AI-based Diagnostic System for Chest X-rays: A Multi-Labeled Classification Approach using Deep Learning

Although chest X-rays (CXRs) are still a vital diagnostic tool for detecting thoracic disease, their interpretation can be challenging because of their multilevel findings and contradictory visual patterns.  As a result, we examine how well deep convolutional neural networks (CNNs) with transfer learning perform automated multi-label classification of CXRs.  Extensive preprocessing and augmentation techniques were used to address class imbalance and normalise image quality using the CheXpert dataset.  Under consistent experimental conditions, several CNN architectures, including CustomNet, DenseNet121, ResNet50, InceptionV3, and VGG16, were trained and evaluated.  With an AUROC of 0.78 and an accuracy of 87% on test data, DenseNet121 performs significantly better than all other models, according to a comparative analysis of AUROC and accuracy. Additional evaluation by disease category on an individual basis showed excellent performance for pleural effusion (AUROC 0.93) and lung opacity (AUROC 0.91). These results indicate the promise of Dense Net-based architectures to deliver accurate, automated diagnostic assistance to clinical radiology. The work emphasizes the utility of transfer learning in enhancing generalization with sparse labeled data and offers pragmatic guidance to model choice in the analysis of medical images.