Electrical Properties Tomography reconstruction technique is highly sensitive to noise, as it requires Laplacian calculations of phase data. To alleviate the noise amplification, large derivative kernels combined with image filters are used. However, this leads to severe errors at tissue boundaries. In this study, we employ a deep learning-based denoising network allowing for noise robust conductivity reconstructions obtained using smaller derivative kernels sizes. This comes with the intrinsic advantage of reduced boundary errors. The feasibility study was performed using cylindrical numerical simulations. Then, the proposed technique was tested using spin echo in-vivo data, and clinical patient data.
This abstract and the presentation materials are available to members only; a login is required.