A Machine Learning Classification Approach for Solving Biomedical Inverse Scattering Problem

This article proposes a U-Net convolution neural network (U-net CNN) with segmentation capability, to classify the types of human tissue in the biomedical inverse scattering scenario. The inverse scattering imaging algorithm, subspace-based optimization method (SOM) is firstly used to obtain distribution of dielectric permittivity of human tissues. The obtained result is input into the pre-trained U-Net CNN, to output the precise segmentation masks (classification images). Numerical results using synthetic data is used to validate the feasibility of precise classification of human tissues imaging.