Comparative Study of Multiple Models for Lung Cancer Detection in Chest CT Images

Lung cancer is among the most prevalent malignant tumors affecting the lungs. Traditional methods of medical imaging diagnosis heavily rely on the expertise and judgment of radiologists, thereby introducing elements of subjectivity and significant time expenditure. This study aims to compare the efficacy of various deep learning models in lung cancer CT image analysis, specifically focusing on the MedMamba model, several conventional Convolutional Neural Network (CNN) models, and Transformer models. Through rigorous experimentation and comparative analysis, this paper elucidates the strengths and limitations of these models and, for the first time, provides an in-depth exploration of the Mamba model’s application potential in lung cancer CT image recognition and diagnosis without necessitating pre-training on extensive datasets. The experimental outcomes indicate that the MedMamba model achieves remarkable accuracy and low computational complexity even in a non-pretrained state, whereas pre-training markedly enhances the performance of all models. This investigation not only aims to optimize existing models but also seeks to explore the future potential of the Mamba model, thereby contributing to advancements in lung CT image classification and enhancing the accuracy and efficiency of early lung cancer diagnosis.