Deep Learning-Based Automated Detection and Classification of Multi-Class Pulmonary Pathologies from High-Resolution Chest CT Imaging

Early and precise diagnosis of pulmonary diseases remains a cornerstone in reducing respiratory morbidity and mortality globally. Traditional radiological interpretations are frequently constrained by observer variability and high diagnostic workloads. This paper presents a novel convolutional neural network architecture, PulmoNet-X, engineered for the automated multi-class segmentation and classification of chronic respiratory pathologies, including pneumonia, interstitial lung disease, and pulmonary nodules. Trained on a curated multi-center dataset of 12,500 annotated high-resolution CT scans, the proposed model integrates spatial attention mechanisms to improve feature localization in ambiguous tissue margins. Experimental evaluations demonstrate an overall classification accuracy of 97.8% with an area under the curve (AUC) of 0.985, significantly outperforming baseline architectures. The framework exhibits robust cross-domain generalization and computational efficiency, establishing its clinical viability as an automated real-time decision-support system in thoracic radiology.