About The Project

This repository contains the code for the publication titled "Metastatic neuroendocrine tumors with unknown site of the primary: a machine-learning based approach to a complex scenario".

Installation

1. Clone the repository.

   git clone https://github.com/IPMI-ICNS-UKE/NET_CUP

2. Install the package and necessary dependencies.

   pip install -e .
   pip install -r requirements.txt

3. Download data:
   • Download the required CSV files and feature vector from Zenodo and extract them into the data/ directory.

Additional Steps for Generating New Feature Vectors

To respect data privacy, only feature vectors generated from random patches of whole-slide images (WSIs) are publicly accessible. The original WSIs can be provided upon reasonable request.

1. Download ResNet weights:
   • Download the MTDP ResNet weights from MTDP repository.
   • Download the RetCCL ResNet weights from RetCCL repository under the "Pre-trained models for histopathological image tasks" section.
   • Rename the downloaded files to mtdp.pth and retccl.pth and place them into weights/.
2. Copy WSIs along with their .geojson segmentation files into data/external_dataset/ and data/uke_dataset/.

Usage

Reproducing results

To reproduce the results from the publication, run the Jupyter Notebooks available in src/NET_CUP/experiments. Within each notebook, you can select different pretrained ResNet backbones and classifiers. Configure these settings in the notebook's "Settings" section by changing the classifier variable or setting the feature_type variable to one of the following options as described in the publication:

• FeatureType.IMAGENET
• FeatureType.RETCCL
• FeatureType.MTDP

Generating new feature vectors

After acquiring access to the original WSIs and following the Installation steps, you can generate new feature vectors:

python src/NET_CUP/feature_extraction/extract_features.py

License

Distributed under the MIT License. See LICENSE.txt for more information.

Contact

Rüdiger Schmitz - [email protected]

Project Link: https://github.com/IPMI-ICNS-UKE/NET_CUP