Authors

Title

World's first covid19 open source Ai model published on February 9, 2020.

Inspiration

A paper by Shan Gui et al, "Time Course of Lung Changes On Chest CT During Recovery From 2019 Novel Coronavirus (COVID-19) Pneumonia" discussing that Covid19 was a form of pneumonia in many cases, inspired Jordan (the originator of the project) to initially prepare a Covid19 open source model.

What it does

As the world's first covid19 open source Ai model published on February 9, 2020, the software does xray based Covid19 diagnosis.

How we built it

  1. Jordan took pretrained artificial neural network pneumonia model from kaggle, and trained on scarce Covid19 data. (Rationalle: The paper in the inspirations section above linked covid19 to pneumonia, so Jordan had decided to use pretrained pneumonia model.)
  2. Jordan buid a user interface atop that scarely trained model.
  3. Interface was eventually extended to add drag/drop capability. This was suggested by a facebook user to Jordan: "Andrei Marinescu". https://www.facebook.com/ProgrammingGodJordan/posts/906357816489912
  4. Jordan welcomed Jeremy Kohn to the team, who later suggested an open source license to the main repository, that was later merged by Jordan.

Challenges we ran into

Scarce covid19 data was an issue. Data was gathered from various covid19 research paper.

Jordan also made a call for local government to use administrative powers to try to acquire covid19 xray data from experienced countries with many more cases:

Jordan's advice to Ministry of Health (February 17, 2020): https://drive.google.com/file/d/1BNXkKJPZuMx64XzwqFmQEpC5s9-C3tJH/view?usp=sharing

Accomplishments that we're proud of

  1. Submitted manual to a caribbean journal, response pending. https://github.com/JordanMicahBennett/SMART-CT-SCAN_BASED-COVID19_VIRUS_DETECTOR/blob/master/data/pdfs/%5BManuscript__Poster%5D%20Convolutional%20Artificial%20Neural%20Network%20for%20Covid19%20Diagnosis%2C%20from%20XRay%20images%20of%20lungs.pdf
  2. On the task of Covid19 detection, so far, with the very limited data available, Sensitivity/Specificity/Accuracy are ~85%/~70%/~77% respectively, as seen in this screenshot, (where the model has been trained on a covid19 dataset Jordan organized).

What we learned

1) Only 6% of actual covid19 infections have been detected by countries worldwide, according to study cited in a new April 7, 2020 Medical Xpress article. https://medicalxpress.com/news/2020-04-covid-average-actual-infections-worldwide.html

2) On March 19, 2020, Epidemiologist Larry Brilliant, (helped to stop smallpox), says we can beat the novel coronavirus—but first, we need lots more testing. https://www.wired.com/story/coronavirus-interview-larry-brilliant-smallpox-epidemiologist/

In summary lots of testing is required is one way we can try to control this coronavirus2019 virus outbreak.

What's next for Smart Covid19 Detector

The next step is reasonably to train the model on even more data, and work with hospitals to deliver this solution.

Try It out

Hackathons

Technologies

keras, python, tensorflow

Devpost Software Identifier

261474