The implementation of the multi-modal fusion models within this repository is based on work done by Sophie Martin during an NHSE PhD internship. Key features include the multi-modal attention gate (MAG+) which was adapted from this repository and is inspired by the work of Zhao et al. to integrate multiple data modalities in and end-to-end training regime.
This model is intended for use in training a classification model for length of stay prediction on the MIMIC-IV dataset.
Data was downloaded from MIMIC-IV. MIMIC is a publically accessible repository (subject to data usage agreements and mandatory training) containing mulimodal data across healthcare centers in the US.