HuLP: Human-in-the-Loop for Prognosis

Muhammad Ridzuan, Mai A. Shaaban, Numan Saeed, Ikboljon Sobirov, Mohammad Yaqub

Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, UAE

---

---

💡 Highlights

• Allowing user interaction and expert intervention at inference time: HuLP facilitates human expert intervention during model inference, empowering clinicians to provide input and guidance based on their domain expertise. This capability significantly enhances the model's decision-making process, particularly in complex prognostic scenarios where expert knowledge is invaluable.
• Capability of handling both missing covariates and outcomes and extract of meaningful vector representations for prognosis: HuLP is equipped with a robust mechanism for handling missing data, ensuring end-to-end reliability in prognostic predictions. By leveraging patients' clinical information as intermediate concept labels, our model generates richer representations of clinical features, thereby enhancing prognostic accuracy.

🛠️ Install

Create environment:
conda create -n hulp python=3.8

Install dependencies: (we assume GPU device / cuda available):

pip install -r requirements.txt

Now, you should be all set.

▶️ Demo

Run the following for Hecktor dataset:

python train.py --model PrognosisModel --fold 0 --seed 0 --dataset_name hecktor --prognosis_loss deephit

Run the following for Lung cancer dataset:

python train.py --model PrognosisModel --fold 0 --seed 0 --dataset_name lung --prognosis_loss deephit

✒️ Citation

If you find our work helpful for your research, please consider citing the following BibTeX entry.

@article{ridzuan2024hulp,
  title={HuLP: Human-in-the-Loop for Prognosis},
  author={Ridzuan, Muhammad and Kassem, Mai and Saeed, Numan and Sobirov, Ikboljon and Yaqub, Mohammad},
  journal={arXiv preprint arXiv:2403.13078},
  year={2024}
}