screening covid-19 suspect based on artificial intelligence

As SARS-CoV-2 is highly contagious and number of suspected patient is growing, we are trying to screening suspected using symptoms, chest x-ray and ct scan. We used machine learning and deep learning.

Comment

Inspiration

Try hard to save a life.

What it does

Model -1: It predicts the possibility of the COVID-19 case based on symptoms. Model -2: It predicts the possibility of the COVID-19 case from a chest x-ray image. Model -3: It predicts the possibility of the COVID-19 case from a CT scan image.

How we built it

We developed three models for predicting COVID-19 cases using machine learning and deep learning algorithms:

  1. A model based on symptoms with a support vector machine algorithm.
  2. A model based on chest x-ray with a deep learning algorithm ( Convolutional Neural network )
  3. A model based on CT images with a deep learning algorithm ( Convolutional Neural network ) We built a web application for hospitals or CVD centers and doctors or operators to access those models through the internet.

For machine learning and deep learning algorithms, we use PyTorch and the TensorFlow framework. And for the web applications, we use node js, angular js, MySQL, Nginx server, HTML and CSS. The web application is fully scalable as based on cloud architecture.

Challenges we ran into

  1. Shortage of data.
  2. Radiologist access.

Accomplishments that we are proud of

Symptoms Model: This model predicts the possibility of getting infection based on symptoms and we achieved 71% accuracy based on our current dataset. X-ray Model: We achieved 93% accuracy based on our current dataset and deployed in the cloud. CT Model: We achieved 78% accuracy based on our current dataset and trying to improve more. Web Application: We developed a web application that manages CVD centers, hospitals or clinics, doctors, and patients.

What we learned

We can do better together.

What's next for screening COVID-19 suspect based on artificial intelligence

We are trying our hard for collecting more x-ray images and more symptoms data. In our model-2 and model-3 cases, we used the whole image for feature extraction and prediction. If we can use masking of the infected areas in the lungs with the help of a radiologist, our model will perform better.

Share this project:

Updates