Testings are essential to detect and containerize COVID-19 pandemic. However, the need for testing is excessive and the equipment is in shortage around the world. Understand the situation, we provide a solution for this rising issue.
What it does
- COVID-19 detector tentatively predicts the probability of an infected patient based on Chest CT scan images (Computed Tomography scan technology is a popular service at most hospitals). We aim to use this project to detect people with high potential of having COVID-19. Thus, health care providers can approach, test, and deliver supports to these patients faster. The detector has a high precision rate at 91% and gives prediction in only seconds.
How we built it
- The backend was built with Python with Flask
- The ML model was trained using
Custom Vision AIfrom Microsoft Azure with a set of data which consists of chest CT scan images of 329 positive COVID-19 patients and 387 negative cases.
- The web app was deployed using Microsoft Azure Web Service with containerized by Docker
Challenges we ran into
- We had a difficult time to find a good, reliable dataset of positive COVID-19 chest CT scan images.
- We had problems with deploying to Azure Web server at first because we can not set up the pipeline with GitHub.
- We also had problems with some stylings with CSS.
Accomplishments that we're proud of
- The model that we trained has a precision rate of 91% and a recall rate of 85.6%.
- We've successfully deployed our web app to azure web service.
- The website has every functionality we planned to implement.
What we learned
- How Flask serves 2 static folders
- How to utilize the Microsoft Azure Custom Vision to quickly train ML model
- How to deploy a web app to Azure Web Service with Docker
What's next for COVID-19
- We look forward to improving the data set to get better predictions.