M.S. AAI Capstone Chronicles 2024

CNN Lung Disease Classification

5

preventing bias and simplifying the model. This reduced the overall complexity by not having to adjust our model to factor imbalances in gender. Similarly, all age groups in the dataset had a prevalence of Atelectasis, Effusion, and Infiltration. The consistency of diseases in both younger and older patients confirmed they were suitable choices for our models. These findings allowed us to narrow the scope of our research and ensure the models are focused primarily on the visual characteristics of our X-rays.

The dataset included a total of 457 X-ray images that include bounding boxes. Bounding boxes are critical in the context of medical image analysis as they delineate the regions of interest (ROIs) within an X-ray, highlighting areas where anomalies, such as signs of disease, are present (Gunnell, 2024). These annotated regions are essential for training and focusing on the precise areas that are clinically relevant, improving the model’s accuracy in detecting and classifying diseases. However, the limited number of X-rays with bounding boxes presents a significant limitation to our project. This restricts

163