Multi-modal Point-of-Care Diagnostic Methods for COVID-19 Based on Acoustics and Symptoms

About:

This software reproduces the results in the manuscript "Multi-modal point-of-care diagnostic methods for COVID-19 based on acoustics and symptoms", submitted to IEEE Journal of Biomedical and Health Informatics. A preprint of the manuscript is available at https://arxiv.org/abs/2106.00639

Directory structure:

LICENSE.md
README.md
local/
- classifier_on_audios.py
- classifier_on_symptoms.py
- feature_extraction.py
- score_fusion.py
- scoring.py
- utils.py
plotscripts/
- JBHI_dataset_metadata_plot.py
- JBHI_symptoms_odds_ratio_plot.py
- JBHI_lr_svm_val_test_ROCs_plot.py
- JBHI_performance_summary_plot.py
- JBHI_fused_test_ROC_confusion_matrices_plot.py
- JBHI_score_analysis_plot.py
LISTS
- train_fold_[1-5]_list
- val_fold_[1-5]_list
- test_list
- recovered_ids
- negatives_after_april2021
- category_to_class
- symptoms
run.sh
REQUIREMENTS.txt

Directory contents:

run.sh [ Master (shell) script to run the codes ]
local/feature_extraction.py [ Extract ComParE2016 features using opensmile ]
local/classifier_on_audios.py [ Train LR, Linear-SVM, RBF-SVM classifiers on audio signals ]
local/classifier_on_symptoms.py [ Train decision tree classifier on Syptoms feature ]
local/score_fusion.py [ Score fusion of results from three audio categories and symptoms ]
local/utils.py [ Commom util functions ]
local/scoring.py [ Performance: computes false positives, true positives, etc., from ground truth labels and the system scores ]
plotscripts/JBHI_dataset_metadata_plot.py [Generate Fig. 1 in the manuscript]
plotscripts/JBHI_symptoms_odds_ratio_plot.py [Generate Fig. 4 in the manuscript]
plotscripts/JBHI_lr_svm_val_test_ROCs_plot.py [Generate Fig. 5 in the manuscript]
plotscripts/JBHI_performance_summary_plot.py [Generate Fig. 7 in the manuscript]
plotscripts/JBHI_fused_test_ROC_confusion_matrices_plot.py [Generate Figs. 8,9 in the manuscript]
plotscripts/JBHI_score_analysis_plot.py [Generate Fig. 10 in the manuscript]
REQUIREMENTS.txt [ Contains a list of dependencies to run the system ]

How to run:

Open shell terminal (Linux), navigate to the directory containing the code
Type the following and hit enter: ./run.sh

Results: Otained for the Logistic regression classifier

Model	AUC	Sensitivity (specificity)
Breathing	0.777	37.90 (95.30)
Cough	0.740	24.10 (95.30)
Speech	0.789	27.60 (95.30)
Symtpoms	0.802	55.20 (95.70)
Acoustics	0.842	44.80 (95.30)
Acoustics & Symptoms	0.924	69.00 (95.30)

Sensitivity computed at 95% specificity
depending on the version of python packages in your system, the performance may be little different in the decimal places

Citation

If you use this software in your work, please cite the relevant paper.

@misc{chetupalli2021multimodal,
      title={{Multi-modal Point-of-Care Diagnostics for COVID-19 Based On Acoustics and Symptoms}},       
      author={Srikanth Raj Chetupalli and Prashant Krishnan and Neeraj Sharma and Ananya Muguli and Rohit Kumar and Viral Nanda and Lancelot Mark Pinto and Prasanta Kumar Ghosh and Sriram Ganapathy},
      year={2021},      
      eprint={2106.00639},
      archivePrefix={arXiv},
      primaryClass={eess.AS}
}

Contact Us:

Please reach out to Sriram Ganapathy, Assistant professor, IISc for any queries.

Authors:

Srikanth Raj Chetupalli | Postdoctoral Researcher, IISc, Bangalore
Prashant Krishnan | Research Assistant, IISc, Bangalore
Neeraj Kumar Sharma | CV Raman Postdoctoral Researcher, IISc, Bangalore
Ananya Muguli | Research Assistant, IISc, Bangalore
Rohit Kumar | Research Associate, IISc, Bangalore
Dr Viral Nanda | P. D. Hinduja National Hospital and Medical Research Center, Mumbai
Dr. Lancelot Mark Pinto | P. D. Hinduja National Hospital and Medical Research Center, Mumbai
Prasanta Kumar Ghosh | Associate Professor, IISc, Bangalore
Sriram Ganapathy | Assistant Professor, IISc, Bangalore

iiscleap / MuDiCov