Lung Cancer Detection by Combination of Capsule Neural Network and Enhanced Walrus Optimization Algorithm

Wenna Sha¹Li Ning¹Xiaowei Ding^2*

• ¹Hangzhou First People's Hospital, Hangzhou, Zhejiang Province, China
• ²Westlake University, Hangzhou, China

Lung cancer is one of the most common cancers in the world and is the leading cause of cancer-related death, underscoring the need for accurate and timely diagnostics. In this study, a new method is proposed to detect lung cancer from CT scans using a Capsule Neural Network (CapsNet) combined with an Enhanced Walrus Optimization Algorithm (EWOA). The proposed model effectively uses the advantages of CapsNet in the representation of meaningful features of lung images, as well as optimizing all of its parameters based on the EWOA. Whereas this metaheuristic algorithm is based on the walrus foraging behavior added with dynamic inertia weight and chaotic maps to help in distinctive searching ability in the solution space and avoid local optima. The performance of the proposed CapsNet/EWOA model was evaluated on the IQ-OTH/NCCD dataset and was compared against several existing methods, which included Deep Neural Networks (DNN/ES), XGBoost, a dung beetle optimizer-enhanced deep feature fusion model (DBOEDFF-LCC), and an enhanced AlexNet architecture combined with Support Vector Machines (SVM-LungNet). Experimental results show that the CapsNet/EWOA model outperforms a variety of metrics. Namely an accuracy of 92.105%, precision of 90.098%, recall of 94.920%, and F-beta score of 93.155%. The model demonstrates a smaller standard deviation for these metrics using k-fold cross-validation compared to other models, suggesting consistent performance across folds. Our CapsNet/EWOA model has less accuracy (than some of the advanced models), but its consistency and robustness indicate that it is a promising tool for lung cancer diagnosis.