gyguo / ie-net

official code of "Eliminating Indefiniteness of Clinical Spectrum for Better Screening COVID-19", accepted by IEEE Journal of Biomedical and Health Informatics (JBHI2021).

IE-Net

official implementation of "Eliminating Indefiniteness of Clinical Spectrum for Better Screening COVID-19", accepted by IEEE Journal of Biomedical and Health Informatics (JBHI2021).

Requirements

To install the environment via Anaconda:

conda env create -f environment.yaml

Prepare Dataset

the original dataset is in https://www.kaggle.com/einsteindata4u/covid19

cd ./data
unzip feature_original.zip

Training and Evaluation of IE-Net

To train the models in this paper, run this command:

cd ./tools
python main_multiple.py

Best Models

The best performed model is in

./checkpoint_best.pth

Comparison

The comparison experiments are

cd ./comparison
python comparison_experiments_fill_0_multimetric.py

Results

For 10-fold cross-validation, our model achieves high performance on the COVID-19 Clinical dataset. The table below shows the results in the paper.

	Accuracy	Recall	Precision	AUC
Results in paper	94.80±1.98	92.79±3.07	92.97±3.06	94.93±2.00

Recently, we introduce F1-score as the metric for selecting the best model, the performance in terms of Recall and Precision has improved. As mentioned in the paper, Recall is the most important metric in this paper.

	Accuracy	Recall	Precision	AUC	F1
Results F1	94.05±2.17	95.99±3.69	94.42±2.26	90.50±3.76	93.81±2.52