AI Predicting COVID-19 using Cough & other vital information

Move fast and treat things, Predicting COVID-19 symptoms using Cough and other vital information

Comment 4

Inspiration

The COVID-19 pandemic is spreading rapidly across the world. There is a growing shortage of medical testing facilities. Tens of thousands of potentially infected people who need to be tested do not have easy access to medical tests.

By April 26, 2020 coronavirus disease 2019 (COVID-19) had 2,910,918 confirmed cases, caused 202,865 deaths.

The losses are compounding every day. Given no vaccination or cure exists as of now, minimizing the spread by timely testing the population and isolating the infected people is the only effective defence, Mankind has at its disposal against the unprecedentedly contagious COVID-19.

One of the significant challenges medics face in the current pandemic is the lack of a fast and reliable testing method for COVID-19. PCR tests work well, but you have to make a trip to a lab and wait for hours to get the results. Many countries don’t have enough tests as well, and an alternative could prove very useful.

What it does

According to the WHO official report, the four major symptoms of the COVID-19 patients are fever, dry cough, tiredness and difficulty in breathing.

It’s difficult to go door to door testing, but a major part of the population has access to mobile phones and the internet.

Our idea is to collect the information about Dry cough using the microphone in the phone, fever using the thermometer, tiredness (using sleep pattern if available from health bands, MI band ), diabetes and blood sugar value if the device is available and breathing difficulty using breathing sounds. We combine all the information and build an A.I model to perform a signal processing on cough sounds and find a pattern in COVID-19 symptoms.

But unfortunately, cough is a common symptom of over two dozen non-COVID-19 related medical conditions. This makes the COVID-19 diagnosis from cough alone an extremely challenging problem. To fix this, we are combining data from different types of coughs along with other vital information to let the model learn a better pattern in collected data.

The work is about diagnostics and in time, with enough information, we hope that cough and other essential information could be used for early diagnosis. Models like these can be lifesavers by directly informing the public health response during a flu epidemic.

How I built it

We are collecting cough data by using app.coughresearch.ai, We are a team of scientists and engineers working on health and A.I solutions. Our goal is to develop an open-source artificial intelligence system for first phase COVID-19 screening using cough sound, fever, other required information, that could potentially reach every person in the world. The model we have built on manually collected cough sounds data, attempts to find the pattern in cough audio signal and identify if a person exhibits different cough symptoms, with reasonable accuracy of 85% ~ 90%. The idea is to collect important information such as fever, diabetes and cough sound using a smartphone and predict the probability of having COVID-19 symptoms as early as possible.

Challenges I ran into

Few doctors are helping us to find significant events in cough audio and to verify the label associated with them.

There is no COVID-19 cough audio data available at the moment, but we have collected different types of cough including dry, wet and a healthy one from different sources. I’ve gone through all possible websites and resources where sound clips available and collected dry,wet cough and healthy cough sounds manually and tagged it. So far, I’ve collected around ~1k sound clips manually. Cleaned the data ( removing silence and noise ) did the audio signal processing and built the model, Right now getting around ~90% accuracy on the data, I’ve collected.

Accomplishments that I'm proud of

One month ago, I started alone and from website to data collection, app building, signal processing, model building, message sharing, building a good team, All I did alone but now we are a group of few good people who also want to help me and work as a team. I'm proud of having them in my team.

What I learned

I have learned how to start something without waiting for help, Do it and people will come, join you and help you. just take the first step and start.

What's next for Predicting COVID-19 using Cough & other vital information

We are planning to release an app which can detect COVID symptoms by using cough sounds, fever and other vital information. read more about our research here: http://www.coughresearch.ai

How can you help us?

We are calling on you to join the initiative, help us to build together and save people lives. We require a large number of cough recordings from COVID-infected subjects, and of uninfected people, both healthy and those with other health problems as well.

We need your help to make our voice reach to those who can help us to collect the COVID patient’s data, any government authority, medical officials, research professors or media.

We are collecting cough data by using https://app.coughresearch.ai It takes less than 2 minutes to collect the data. We respect your privacy and contribution. We do not collect any sensitive personal data (e.g., name, telephone number, email, location). We need your support and contribution in the project, please help us in the fight against COVID-19 by filling the form with correct detail. Stay Home and Join us in this fight against COVID-19 by sharing this app with your friend circle, doctors and social media.

Because TOGETHER, WE CAN! We are looking for help in other fields too. If you can help us in any of the following, You are welcome to join and contribute

  1. Audio signal processing
  2. Advanced Image processing with spectrogram
  3. Feature extraction from multidimensional data
  4. Biomedical field
  5. Government officer
  6. React developer ( Front-end help )
  7. Ideas explorer

Built With

Share this project:

Updates