Forecasting MASH Patient Trajectories on the Liver Transplant Waitlist using a Neural Network-based Competing Risk Analysis
DeepHit for MASH has been trained using SRTR data of MASH patients at the time of listing and the DeepHIT model structure to predict the monthly risk of dying on the death and receiving a liver transplant [1].
Authors: Gopika Punchhi, Yingji Sun, Sumeet K. Asrani, Sirisha Rambhatla, Mamatha Bhat
- Reference: G. Punchhi, Y. Sun, S. K. Asrani, S. Rambhatla, M. Bhat, "Forecasting MASH Patient Trajectories on the Liver Transplant Waitlist using a Neural Network-based Competing Risk Analysis "
[1] C. Lee, W. R. Zame, J. Yoon, M. van der Schaar, "DeepHit: A Deep Learning Approach to Survival Analysis with Competing Risks," AAAI Conference on Artificial Intelligence (AAAI), 2018
The processing of the SRTR data (CAND_LIIN dataset) can be done using data_processing.py with the re-categeorizaiton of features based on clinical definitions, missing data imputation, and one hot encoding. Sample data post-processing can be viewed here sample data. The template of input features,data upload template, can be viewed and downloaded to be filled and served as input for the model.
The model has been implemented in a web app for direct use from browsers. The Streamlit web app include functions such as downloading the data template for reference, uploading patient data as .csv files, and downloading the visualization and forecasted risk trajectories plots predicted with DeepHit for MASH.
The trained DeepHit model is availble for use. If you need help with importing the model files in tensorflow, check our code in DeepHit_Prediction.py for reference.
To run the web app on your own server, first install the packages with the command
pip install -r requirements.txt
Then use the following command to launch the streamlit app on your localhost
streamlit run DeepHit_Prediction.py