ConRad

This repository contains the code to reproduce the results of our ConRad paper: Arxiv preprint.

Requirements:

• Download LIDC-IDRI dataset
• PyTorch
• PyTorch Lightning
• pylidc
• TorchIO

The notebooks should be executed in the following order:

1. create_dataset.ipynb, extract nodule volumes and associated annotations (segmentations, biomarkers, malignancy)
2. finetune_models.ipynb, finetune ResNet end-to-end CNN model and biomarker regression model
3. extract_radiomics.ipynb, extract radiomics features using nodule volumes and segmentations
4. classifiers.ipynb, train classifiers on various feature sets

The repository contains all the necessary code for finetuning the CNN and biomarker regression models. To allow for a quick reproduction of our results we also make pre-trained weights available. They should be placed in a folder called "weights".

The model can be interacted with here.