MHub / GC - Add LUNA22 ISMI baseline model implementation

models/gc_luna22_ismi_baseline/config/default.yml

@@ -0,0 +1,36 @@
+general:
+  data_base_dir: /app/data
+  version: 1.0
+  description: LUNA22 ISMI Baseline lung CT nodule malignancy and nodule type classification (dicom to json)
+
+execute:
+- DicomImporter
+- MhaConverter
+- Luna22IsmiBaselineRunner
+- ReportExporter
+- DataOrganizer
+
+modules:
+  DicomImporter:
+    source_dir: input_data
+    import_dir: sorted_data
+    sort_data: True
+    meta: 
+      mod: ct
+
+  MhaConverter:
+    engine: panimg
+
+  ReportExporter:
+    format: compact
+    includes:
+      - data: malignancy_risk
+        label: malignancy_risk
+        value: value
+      - data: texture
+        label: texture
+        value: value
+
+  DataOrganizer:
+    targets:
+      - json:mod=report-->[i:sid]/nodule_predictions.json

models/gc_luna22_ismi_baseline/dockerfiles/Dockerfile

@@ -0,0 +1,45 @@
+FROM mhubai/base:latest
+
+# Update authors label
+LABEL authors="[email protected]"
+
+# Install git-lfs (required for unpacking model weights)
+RUN apt update && apt install -y --no-install-recommends git-lfs && rm -rf /var/lib/apt/lists/*
+
+# build/install CUDA 11 toolkit cudnn 8 and tensorflow 2.11.0 with GPU support (without conda)
+RUN pip3 install --no-cache-dir \
+    nvidia-cuda-runtime-cu11 \
+    nvidia-cusolver-cu11 \
+    nvidia-curand-cu11 \
+    nvidia-cufft-cu11 \
+    nvidia-cublas-cu11 \
+    nvidia-cusparse-cu11 \
+    nvidia-cudnn-cu11 \
+    tensorflow==2.11.0 \
+    nvidia-tensorrt==7.2.3.4 \
+    --extra-index-url https://pypi.ngc.nvidia.com
+
+# Configure required paths for tensorflow with GPU support
+ENV NVIDIA_DIR /usr/local/lib/python3.8/dist-packages/nvidia
+ENV LD_LIBRARY_PATH /usr/local/lib/python3.8/dist-packages/tensorrt:$NVIDIA_DIR/cublas/lib:$NVIDIA_DIR/cuda_runtime/lib:$NVIDIA_DIR/cudnn/lib:$NVIDIA_DIR/cufft/lib:$NVIDIA_DIR/curand/lib:$NVIDIA_DIR/cusolver/lib:$NVIDIA_DIR/cusparse/lib
+
+# Clone the main branch of MHubAI/models TODO enable
+#RUN git stash \
+#    && git fetch https://github.com/MHubAI/models.git main \
+#    && git merge FETCH_HEAD \
+#    && git sparse-checkout set "models/gc_luna22_ismi_baseline" \
+#    && git fetch https://github.com/MHubAI/models.git main
+
+# Install luna22 ismi baseline algorithm and model weights (main branch, commit 68c5e93be62dae54d15a8a3d8821ca2138e0fe2e)
+RUN git clone https://github.com/DIAGNijmegen/bodyct-luna22-ismi-algorithm-baseline.git /opt/algorithm && \
+    cd /opt/algorithm && git reset --hard 68c5e93be62dae54d15a8a3d8821ca2138e0fe2e
+
+# Install algorithm requirements
+RUN pip3 install --no-cache-dir evalutils==0.3.1
+
+# Set PYTHONPATH to include algorithm code
+ENV PYTHONPATH "/app:/opt/algorithm"
+
+# Default run script
+ENTRYPOINT ["python3", "-m", "mhubio.run"]
+CMD ["--config", "/app/models/gc_luna22_ismi_baseline/config/default.yml"]

models/gc_luna22_ismi_baseline/utils/Luna22IsmiBaselineRunner.py

@@ -0,0 +1,73 @@
+"""
+------------------------------------------------------------------
+Mhub / DIAG - Run Module for the GC Luna22 ISMI baseline Algorithm
+------------------------------------------------------------------
+
+------------------------------------------------------------------
+Author: Sil van de Leemput
+Email:  [email protected]
+------------------------------------------------------------------
+"""
+
+import json
+from pathlib import Path
+from typing import Optional
+
+from mhubio.core import Instance, InstanceData, IO, Module, ValueOutput, ClassOutput, Meta
+
+import SimpleITK
+
+import tensorflow as tf
+
+# Import the Luna22 baseline algorithm class from the luna22-ismi-algorithm repository
+from process import Nodule_classifier as NoduleClassifier
+
+
+@ValueOutput.Name('malignancy_risk')
+@ValueOutput.Meta(Meta(key="value"))
+@ValueOutput.Label('MalignancyRisk')
+@ValueOutput.Type(float)
+@ValueOutput.Description('Probability of the nodule malignancy.')
+class MalignancyRiskOutput(ValueOutput):
+    pass
+
+
+@ClassOutput.Name('texture')
+@ClassOutput.Label('NoduleType')
+@ClassOutput.Description('Prediction of the nodule type.')
+@ClassOutput.Class(0, 'Non-solid', 'Non-solid.')
+@ClassOutput.Class(1, 'Part-solid', 'Part-solid.')
+@ClassOutput.Class(2, 'Solid', 'Solid.')
+class NoduleTypeOutput(ClassOutput):
+    pass
+
+
+class Luna22IsmiBaselineRunner(Module):
+
+    _cached_classifier: Optional[NoduleClassifier] = None
+
+    @IO.Instance()
+    @IO.Input('in_data', 'mha|nrrd|nifti:mod=ct', the='input volume centered around a nodule from a lung CT image')
+    @IO.OutputData('malignancy_risk', MalignancyRiskOutput, data='in_data', the='Luna22 Ismi nodule malignancy probability')
+    @IO.OutputData('texture', NoduleTypeOutput, data='in_data', the='Luna22 Ismi nodule type classification')
+    def task(self, instance: Instance, in_data: InstanceData, malignancy_risk: MalignancyRiskOutput, texture: NoduleTypeOutput) -> None:
+        assert len(tf.config.list_physical_devices("GPU")) >= 1, \
+            "Error: Luna22IsmiBaselineRunner must be run on a GPU! Because: " \
+            "The Conv2D op currently only supports the NHWC tensor format on" \
+            " the CPU. The op was given the format: NCHW [Op:Conv2D]"
+
+        # Read input image
+        input_image = SimpleITK.ReadImage(in_data.abspath)
+
+        # Create classifier and load model weights if not already loaded
+        if self._cached_classifier is None:
+            self.v("Luna22IsmiBaselineRunner - Loading and caching model weights")
+            self._cached_classifier = NoduleClassifier()
+        classifier = self._cached_classifier
+
+        # Run the classifier on the input image
+        predictions = classifier.predict(input_image=input_image)
+
+        # Output the predicted values
+        malignancy_risk.value = predictions["malignancy_risk"]
+        texture.value = predictions["texture"]

models/gc_luna22_ismi_baseline/utils/__init__.py

@@ -0,0 +1 @@
+from Luna22IsmiBaselineRunner import *