Breast Cancer Classification Using DDPM and Holographic Guidance Diffusion Framework

Breast cancer is a serious health threat to women. The early detection of breast cancer is crucial to improving survival rates and reducing the difficulty of treatment. In recent years, denoising diffusion probability models (DDPM) have performed well in image generation, but their application to breast cancer classification remains unknown. Additionally, conditional priors are often the most important source of information for diffusion models to solve problems, and fuzzy lesions or high-noise images pose a significant challenge for the model to extract accurate prior information. To address these issues, we propose a new classification model based on DDPM and combine it with the holographic guided diffusion framework (HGDF) to remove noise from images through specific guidance and make full use of HPC resources to improve the classification performance of breast cancer images. Results from the experiment indicate that the model has excellent accuracy in ultrasound images (BUS dataset) and mammography images (MAMMO dataset), with accuracy of 86.17 % ± 1.74% and 94.32% ± 2.31%, respectively.