This repository contains the code to reproduce the results of our ConRad paper: Arxiv preprint.
Requirements:
The notebooks should be executed in the following order:
- create_dataset.ipynb, extract nodule volumes and associated annotations (segmentations, biomarkers, malignancy)
- finetune_models.ipynb, finetune ResNet end-to-end CNN model and biomarker regression model
- extract_radiomics.ipynb, extract radiomics features using nodule volumes and segmentations
- 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.