Deep Learning-Driven Liver Cancer Detection Enhanced by Particle Swarm Optimization (PSO) Algorithm

The Abstract— The early and accurate detection of liver cancer is paramount for effective diagnosis and treatment planning. This paper proposes a robust computer-aided diagnosis (CAD) framework for the classification of liver tumors by integrating a Convolutional Neural Network (CNN) with Particle Swarm Optimization (PSO). The implemented MATLAB pipeline begins with the acquisition of liver images, which are first enhanced through pre-processing techniques to improve quality and reduce noise. Subsequently, tumor regions are precisely delineated using Fuzzy C-Means (FCM) clustering for segmentation. Critical texture features are then extracted from the segmented regions using the Gray Level Co-occurrence Matrix (GLCM). The PSO algorithm is employed to optimize this feature set, selecting the most discriminative attributes to improve classification efficiency. The optimized features are used to train a CNN classifier for the binary classification of tumors into benign or malignant categories. The proposed model was rigorously evaluated, demonstrating high performance with accuracies of 94.99% for benign and 94.19% for malignant cases. Metrics including sensitivity, specificity, and precision further confirm the system's robustness and reliability. The synergy of PSO-based feature optimization and deep learning classification presents a powerful and efficient tool for assisting clinicians in making informed diagnostic decisions.