This abstract presents a combined deep learning framework SuperMAP to generate MR parameter maps from very few subsampled echo images. The method combines deep residual convolutional neural networks (DRCNN) and fully connected networks (FC) to exploit the nonlinear relationship between and within the combined subsampled T1rho/T2 weighted images and the combined T1rho/T2 maps. Experimental results show that the proposed combined network is superior to single CNN network and can generate accurate T1rho and T2 maps simultaneously from only three subsampled echoes within one scan with results comparable to reference from fully sampled 8-echo images each for two separate scans.
This abstract and the presentation materials are available to members only; a login is required.