Enhancing Lung Cancer Detection through a Novel CapsuleNet-ResNet Fusion Model: A Comparative Study on Accuracy and Robustness

Abstract

Lung cancer, a globally prevalent and highly lethal malignancy, requires precise and robust detection methods for timely intervention and treatment. This study presents a novel approach to improve lung cancer detection by combining CapsuleNet and ResNet architectures in a unique fusion model. The CapsuleNet-ResNet fusion model harnesses the distinct strengths of both architectures, merging CapsuleNet's capability to capture hierarchical feature relationships with ResNet's expertise in learning intricate patterns through residual learning. The methodology involves a diverse dataset of lung images encompassing various pathological conditions and stages of cancer progression. Through thorough preprocessing and augmentation, the dataset is prepared to ensure model generalization and resilience. Training incorporates an optimized fusion strategy, integrating CapsuleNet and ResNet at feature levels to facilitate seamless information exchange while preserving individual characteristics. Leveraging transfer learning and fine-tuning techniques, the fusion model is skillfully trained on complex lung cancer patterns. Rigorous cross-validation and standard performance metrics validate and assess the model, demonstrating superior performance compared to individual models and existing methods. The fusion model exhibits exceptional accuracy, sensitivity, and specificity across varied datasets and real-world scenarios. Comparative analysis with established methodologies emphasizes the fusion model's superiority and robustness, highlighting its potential as a reliable tool for early and precise lung cancer detection. In conclusion, the fusion of CapsuleNet and ResNet architectures signifies a promising advancement in lung cancer detection, offering more accurate, efficient, and scalable diagnostic tools for clinical applications.