Combined Machine Learning- Inversion Scheme for Super-Resolution 3-Dimensional Microwave Human Brain Imaging

Abstract

To realize three-dimensional (3-D) super-resolution whole brain imaging with a low high efficiency and computational cost, a new machine learning based-inversion method with the resolution enhancement technique is proposed in this work. This method consists three parts: a parallel semi-connected back-propagation neural network (SJ-BPNN) scheme, a U-Net scheme and a modified Akima segmented cubic Hermite interpolation (MASCHI) scheme. The parallel SJ-BPNN scheme is employed to map the measured scattering field data to the initial electrical properties distribution of human brain, firstly. Then, U-Net is used to improve the quality of preliminary reconstruction results obtained from SJ-BPNN. Finally, MASCHI scheme is adopted to greatly improve the resolution of reconstruction results with a very low computational cost. Numerical examples of normal human brain and human brain with abnormal scatterers show that the proposed model can achieve accurate human brain imaging with 1024×1024×1024 voxels resolution with a very low computational cost.