Detecting covid-19 in x-ray images with deep learning

Aw, Ethel Miao Han (2023) Detecting covid-19 in x-ray images with deep learning. Final Year Project, UTAR.

Preview

PDF Download (2380Kb) | Preview

Abstract

The Corona Virus Disease-2019 (COVID-19) has had a profound impact on the world and thus creates awareness of the need for a fast and accurate diagnosis if a similar outbreak occurs again. Chest X-Ray (CXR) is widely used to detect COVID-19 manually, but it is time-consuming and prone to errors, especially when the outbreak is severe. Deep Learning (DL) algorithms, i.e., Convolutional Neural Networks (CNNs), have shown promising results in automatically detecting COVID-19. This project used (i) single CNNs, (ii) incrementally learned CNNs, and (iii) incrementally learned multiple CNNs with majority voting to extract features from CXR images. Then, an XGBoost classifier was used with each of these CNNs to detect COVID-19. A dataset consisting of 22,900 images was used for training (66.67%), validation (16.67%), and testing (16.67%). The results show that using XGBoost classifier with incrementally learned and incrementally learned multiple CNNs gave good and comparable detection accuracy (94.56% and 94.58%). The best performer – incrementally learned multiple CNNs with majority voting used ResNet152, DenseNet201, and VGG16.

Actions (login required)