AAI_2025_Capstone_Chronicles_Combined

The end users of this AI-powered diagnostic tool are medical professionals, particularly

radiologists and physicians, who can utilize the model to assist in pneumonia diagnostics. By

integrating synthetic image generation via GANs, the project aims to create a more accurate and

robust pneumonia detection system, addressing the limitations posed by data scarcity in

medical AI applications. This approach demonstrates that augmenting real medical datasets

with GAN-generated images can serve as a useful tool for developing high-accuracy diagnostic

models, contributing to improved patient outcomes and optimized healthcare workflows.

Data Summary

The dataset contains 5,856 grayscale X-ray images divided into training (5,216 images),

testing (624 images), and validation (16 images) subsets. The data is pre-sorted into

“PNEUMONIA” and “NORMAL” folders, which serve as class labels. The dataset is designed to

support the development of a CNN classification model to differentiate normal chest X-rays

from those showing pneumonia. It also serves as the basis for generating synthetic images using

a GAN to supplement the CNN model and improve performance metrics such as accuracy and

F1 score.

Variables and Data Characteristics

The dataset consists of X-ray images varying in size and resolution, depicting the chest

cavities of patients (Figure 1). Differences in patient posture and positioning are present,

including arms positioned above or below the head, slight arm rotations, and tilted or off-center

96