AAI_2025_Capstone_Chronicles_Combined

anatomical variations, such as different chest cavity sizes or patient poses, enriching the training

set with underrepresented features. This increased variability supports better generalization of

CNN models across diverse populations and imaging conditions, enhancing classification

performance, and reducing bias (Che Azemin et al., 2024).

To further expand the dataset, data augmentation techniques will be applied during

preprocessing. These augmentations include flipping images horizontally, applying random

shear, varying the brightness of images, and applying an optional Gaussian blur to sharpen

features. These data augmentation improve model quality and reduce overfitting (Lakshmanan

et al., 2021).

Background Information

The COVID-19 pandemic posed significant challenges for researchers across multiple

disciplines, necessitating rapid analysis of large volumes of medical data, including pathological

findings, scans, and X-rays (Born et al., 2021). This urgency led to a surge in the application of

neural networks, particularly convolutional neural networks (CNNs), for medical image analysis

(Muhamed et al., 2024). CNNs offer distinct advantages over traditional techniques by

automatically identifying relevant features from raw data, eliminating the need for manual

feature extraction, and significantly enhancing diagnostic precision and speed (Kourounis et al.,

2023). Their widespread adoption soon extended beyond COVID-19 diagnostics to broader

applications, including the detection and classification of pulmonary and ocular diseases,

demonstrating the versatility and effectiveness of CNNs in medical imaging.

100