Artificial Enhanced MRI Imaging Generative Model

Abstract

[background]Breast MRI screening for breast cancer has been recommended for those with BRCA1/2 mutations. However, there are disadvantages having an annual MRI scan. The MRI imaging model which generated enhanced MRI images from the simple MRI without contrast agent was constructed. [materials&methods]Of the contrast-enhanced MRI images of the cases diagnosed as primary breast cancer, fat suppression T1-weighted images(T1), fat suppression T2-weighted images(T2), diffusion weighted images(DWI) and contrasted early-phase images(early phase) in which cut planes match in all types were used. U-Net algorithm was trained to estimate the early phase from T1, T2, and DWI. [result]The mean squared error was decreased down to 264. The peak signal-to-noise ratio was 55.1 dB, indicatig that the images were properly generated. [discussion]MRI screening can be performed more safely and efficiently without contrast agent.