Merge pull request #1 from frankkramer-lab/rework_stain_normalization

Rework stain normalization

.github/workflows/build-package.yml

@@ -0,0 +1,36 @@
+# This workflow will install Python dependencies and run tests with a variety of Python versions
+# For more information see: https://help.github.com/actions/language-and-framework-guides/using-python-with-github-actions
+# CUSTOM Modifications by Dominik Müller (2022)
+
+name: Python Package - Build
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+jobs:
+  requirements:
+    name: Requirements
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ['3.8', '3.10']
+    steps:
+    - uses: actions/checkout@v2
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        python -m pip install pytest
+        sudo apt -y install libvips-dev
+        if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
+    - name: Test with pytest
+      run: |
+        git lfs pull
+        pytest -v

.github/workflows/release.yml

@@ -0,0 +1,89 @@
+#==============================================================================#
+#  Author:       Dominik Müller                                                #
+#  Copyright:    2024 AG-RAIMIA-Müller, University of Augsburg,                #
+#                University of Augsburg                                        #
+#                                                                              #
+#  This program is free software: you can redistribute it and/or modify        #
+#  it under the terms of the GNU General Public License as published by        #
+#  the Free Software Foundation, either version 3 of the License, or           #
+#  (at your option) any later version.                                         #
+#                                                                              #
+#  This program is distributed in the hope that it will be useful,             #
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of              #
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
+#  GNU General Public License for more details.                                #
+#                                                                              #
+#  You should have received a copy of the GNU General Public License           #
+#  along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
+#==============================================================================#
+# This workflow will:
+# - create a CHANGELOG between latest tag and current tag
+# - release the package on GitHub
+# - dockerize the package on GitHub (GitHub Container Registry)
+# - upload it to PyPI
+
+name: Package Release
+on:
+  push:
+    tags:
+      - 'v[0-9]+.[0-9]+'
+
+jobs:
+  release:
+    name: Release on GitHub
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v2
+
+      - name: Create CHANGELOG
+        id: changelog
+        uses: Requarks/changelog-action@v1
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          tag: ${{ github.ref_name }}
+          excludeTypes: other,style
+
+      - name: Create Release
+        uses: actions/create-release@v1
+        with:
+          tag_name: ${{ github.ref }}
+          release_name: aucmedi_${{ github.ref }}
+          body: ${{ steps.changelog.outputs.changes }}
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  dockerize:
+    name: Dockerize on GitHub
+    runs-on: ubuntu-latest
+    env:
+      REGISTRY: ghcr.io
+      IMAGE_NAME: ${{ github.repository }}
+    permissions:
+      contents: read
+      packages: write
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v3
+
+      - name: Log in to the Container registry
+        uses: docker/login-action@v2
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata (tags, labels) for Docker
+        id: meta
+        uses: docker/metadata-action@v4
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+
+      - name: Build and push Docker image
+        uses: docker/build-push-action@v3
+        with:
+          context: .
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}

Dockerfile

@@ -0,0 +1,59 @@
+#==============================================================================#
+#  Author:       Dominik Müller                                                #
+#  Copyright:    2024 AG-RAIMIA-Müller, University of Augsburg,                #
+#                University of Augsburg                                        #
+#                                                                              #
+#  This program is free software: you can redistribute it and/or modify        #
+#  it under the terms of the GNU General Public License as published by        #
+#  the Free Software Foundation, either version 3 of the License, or           #
+#  (at your option) any later version.                                         #
+#                                                                              #
+#  This program is distributed in the hope that it will be useful,             #
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of              #
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
+#  GNU General Public License for more details.                                #
+#                                                                              #
+#  You should have received a copy of the GNU General Public License           #
+#  along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
+#==============================================================================#
+#-----------------------------------------------------#
+#              Information & System Base              #
+#-----------------------------------------------------#
+# Base image
+FROM tensorflow/tensorflow:latest-gpu
+
+# Meta information
+LABEL authors="Dominik Müller"
+LABEL contact="dominik.mueller@informatik.uni-augsburg.de"
+LABEL repository="https://github.com/frankkramer-lab/DeepGleason"
+LABEL license="GNU General Public License v3.0"
+
+#-----------------------------------------------------#
+#                        Setup                        #
+#-----------------------------------------------------#
+# Setup system environment variables
+ENV DEBIAN_FRONTEND=noninteractive
+ENV PIP_ROOT_USER_ACTION=ignore
+
+# Copy git repository into container
+ADD . /root/DeepGleason
+
+# Install required software dependencies (cv2)
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends python3-opencv && \
+    apt-get clean && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
+
+# Update Python pip
+RUN python -m pip install pip --upgrade
+
+# Install AUCMEDI from local git repo
+RUN pip install -r /root/DeepGleason/requirements.txt
+
+# Create working directory
+VOLUME ["/data"]
+WORKDIR /data
+
+#-----------------------------------------------------#
+#                       Startup                       #
+#-----------------------------------------------------#
+ENTRYPOINT ["/root/DeepGleason/code/main.py"]

README.md

@@ -4,13 +4,22 @@ Welcome to DeepGleason, an advanced software solution developed for inferring Gl
 
 ![viz](docs/viz.png)
 
-It computes a one of the following 7 classes per 1024 x 1024 tile: 
+It computes a one of the following 6 classes per 1024 x 1024 tile: 
 Artifact - Sponge, Artifact Dirt, PIN, precursor or unclear tissue, Regular Tissue, Gleason 3, Gleason 4, Gleason 5
 
 ![viz](docs/samples.png)
 
 ## Usage
 
+**Example:**  
+```sh
+cd DeepGleason/
+python code/main.py   --input /sandbox/9f3fae7a6ff12c76fbbf89b9c50c66b6.ome.tiff \
+                      --output /sandbox/ \
+                      --model models/model.ConvNeXtBase.hdf5 \
+                      -p /sandbox/predictions.csv
+```
+
 **Input**: 
 - Whole-slide image of prostata cancer
 - DeepGleason deep neural network model
@@ -20,7 +29,7 @@ Artifact - Sponge, Artifact Dirt, PIN, precursor or unclear tissue, Regular Tiss
 - CSV file containing class predictions & confidence for each tile
 
 ```sh
-usage: scripts/pred_cli.py [-h] [-g GPU] [--cache CACHE] -i INPUT [-o OUTPUT] [--model MODEL] [--generate_overlay] [-p PREDICTION]
+usage: code/main.py [-h] [-g GPU] [--cache CACHE] -i INPUT [-o OUTPUT] [--model MODEL] [--generate_overlay] [-p PREDICTION]
 
 DeepGleason: Prediction
 
@@ -60,7 +69,7 @@ This program utilizes the PyVIPS library to load and store images and AUCMEDI to
 The implemented medical image classification pipeline can be summarized in the following core steps:
 - Slide preparation with PyVIPS: 1024x1024 pixel tiles
 - Tile preprocessing: Padding, resize, pixel intensity normalization
-- Deep Learning model architecture: ResNeXt101
+- Deep Learning model architecture: ConvNeXt
 - Slide reconstruction to bigTIFF overlay
 
 ![workflow](docs/workflow.png)

code/main.py

@@ -1,6 +1,6 @@
 # =============================================================================#
 #  Author:       Dominik Müller, Philip Meyer                                  #
-#  Copyright:    2023 AG-RAIMIA-Müller, University of Augsburg                 #
+#  Copyright:    2024 AG-RAIMIA-Müller, University of Augsburg                 #
 #                                                                              #
 #  This program is free software: you can redistribute it and/or modify        #
 #  it under the terms of the GNU General Public License as published by        #
@@ -131,7 +131,7 @@
 
 
 config = {}
-COL_NAMES = ["A_S", "A_D", "R", "Q", "G3", "G4", "G5"]
+COL_NAMES = ["A_S", "A_D", "R", "G3", "G4", "G5"]
 
 # pyvips.cache_set_max_mem(0) #This may be necessary to cache operations. 
 # On the other hand this is incredibly useful to accelerate null computations

code/model.py

@@ -1,6 +1,6 @@
 # =============================================================================#
 #  Author:       Dominik Müller, Philip Meyer                                  #
-#  Copyright:    2023 AG-RAIMIA-Müller, University of Augsburg                 #
+#  Copyright:    2024 AG-RAIMIA-Müller, University of Augsburg                 #
 #                                                                              #
 #  This program is free software: you can redistribute it and/or modify        #
 #  it under the terms of the GNU General Public License as published by        #
@@ -20,35 +20,42 @@
 # -----------------------------------------------------#
 import pandas as pd
 import tensorflow as tf
-
+from PIL import Image
+import os
 # AUCMEDI libraries
 from aucmedi import DataGenerator, NeuralNetwork
 from aucmedi.data_processing.subfunctions import Padding
+from stain_normalization import StainNormalization
 
 # -----------------------------------------------------#
 #                   AUCMEDI Pipeline                   #
 # -----------------------------------------------------#
-COL_NAMES = ["A_S", "A_D", "R", "Q", "G3", "G4", "G5"]
+COL_NAMES = ["A_S", "A_D", "R", "G3", "G4", "G5"]
 
 def run_aucmedi(x, architecture, config):
+    # Load stain normalization
+    dir_path = os.path.dirname(os.path.realpath(__file__))
+    path_stain_target = os.path.join(dir_path, "stainnormalize_target.png")
+    target_image = Image.open(path_stain_target)
     # Define Subfunctions
-    sf_list = [Padding(mode="square")]
-    # Set activation output to softmax for multi-class classification
-    activation_output = "softmax"
+    sf_list = [Padding(mode="square"), StainNormalization(target_image)]
+
+    # identify architecture
+    arch_name = architecture.split(".")[-2]
 
     # Initialize model
     model = NeuralNetwork(
         config["nclasses"],
         channels=3,
-        architecture="2D.ResNeXt101",
-        multiprocessing=False,
-        activation_output=activation_output,
+        architecture="2D." + arch_name,
+        workers=16, 
+        multiprocessing=True,
     )
 
-    # Dump latest model
-    model.load(architecture)
+    # Load model
+    model.model.load_weights(architecture)
 
-    # Initialize training and validation Data Generators
+    # Initialize Data Generator
     gen = DataGenerator(
         x,
         config["path_images"],
@@ -62,6 +69,7 @@ def run_aucmedi(x, architecture, config):
         sample_weights=None,
         seed=123,
         image_format=config["image_format"],
+        workers=6,
     )
 
     # generate predictions

code/proc.py

@@ -1,6 +1,6 @@
 # =============================================================================#
 #  Author:       Dominik Müller, Philip Meyer                                  #
-#  Copyright:    2023 AG-RAIMIA-Müller, University of Augsburg                 #
+#  Copyright:    2024 AG-RAIMIA-Müller, University of Augsburg                 #
 #                                                                              #
 #  This program is free software: you can redistribute it and/or modify        #
 #  it under the terms of the GNU General Public License as published by        #
@@ -21,7 +21,7 @@
 import os
 import pyvips
 
-os.environ["VIPS_CONCURRENCY"] = "64"
+os.environ["VIPS_CONCURRENCY"] = "0"
 import numpy as np
 import pandas as pd
 import math
@@ -30,12 +30,14 @@
 #------------------------------------------------------#
 #             Processing Utility Functions             #
 #------------------------------------------------------#
-def eval_cb(image, progress):
-    print(f"\reval: percent = {progress.percent}", end="\n")
-
+def eval_handler(image, progress):
+    print('run time so far (secs) = {}'.format(progress.run))
+    print('estimated time of arrival (secs) = {}'.format(progress.eta))
+    print('total number of pels to process = {}'.format(progress.tpels))
+    print('number of pels processed so far = {}'.format(progress.npels))
+    print('percent complete = {}'.format(progress.percent))
 
 def gen_tiles(patch_path, slide, name, PATCH_SIZE):
-    os.environ["VIPS_CONCURRENCY"] = "128"
     props = {
         "compression": "jpeg",
         "xres": 4000,
@@ -53,9 +55,8 @@ def gen_tiles(patch_path, slide, name, PATCH_SIZE):
     img = img_r.bandjoin([img_g, img_b])
     img = img.copy(interpretation="rgb")
 
-    img.set_progress(True)
-    # img.signal_connect('preeval', preeval_cb)
-    img.signal_connect("eval", eval_cb)
+    #img.set_progress(True)
+    #img.signal_connect("eval", eval_handler)
 
     width = img.width
     height = img.height
@@ -95,14 +96,12 @@ def class_reassemble(max_X, max_Y, slide_name, df_res, PATCH_SIZE):
                 small_version[x, y] = [0, 0, 0]  # White
             elif samp == "R":
                 small_version[x, y] = [0, 1, 0]  # Green
-            elif samp == "Q":
-                small_version[x, y] = [0, 1, 1]  # Cyan
             elif samp == "G3":
-                small_version[x, y] = [1, 1, 0]  # Yellow
+                small_version[x, y] = [1, 1, 0]     # Yellow
             elif samp == "G4":
-                small_version[x, y] = [1, 0.5, 0]  # Orange
+                small_version[x, y] = [1, 0.5, 0]   # Orange
             elif samp == "G5":
-                small_version[x, y] = [1, 0, 0]  # Red
+                small_version[x, y] = [1, 0, 0]     # Red
 
             cntr.update(1)
     del cntr