A Robust Ensemble Learning Framework for High-Precision Diagnosis of Pulmonary Diseases using CT Imaging

Lung diseases, including lung cancer and COVID-19, are the greatest reason for transnational mortality, necessitating laborsaving diagnostic tools for early detection. This paper offers a weighted ensemble model incorporating DenseNet-121, RegNetX-032, and Vision Transformer (ViT) to organize lung diseases from computed tomography (CT) scans. Leveraging rampant connectivity, scalable convolutional techniques, and attention mechanisms, the model utilizes a semi-supervised framework with pseudo-labeling and weighted prediction to accomplish a test accuracy of 94.95%. Assessed on 15,000 CT scans (11,250 labeled, 3,750 unlabeled), the ensemble outperforms unique models (DenseNet-121: 60.15%, RegNetX-032: 82.66%, ViT: 66.79%) and benchmarks (ResNet-50: 90.1%). Progressive preprocessing, model-specific optimizations, and dynamic weighting ensure scalability and precision, positioning the model as a robust tool for clinical diagnostics.