State-of-the-Art Deep Learning Strategies for Multi-Label Classification of Chest X-Ray Thoracic Disorders

This study tackles the challenge of multi-label classification of chest X-rays from the NIH dataset, focusing on diagnosing 14 thoracic pathologies. It employs advanced deep learning models and ensemble techniques to enhance diagnostic accuracy. Three key CNN architectures are used: MobileNet, EfficientNet, and Extreme Learning Machine (ELM). MobileNet is efficient for resource-limited environ-ments, EfficientNet balances accuracy and computational efficiency, while ELM offers fast learning for large datasets. Ensemble methods like average, weighted average, and max vote combine these models to boost robustness and reliability. The performance is evaluated using metrics such as AUC, Fl score, and Hamming loss, with EfficientNet emerging as the top performer. The weighted average ensemble achieves the best results with an AUC of 0.8721, an Fl score of 0.8245, and a Hamming loss of 0.0987. This research demonstrates the potential of CNNs and ensemble methods in medical image analysis, advancing diagnostic tools for better clinical outcomes and improved healthcare.