Track: COVID-19 Analytics Competition
Abstract
The most challenging task of the Covid-19 pandemic is the accurate detection of Covid-19 affected people. To fight against this challenge, automatic detection of Covid-19 can help to diagnose patients with more reliability. In this work, we have collected chest X-rays of Covid-19 affected patients and processed the dataset for training purpose. As we have imbalanced datasets as well as data scarcity for Covid-19 class, these might create problems during training of deep neural networks. To overcome that, we have tried to develop a simple CNN model by hyperparameter optimization so that this shallow network can give better results in case of data limitation. Later we compared the performance of our developed model and the existing VGG16 model. Our model has an overall accuracy of 88.12% whereas the VGG16 model has an accuracy of 84.34%.