Authors

X-COVID AI Assistant

A Web Application to detect signs of COVID-19 presence from Chest X-Rays using Deep Learning

Motivation

With shortages and delays in PCR tests, chest x-rays have become one of the fastest and most affordable ways for doctors to triage patients. In many hospitals, patients often have to wait six hours or more for a specialist to look at their x-rays. If an emergency room doctor could get an initial reading from an AI-based tool, it could dramatically shrink that wait time. Before the pandemic, health-care AI was already a booming area of research. Deep learning, in particular, has demonstrated impressive results for analyzing medical images to identify diseases like breast and lung cancer or glaucoma at least as accurately as human specialists. Therefore, what we envision through our Project, is to follow this paradigm, and equip doctors with a powerful AI tool in the fight against COVID-19.

What it does

X-COVID AI Assistant is a simple and effective application that could be used by health care providers as a supportive tool on the examination process and patient triage. The use of this application would be very beneficial in the following cases:

  • Use with portable x-ray devices to fast-track test results and decision-making in advance to the PCR results
  • Use as a supportive tool during case overload, which can likely lead to low reading quality
  • Use when regular monitoring is required among patients showing mild symptoms in order to identify and/or triage patients by the progression of symptoms

How it works

Our solution is based on a Deep Learning model that was trained on a dataset of Chest X-Rays images from real patients. This model, known as COVID-Net, was implemented and tested by a research team from the University of Waterloo. Our challenge is to correctly incorporate that model into the heart of a web-based, privacy-aware, platform that could be easily used by health care providers.

Through our application a health care provider can insert a Chest X-Ray image of a suspected COVID-19 patient and get the model's prediction in less than 10 seconds.

Demo

A live demo of the web application is currently running here.

Proposed Business Model Canvas

A Business Model Canvas prototype for X-COVID AI Assistant is presented in this video here.

Technologies used

The user interface was built with React.js, HTML and CSS. On the backend, we developed a Flask API using Python and Tensorflow to serve the Deep Learning Model. We also used Docker and Nginx to deploy the application on Google Cloud Platform.

What's next for X-COVID AI Assistant

  • We will work on adding a new feature that interprets the model's decision in a human understandable way. We will utilize LIME (Local Interpretable Model-agnostic Explanations) algorithm to visualize the image areas that played the most important role in the model's prediction.

Example:

  • We will reach out to more health care providers and hospitals to test our application.

The Day After Corona

As we mentioned before, Deep Learning has demonstrated impressive results for analyzing medical images to identify diseases like breast and lung cancer or glaucoma at least as accurately as human specialists. Our application can support any type of machine learning model and can easily adjust to serve as a supportive tool for a different application.

Try It out

Hackathons

Technologies

css, docker, flask, github, html, lime, nginx, opencv, python, react.js, tensorflow, web-app

Devpost Software Identifier

263612