This study presents a hybrid deep learning model, HHO-LOA-DCNN-LSTM, optimized using Horse Herd Optimization (HHO) and Lion Optimization Algorithm (LOA), for classifying lung cancer from CT scans. The model combines a Deep Convolutional Neural Network (DCNN) for feature extraction with Long Short-Term Memory (LSTM) networks to capture sequential dependencies, with hyperparameters fine-tuned via the hybrid optimizer to improve accuracy, convergence, and generalization. Evaluated on benchmark datasets and an external cohort, the model achieved superior performance metrics, including an accuracy of 99.65%, precision of 99.75%, recall of 99.23%, and robust calibration scores, outperforming several state-of-the-art DL models such as ConvNeXt, EfficientNet, and Vision Transformer. The study also addressed class imbalance using SMOTE, class weighting, data augmentation, and stratified cross-validation, ensuring reliable detection of malignant nodules.

Extensive ablation studies confirmed the critical role of each component—DCNN, LSTM, and hybrid optimization—in achieving high accuracy, while learning curves demonstrated stable convergence without overfitting. The model showed computational efficiency with lower CPU utilization and inference time compared to alternatives, making it suitable for clinical deployment. Failure mode analysis highlighted common misclassifications and proposed uncertainty-aware reporting to improve clinical safety. Despite limitations related to input quality and computational demands, the model’s strong generalizability and performance suggest it is a promising tool for early, accurate, and scalable lung cancer diagnosis from CT imaging, with future work aimed at reducing resource requirements and exploring ensemble methods.