M.S. AAI Capstone Chronicles 2024

CNN Lung Disease Classification

7

Background Information

Studying Effusion, Atelectasis, and Infiltration poses a substantial challenge in medical diagnoses due to their overlapping symptoms and complex manifestations in chest X-rays. Effusion, the abnormal accumulation of fluid in the pleural space, can compromise respiratory function and is often a sign of underlying systemic diseases such as heart failure (Krishna, Antoine, & Rudrappa, 2024). Atelectasis, the collapse or incomplete expansion of lung tissue, can lead to severe respiratory distress if not promptly addressed (Johns Hopkins Medicine, n.d.). Infiltration, characterized by the presence of substances like fluid or cells within the lung parenchyma, is frequently associated with infections like pneumonia (Core Concepts in Clinical Infectious Diseases (CCCID), 2016). By selecting these diseases, we aimed to leverage CNNs and transfer learning in order to identify these conditions individually while also recognizing their co-occurrence. Transfer learning is a technique widely used in medical imaging that improves performance on models with limited data. Salehi et. al. (2023) address the challenges of small datasets and computational limitations in A Study of CNN and Transfer Learning in Medical Imaging . By reusing models that are pre trained on large datasets, such as ImageNet or our NIH Chest X-Ray dataset, transfer learning can reduce the need for extensive labeled data. This can improve accuracy, shorten training time and manage class imbalances (Salehi et. al., 2023). However despite these advantages, transfer learning may still transfer biases from pre-trained models, and requires significant computational resources for fine-tuning.

165