Efficient hybrid CNN–ViT framework for pulmonary nodule classification in CT images using feature-level optimization

Lung cancer, particularly cases associated with pulmonary nodules, remains one of the leading causes of cancer-related mortality worldwide, necessitating the development of reliable and computationally efficient computer-aided diagnosis (CAD) systems for early detection from thoracic CT scans. In this study, we propose a compact and intelligent hybrid deep learning framework that integrates convolutional neural networks (CNNs) with Vision Transformers (ViTs) to achieve high classification performance while maintaining deployment efficiency. The proposed methodology is evaluated on the LUNA16 benchmark dataset, derived from the LIDC-IDRI repository. Preprocessing involves resizing CT slices to \(64 × 64\) pixels, grayscale standardization, and extensive data augmentation including horizontal flipping, rotation, and scaling to enhance model generalization. The architecture employs EfficientNetV2B1 as a convolutional backbone for extracting fine-grained local features, followed by a lightweight Vision Transformer module with multi-head self-attention to model global contextual dependencies. To further optimize computational performance, structured model pruning is applied, significantly reducing parameter redundancy and inference latency. Experimental results demonstrate strong classification performance, achieving \(95.16%\) accuracy, \(97.86%\) precision, a \(96.13%\) F \(_1\) -score, and an AUC of \(99.00%\) . Model compression reduces the parameter count to 12.5 million while improving inference efficiency without compromising predictive capability. These findings indicate that the proposed hybrid CNN–ViT framework provides an accurate, scalable, and computationally efficient solution for automated lung nodule screening and classification in clinical settings.