Muhammad Zeeshan KaramatMelia FleischmannMohamed AmrGabriel Brandle
Published © MIT

DetectNow

Deep learning to detect biosignals of the cough of COVID-19 from recordings.

ExpertWork in progress1,577

Things used in this project

Software apps and online services

App link
Web app for data collection . We are in the process of conducting a medical clinical trials first, before releasing our testing functionality, as this medical domain is a extremely sensitive topic.

Story

Read more

Schematics

Deep Learning architecture

DetectNow is implemented on HIPAA and GDPR compliant infrastructure and all user data and trained models are using anonymised user data. Necessary permission and instructions are given to use for safe usage of DetectNow.
Technically, we use web app to record cough, individual variables (Age. Sex, Smoking history, Height / weight, Ethnicity) and supportive data symptoms ( Fever, headache, runny nose, joint pain / body aches, diarrhoea, vomiting, effort tolerance, conjunctivitis) to give diagnostic prediction about COVID19 at zero cost and at the comfort of home. We use deep learning and audio signal processing at the backbone of our big data realtime data pipeline. First signal is cleaned by removing unwanted noises, after the long audio file is segmented with a cough detection model, after that we use feature engineering (MEL spectrograms, episodic, and time series features) on this segmented audio chunk, which is then finally fed to the COVID19 classification model.

Code

Github Repository

Steps to Reproduce Clone The repo `git clone https://github.com/shresthagrawal/covid19.git` `cd covid19` Create a virtual env `virtualenv venv` `source venv/bin/activate` Clone `git clone https://github.com/tyiannak/pyAudioAnalysis.git` Install Dependencies `pip3 install -r ./App/requirements.txt` `pip3 install -r ./pyAudioAnalysis/requirements.txt` `brew install ffmpeg` Install `pip3 install -e pyAudioAnalysis` Using Test.py `python3 App/app.py` Alternative `chmod +x ./auto.sh` `./auto.sh` Training on the original dataset `python3 pyAudioAnalysis/pyAudioAnalysis/audioAnalysis.py trainClassifier -i data/cough/not_sick data/cough/sick --method randomforest -o model_new` Training on Uploads `python3 pyAudioAnalysis/pyAudioAnalysis/audioAnalysis.py trainClassifier -i data/uploads/not_sick data/uploads/sick --method randomforest -o model/model_new` Training Dataset [data-set](https://osf.io/4pt2s/)

Credits

Muhammad Zeeshan Karamat

Muhammad Zeeshan Karamat

1 project • 4 followers
Software engineer with extensive experience in leading successful large scale machine learning solutions.
Melia Fleischmann

Melia Fleischmann

0 projects • 1 follower
Medical doctor, resident surgeon
Mohamed Amr

Mohamed Amr

0 projects • 2 followers
Gabriel Brandle

Gabriel Brandle

0 projects • 1 follower
Thanks to Melia Flieshmann, Gabriel Brandel, Pia Eggert, Patrick Betz, Shresth Agrawal, Vlada Petrusenko, Mohammed Amr, Shizhe He, Alexandre Micheloud, Stephana Muller, Simon Hofer , and Muhammad Zeeshan Karamat.

Comments