AAI_2025_Capstone_Chronicles_Combined

However, limited availability of high-quality, well-annotated medical datasets remains a

major obstacle in applying AI to healthcare (Arora et al., 2023). Many datasets suffer from class

imbalance, inconsistent labeling, and lack of diverse representation, hindering model

generalization and real-world applicability (Salmi et al., 2024). Generative Adversarial Networks

(GANs) have emerged as a promising solution by generating realistic synthetic medical images to

augment datasets, improving model robustness and performance (Yi et al., 2019). Transfer

learning has also been employed effectively, enabling models trained on larger datasets to be

fine-tuned on smaller medical imaging datasets, significantly enhancing performance and

adaptability (Esteva et al., 2019). Although Reinforcement Learning, Meta-Learning, and

Federated Learning offer potential applications in healthcare, they are beyond the scope of this

project.

By leveraging GANs and transfer learning, researchers can create additional training

samples and utilize existing knowledge from larger datasets, capturing a wider range of

variations without the need for further patient data collection. This combined approach

enhances the ability of deep learning models to detect and diagnose diseases more accurately.

Convolutional Neural Networks for Medical Analysis

This project draws inspiration from Kermany et al.'s (2018) research and uses the same

dataset. Their study demonstrated the potential of CNNs in medical diagnostics by successfully

identifying macular degeneration in eye scans and pneumonia in chest X-rays. Published prior to

the COVID-19 pandemic, their work highlighted the early promise of CNNs in the medical field.

This project specifically utilizes the chest X-ray data from Kermany et al.'s study.

101