Pytorch implementation for the LRH-Net: A Multi-Level Knowledge Distillation Approach for Low-Resource Heart Network (FAIR, MICCAI 2022).
Parallel-MLKD | Sequential-MLKD |
---|---|
Install anaconda/miniconda
Required packages
$ conda env create --name lrhnet --file env.yml
$ conda activate lrhnet
Install PyTorch
Download Datasets from https://moody-challenge.physionet.org/2020 and https://moody-challenge.physionet.org/2021
Main contributions of this paper are:
- A real-time cardiovascular disease detection model which is 106x smaller than a large-scale model and 12x times smaller than the existing low-scale model.
- A Multi-Level knowledge distillation approach to improve the performance of LRH-Net (student model) and to reduce the number of electrodes and input leads data required for the student model .
- Performed evaluation on a very diverse, publicly available and combination of multiple datasets to increase its desirability.
Further research can focus on lowering the performance gap in low-lead configurations by optimizing the number of steps required to distill majority of the critical information by varying the levels of MLKD so that the classification performance on hard-to-classify diseases does not get severely affected.
If you use the code or results in your research, please use the following BibTeX entry.
@InProceedings{10.1007/978-3-031-18523-6_18,
author="Chauhan, Ekansh
and Guptha, Swathi
and Reddy, Likith
and Raju, Bapi",
title="LRH-Net: A Multi-level Knowledge Distillation Approach for Low-Resource Heart Network",
booktitle="Distributed, Collaborative, and Federated Learning, and Affordable AI and Healthcare for Resource Diverse Global Health",
year="2022",
publisher="Springer Nature Switzerland",
address="Cham",
pages="190--201",
isbn="978-3-031-18523-6"
}