StyleGAN3-Based Lung Nodule Sample Generation and Classification Study

Lung cancer is one of the cancers with the highest mortality rate in the world, while lung nodules are potential manifestations of lung cancer, and categorizing lung nodules in CT images is a critical step for early detection of lung cancer. Although deep learning methods have shown great potential in classifying lung nodules, they still face challenges such as small datasets, imbalanced data categories, and poor generalization performance. In this paper, we use the StyleGAN3 model to generate lung nodule samples and combine two pre-trained classifiers to test and analyze them for different data enhancement strategies and proportions of generated data. The experiments show that the classification ACC accuracy has reached a level comparable to that of the real dataset when only using an equal amount of generated samples as the real dataset. Although the generated data cannot completely replace the real data, the appropriate introduction of generated samples is an effective and feasible data enhancement strategy when the dataset is small.