Data Augmentation Method for Classification of Lymph Node Metastasis using Diffusion Model

Automatic judgment of the presence or absence of cancer metastasis to lymph nodes is an important topic in cancer staging to reduce the burden of medical doctors to check a large number of CT images. Therefore, the machine learning approach becomes helpful and desired. However, using CT images for machine learning requires annotation by a special radiologist, and it is difficult to prepare a large dataset. In recent years, many approaches to the generation of images by machine learning have been proposed and applied to medical images. In this paper, we propose a method to generate images around a lymph node in an arbitrary metastatic state using a diffusion model, which is a generative model that is gaining attention on behalf of GAN. Experiments were also conducted to learn the classification of cancer metastasis using the generated images for data augmentation, showing that the generated images are effective in improving classification performance.