COVID-19 Severity Detection from Lung CT-Scan Images using CNN

The severity of COVID-19 was detected in this project from Computed Tomography Scan (CT-scan) images of lungs using CNN. The world is devastated by the COVID- 19 epidemic, and the economy is in deep crisis. COVID-19 is one of the most recent high-risk issues worldwide. It is more risky and spreads very quickly through the respiratory tract. Anyone can get sick with COVID-19 and become seriously ill or die at any age. People who suffer from various diseases are more likely to become infected and develop more serious conditions. Most of the seniors are suffering from various lung problems, so they are becoming more infected and most of them have lost their lives. However, medical agencies do not have adequate equipment to detect COVID- 19 and its severity. Moreover, sometimes this tool has given wrong results. Therefore, it is very important to identify corona positive patients to prevent the spread of this virus. The best way to prevent this is to detect the severity of COVID-19 from a CT scan because it gives more accurate results and treats accordingly. People will be able to recognize the status of the infected person through this application. The condition may vary from critical, extent, minimal, moderate, normal or severe. The dataset was collected from the COVID-19 Infection Percentage Estimation Challenge arranged by CodaLab. They provided 3053 CT-scan images of lungs. These images have been classified into 6 classes. The classes are balanced by augmentation of the images. After augmentation 8614 images are created in the dataset. Some pre-trained models of CNN architecture like VGG16 and MobileNet have been used. Also, Convolutional Neural Network, densenet121 and Sequential CNN models have been applied for training. The accuracy found after applying these models are 85.92%, 80%, 76%, 69% and 63.87% from VGG16, Densenet121, Convolutional Neural Network, Mobilenet and Sequential models respectively. Among all the models the VGG16 has scored the highest accuracy. In the future, more models will be applied for training and for better accuracy.
TECHNOLOGY::Electrical engineering, electronics and photonics::Electrical engineering
Department Name
North South University
Printed Thesis