Deep learning frameworks are emerging methods for solving ill-posed inverse problems in medical imaging, including Quantitative Susceptibility Mapping (QSM). Previously, U-net has been successfully trained on susceptibility maps to learn the dipole inversion process; however, susceptibility contrast loss was observed in iron-rich deep grey matter regions. In this study, we propose an enhanced deep learning network “xQSM” using the state-of-the-art Octave Convolution, which shows more accurate susceptibility contrasts than the original U-net in both simulated and in vivo datasets.
This abstract and the presentation materials are available to members only; a login is required.