Revealing New Possibilities for Breast MRI Enhancement: Mamba-Driven Cross-Attention GAN with VMKANet

Breast cancer poses a serious risk to women’s health, making early detection critical. Dynamic Contrast-Enhanced MRI (DCE-MRI) aids diagnosis by using a contrast agent to highlight tissue details, but its cost and health risks limit accessibility. This study proposes a conditional Generative Adversarial Network (cGAN) to enhance MRI images, incorporating a novel generator (VMKANet, based on Vision Mamba and Kolmogorov-Arnold Network) and a discriminator (CAViT, using Vision Transformer with Cross Attention). To our knowledge, this is the first method to generate Breast DCE-MRI images from single-modality MRI data using Mamba. Our approach shows significant improvements in image quality, validated by strong Spearman correlation (R), high structural similarity index (SSIM), enhanced mutual information (MI), reduced normalized root mean square error (NRMSE), and lower symmetric mean absolute percent error (SMAPE). Results demonstrate that VMKANet and CAViT effectively enhance breast MRI images, promising safer and more affordable diagnostic capabilities.

Duke Scholars

Author Jing Cai Radiation Oncology