AI-assisted pulmonary disease diagnosis model based on dual-stage feature selection

Abstract
Pneumonia and other lung diseases pose significant global health challenges. Deep learning approaches have shown potential in medical image analysis, but high-dimensional feature redundancy and dataset-specific variations can limit performance. This study presents a dual-stage framework that combines convolutional neural networks (CNNs) with feature selection and classifier optimization. Pre-trained CNNs are used as feature extractors for chest X-ray and CT images, followed by a two-stage feature selection process—mutual information filtering and recursive feature elimination with cross-validation (MIRFE-CV)—to reduce dimensionality while retaining discriminative information. An Archimedes Optimization Algorithm (AOA) is then applied for classifier hyperparameter tuning. Experiments on public pneumonia X-ray and COVID-19 CT datasets demonstrate competitive classification results with reduced features, highlighting the value of systematic feature-level refinement for decision-oriented CNN analysis in pulmonary disease imaging.