Hybrid Deep Learning Framework Using AlexNet, VGG16, and PSO–GA Optimization for Lung Cancer Detection from CT Scans

Lung cancer remains one of the leading causes of cancer-related deaths worldwide, primarily due to delayed diagnosis and limitations in manual interpretation of CT scans. This study proposes a hybrid deep learning framework that integrates AlexNet and VGG16 for feature extraction with a hybrid Particle Swarm Optimization–Genetic Algorithm (PSO–GA) for feature selection and hyperparameter optimization. The framework was evaluated using publicly available LIDC-IDRI and NSCLC-Radiomics datasets containing benign and malignant CT images. Preprocessing techniques including normalization, denoising, resizing, and augmentation were applied to improve data quality and reduce overfitting. Deep features extracted from AlexNet and VGG16 were optimized through the PSO–GA mechanism to retain the most discriminative features while reducing redundancy. A dual-stage classifier was then employed for lung nodule classification. Experimental results demonstrated classification accuracies of 95.2% and 94.7% on LIDC-IDRI and NSCLC-Radiomics datasets respectively, outperforming baseline CNN approaches. Grad-CAM visualization further improved interpretability by highlighting clinically relevant regions in CT scans. The proposed framework provides an accurate, scalable, and clinically interpretable solution for automated lung cancer detection.