This study by Wang et al. presents RISNet, a variable multi-modal image feature fusion adversarial neural network designed to generate specific dMRI images of macaque brains, specifically to compensate for the insufficient number of lower-b-value images in existing datasets. The research was conducted by feeding T1 images and higher-b-value images into the RISNet network, which uses a rapid insertion structure (RIS) to combine multi-modal features and then passes them through a general residual decoding structure. The methodology involved multi-center joint training using four macaque dMRI datasets from the PRIMatE Data Exchange (PRIME-DE). The experimental results showed that RISNet outperformed the other comparison methods, improving the Peak Signal-to-Noise Ratio (PSNR) by 1.8211 on average and the Structural Similarity Index (SSIM) by 0.0111. RISNet also reached the best results across all datasets when using both T1 and high-b images as inputs. Diffusion Tensor Imaging (DTI) estimation experiments highlighted RISNet's effectiveness, as it reduced noise and better preserved the texture details of Fractional Anisotropy (FA) images, and Xtract tracking results closely resembled the original dMRI data.