Figure 6:

T_1rho maps reconstructed using SuperMAP and MANTIS methods with down-sampled data (top) and corresponding error maps compared to the reference map generated with fully-sampled data (bottom). SuperMAP is a novel deep learning framework that directly converts a series of undersampled (both in k-space and in parameter dimension), parameter-weighted images (e.g. T_1rho or T₂-weighted images) into quantitative maps (e.g. T_1rho or T₂ maps), bypassing the conventional exponential fitting procedure. This network incorporates patch wise training with the entire image as the backward cycle (model-data) for consistency. Maps with joint acceleration factor (J-AF) of 16, 20 and 24 using SuperMAP are demonstrated, which provide more superior performance compared to maps reconstructed with MANTIS using J-AF 16. NMSE, normalized mean squared error; PSNR, peak signal to noise ratio; SSIM, structural similarity index. MANTIS: Model-Augmented Neural neTwork with Incoherent k-space Sampling (66). Figures edited from reference (61) with permission.