_object_size_um2=200**2,
+        min_hole_size_um2=190**2,
     )
 
     click.secho("\nRunning model inference.\n", fg="green")