CoviScan | Devpost

What is CoviScan

CoviScan is COVID19/Pneumonia detection application through chest radiographs. CoviScan consist of a Web interface and an iOS app which lets user to upload their chest x-rays. These two application works on the Computer Vision Classifier which is developed by us too.

How it helps

Currently, Covid-19 can be diagnosed via RT-PCR to detect genetic material from the virus . However, it can take a few hours and sometimes days before the molecular test results are back. By contrast, chest radiographs can be obtained in minutes. And this is what CoviScan does. It allows user to upload their chest x-rays and predicts whether they are suffering from COVID-19/Pneumonia.

Through this application we will be helping in differentiating those patients which need more attention(severe cases) than others and reducing pressure on doctors to check every patients. With this application, doctors can identify critical patients instantly without wasting any minute.

This application not only helps doctors but helps patients too. With one tap, patients can get to know what they are suffering from. Patients only need their X-ray images and rest of the work is done by the app. Patients can self assess themselves and can act accordingly without wasting any second.

What it does

It is an automated imaging tool which process chest x-ray images and predicts whether the person has Covid-19 disease or Pneumonia or Normal.

How we built it

In this project we have developed a Computer Vision Classifier which is based on Convolution Neural Network(CNN) in Deep Learning it scans the images and publishes the result which is displayed on the user screen. We used Deep Learning techniques, where we defined a custom CNN model for the X-Ray validator. We used transfer learning on two different architectures i.e., DenseNet121, MobileNetV2 for classification and we used state-of-art U-Net architecture for extracting the lungs from the x-ray image. For training these models we collected the data from different sources then organised them in one collection for training. We also used different python libraries like Open-CV, Numpy, PIL etc. for image processing. And then for deployment, we integrated this tool into a web application written using Django framework in python.

In this project, we have also developed an iOS app. It is beautifully designed app with great UI which makes it very user friendly. Mostly the app is built on SwiftUI framework. Some parts of the app is built on Swift and UIKit especially the ImagePicker code and code related to displaying Lottie Animation. The app contains Login Page which is powered by Firebase. Features like Dark Mode and Persistent Login are also available. The app works on any iPhone and iPad which supports iOS 14 or above.

Challenges we ran into

The first challenge was to collect data for training. It was hard to find organised data in one place especially the COVID positive x-ray images. So we collected data from different sources but still, the data was very imbalanced and we used data augmentation techniques for balancing the data. The second challenge was to find the best model/CNN whether its custom defined or available state-of-art CNN architectures. After training and evaluating various models we ended up selecting DenseNet121 architecture as the best performing model for classification. The third challenge we faced was during the development of our iOS app. We have developed the app in native way by using Swift. We faced problems while uploading the image and sending it to our server but fortunately we were able to solve it using URLSessionUploadTask.

What's next for CoviScan

This is just the beginning. We want to make this project as extensive as possible. We will be adding more functionalities to our app like User scan log which provide all the scans of a particular user and many more.