Metaheuristic optimization of deep CNNs for multi-class diagnosis of cervical cancer and lymphoma

Accurate and early diagnosis of cancer is critical for determining effective treatment strategies and improving patient survival rates. However, automated multi-class cancer detection remains an enormous clinical and computational challenge due to the high visual heterogeneity within specific cancer classes and the morphological similarities across different types of malignancies. Deep convolutional neural networks (CNNs), particularly the VGG-16 architecture, offer robust feature extraction capabilities for medical imaging; yet, their diagnostic performance is heavily restricted by suboptimal hyperparameter tuning and inefficient feature utilization. To solve this problem, this article proposes a comprehensive, dual-strategy deep learning framework that integrates both pre-trained and fine-tuned VGG-16 models with six nature-inspired metaheuristic optimization algorithms. By employing the Whale Optimization Algorithm (WOA), Grey Wolf Optimizer (GWO), Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Ant Colony Optimization (ACO), and Modified PSO (MPSO), the framework autonomously optimizes critical hyperparameters to maximize classification accuracy. The proposed methodology was rigorously evaluated on two complex imaging datasets: a five-class dataset for cervical cancer (a leading global cause of female cancer-related mortality) and a three-class dataset for lymphoma (a complex malignancy of the lymphatic system). The experimental results demonstrated that integrating pre-trained VGG-16 networks with metaheuristic optimizers significantly outperformed baseline models across both datasets. Notably, the Whale Optimization Algorithm (WOA) exhibited superior performance, achieving up to 100% in accuracy, precision, recall, and specificity during the testing phase for both datasets. These findings confirm that optimizing deep CNNs with metaheuristic algorithms provides a highly adaptable, reliable, and precise framework capable of resolving the complexities of high-dimensional multi-class cancer diagnosis.