Document Types
Paper Presentation
School Code
N/A
School Name
Philippine Science High School - CALABARZON Region Campus, Batangas City, Batangas
Abstract/Executive Summary
The COVID-19 pandemic has heavily affected the well-being of people worldwide. Current diagnostic tools, like the RT-PCR, are expensive and time-consuming; thus, there is a need for cheaper and faster means of COVID-19 detection. This study proposes using a desktop application with a convolutional neural network (CNN) and visual analysis as a supplementary diagnostic tool for detecting COVID-19 pneumonia in chest X-ray images. The CNN used is a sequential Keras model that was trained and tested through eight epochs using an augmented dataset. Random data augmentation techniques applied were rotation and horizontal flipping, which increased the total images used to 13,584. Visual analysis was created using the Grad-CAM algorithm to determine patterns in chest X-ray images. These were implemented in a desktop application and evaluated by a professional pulmonologist. Results showed that the CNN achieved an average accuracy rate of 97.96% among the three classes, which was superior among related studies. The CNN also achieved a precision, recall, and F1-score of 99.67%, 99.62%, and 99.64% respectively for COVID-19 pneumonia, 99.26%, 94.83%, and 96.99% respectively for viral pneumonia, and 95.12%, 99.42%, and 97.22% respectively for normal chest X-ray images. Meanwhile, the visual analysis was also accurate, as evaluated by a professional pulmonologist, where patterns of haziness were determined. Hence, this could serve as an effective supplementary diagnostic tool for healthcare professionals for faster and more accurate diagnosis of COVID-19 and viral pneumonia patients.
Keywords
COVID-19; pneumonia; convolutional neural network; chest x-ray image; desktop application; Grad-CAM
Research Theme (for Paper Presentation and Poster Presentation submissions only)
Computer and Software Technology, and Robotics (CSR)
Initial Consent for Publication
yes
COVIDetect: A Desktop Application as a Diagnostic Tool for Novel Coronavirus (COVID-19) Pneumonia in Chest X-ray Images Using Convolutional Neural Network
The COVID-19 pandemic has heavily affected the well-being of people worldwide. Current diagnostic tools, like the RT-PCR, are expensive and time-consuming; thus, there is a need for cheaper and faster means of COVID-19 detection. This study proposes using a desktop application with a convolutional neural network (CNN) and visual analysis as a supplementary diagnostic tool for detecting COVID-19 pneumonia in chest X-ray images. The CNN used is a sequential Keras model that was trained and tested through eight epochs using an augmented dataset. Random data augmentation techniques applied were rotation and horizontal flipping, which increased the total images used to 13,584. Visual analysis was created using the Grad-CAM algorithm to determine patterns in chest X-ray images. These were implemented in a desktop application and evaluated by a professional pulmonologist. Results showed that the CNN achieved an average accuracy rate of 97.96% among the three classes, which was superior among related studies. The CNN also achieved a precision, recall, and F1-score of 99.67%, 99.62%, and 99.64% respectively for COVID-19 pneumonia, 99.26%, 94.83%, and 96.99% respectively for viral pneumonia, and 95.12%, 99.42%, and 97.22% respectively for normal chest X-ray images. Meanwhile, the visual analysis was also accurate, as evaluated by a professional pulmonologist, where patterns of haziness were determined. Hence, this could serve as an effective supplementary diagnostic tool for healthcare professionals for faster and more accurate diagnosis of COVID-19 and viral pneumonia patients.
