An Enhanced Framework for Pulmonary Emphysema Classification Using Advanced Optimization Algorithms and Deep Learning Techniques

Background Pulmonary emphysema, is a critical manifestation of chronic obstructive pulmonary disease (COPD) and a precursor to long tumors, necessitate accurately early detection for effective clinical management. Current diagnostic methods often lack precision in subclassifying emphysema severity. To address this, we propose a deep learning framework that integrates advanced segmentation and classification model to improve diagnostic accuracy and support clinical decision making. Material and Methods A benchmark dataset of pulmonary emphysema images was utilized. The proposed framework involves three stages (1) Segmentation using the novel adaptive Trans Residual DenseNet Unet (ATRDuNet). (2) Parameter optimization via the fitness revised position of Wild Horse Optimizer (FRP-WHO), and Classification through a Vision Transformer-based residual DenseNet with an integrated LSTM layer (ViTrans-RDLSTM). The performance of the framework rigorously evaluated against state-of-art methods in terms of segmentation accuracy, computational efficiency, and classification matrices. Results The ATRDUNet achieved a segmentation Dice score of 94.2%, outperforming traditional UNet architectures by 6.5%. the FPR-WHO optimizer reduced training convergence time by 22% compared to standard optimizers. The ViTrans-RDLSTM classifier attained an accuracy of 98.1% and an F1-score of 97.8%, surpassing existing models like ResNet50 (92.3%) and Inception-V3 (93.7%). Statistical analysis confirmed the superiority of the proposed framework (p-value < 0.01) Conclusion The Integrated ATRDUNet-FPR-WHO and ViTrans-RDLSTM framework demonstrates exceptional in pulmonary emphysema classification, offering high precision and reduced computational complexity. Its robustness and efficiency make it suitable for deployment in clinical systems, adding radiologists in early diagnosis and subclassification of emphysema severity.