Machine-Learning-Based Inversion Scheme for Super-Resolution Three-Dimensional Microwave Human Brain Imaging

To realize efficient three-dimensional (3-D) super-resolution whole brain microwave imaging, a new machine- learning-based inversion method with a resolution enhancement technique is proposed. It consists of three parts: a parallel semiconnected backpropagation neural network (SJ-BPNN) scheme, a U-Net scheme, and a modified Akima piecewise cubic Hermite interpolation (MAPCHI) scheme. The parallel SJ-BPNN scheme is first employed to map the measured scattered field data to the preliminary electrical properties' distribution of human brain. Then, U-Net is used to improve the quality of these preliminary reconstruction results. Finally, the MAPCHI scheme is adopted to greatly improve the resolution of reconstruction results with a very low computational cost. Numerical examples of a normal human brain and a human brain with abnormal scatterers show that the proposed method can achieve accurate high-resolution human brain imaging with 1024 x 1024 x 1024 voxels with a very low computational cost.