A Machine Learning-based Inverse Scattering Method for Biomedical Imaging Segmentation

This paper proposes a transformer-based neural network, for segmenting the images of human tissues obtained by inverse scattering method. Firstly, the distribution image of relative permittivity for the human tissue is obtained by the subspace-based optimization method (SOM). Then the obtained results are fed into a transformer-based network to output a segmentation mask. Numerical results verify that this method can get clear edges for different tissues, and it can achieve accurate classification for human tissue imaging.