Detecting COVID-19 using Chest X-Rays
Introduction
So, midst the recent pandemic COVID-19 has surfaced and has devasted the world in many different ways.
There is a lot of ongoing research to detect and to cure COVID-19 but, and a lot of progress is being made on a daily basis.
So, I tried creating a model that would take chest X-Rays as images and would predict if they are COVID-19 positive or not.
You might think, How? Well, the answer would be that when someone contracts COVID-19, the virus shoots itself into the lung’s lining of mucus which finds it as a violation and in retaliation, it starts filling the lungs up with mucus(which is also called pneumonia), and leads to dire consequences.
That’s how COVID-19 becomes very fatal.
Let’s Try to Answer some of the pressing questions.
- How will Chest X-Rays help in predicting COVID-19?
- How efficient is the model?
- And can we rely on it as a secondary source of detection?
Well, Chest X-Rays will highlight the inner mucus lining and as we know that it will be different from that of a healthy lung. Which can help us predict it.
The model reaches an efficiency of almost 95%, because of the use of one the best image-net classifier VGG16.
In third world countries where the testing kit and facilities have still not been able to establish themselves, this might become handy for people. Also, where a lot of patients are coming in on a daily basis, the following prediction can be run to make a list of possible COVID-19 patients which would make the screening process much easier and efficient.
As you can see in the above plot, the number of tests being done is not enough.
Chest X-Rays compared with that of an infected patient. We can notice a lot of difference in the lung patterns.
Conclusion
In this story, we took a look at how neural nets(particularly CNN’s) can be useful in predicting the disease COVID-19 and can be used as a secondary means of detection.
The dataset was acquired from https://github.com/ieee8023/covid-chestxray-dataset, so you can go and check it out.
- We found out that image-net classifiers like VGG-16 can be used in the detection of positive patients.
- The amount of screening possible by implementing this method will be immense.
To find the code of the above observation, visit the link https://github.com/mayukhsil/DSND-Project-CRISP-DM