Optimizing CNN Architectures for Advanced Thoracic Disease Classification
Tejas Mirthipati
TL;DR
The paper addresses thoracic disease classification from chest X-ray images under challenges of dataset imbalance, multi-label outputs, and imaging variations. It evaluates binary, multi-label, and ResNet50 CNN architectures, augmented with PCA-based image compression and a class-weighted loss to mitigate bias. Key findings show 66% accuracy and an AUC of 0.708 for binary classification, while multi-label classification remains challenging due to rare diseases and visually similar conditions, though ResNet50 offers better generalization under computational constraints. The work suggests future directions including data augmentation, synthetic data generation, transfer learning, and incorporating patient demographics and imaging conditions to improve robustness and clinical applicability.
Abstract
Machine learning, particularly convolutional neural networks (CNNs), has shown promise in medical image analysis, especially for thoracic disease detection using chest X-ray images. In this study, we evaluate various CNN architectures, including binary classification, multi-label classification, and ResNet50 models, to address challenges like dataset imbalance, variations in image quality, and hidden biases. We introduce advanced preprocessing techniques such as principal component analysis (PCA) for image compression and propose a novel class-weighted loss function to mitigate imbalance issues. Our results highlight the potential of CNNs in medical imaging but emphasize that issues like unbalanced datasets and variations in image acquisition methods must be addressed for optimal model performance.