Learning Disease State from Noisy Ordinal Disease Progression Labels

This repository contains the code for the MICCAI Submission "Learning Disease State from Noisy Ordinal Disease Progression Labels".

Installation

Set up a python environment with a python version 3.13. Then, download the repository, activate the environment and install all other dependencies with

cd Learning-Disease-State
pip install --editable . 

This installs the code in src as an editable package and all the dependencies in requirements.txt.

Running the Model

Mario Challenge Dataset

To train the model on the Mario Challenge dataset:

1. Update the dataset path in the train_mario.yaml config file.

2. Run the following command:

   python src/train_mario.py

Custom Dataset

To train the model on a different dataset:

1. Create your own pytorch dataset.

2. Update the dataset path and pretrained model path in the train_internal.yaml config file.

3. Run the following command:

   python src/train_internal.py

Organization of the repo

• configs: Configuration files for both mario and internal experiments.
• src: Main source code to run the experiments.
• train_mario.py: Training on the Mario challenge dataset.
• train_internal.py: Running the trained models on out-of-domain dataset.
• loss.py: Contains the loss function.
• dataset.py: Contains mario dataset.