Learning-Assisted Multimodality Dielectric Imaging

We introduce a convolutional neural network (CNN)-assisted dielectric imaging method, which uses CNN to incorporate the abundant image information from magnetic resonance (MR) images into the model-based microwave inverse scattering imaging process and generate high-fidelity dielectric images. A CNN is designed and trained to learn the complex mapping function from MR T1 images to dielectric images. Once trained, the new patients' MR T1 images are fed into the CNN to generate predicted dielectric images, which are used as the starting image for the microwave inverse scattering imaging. The CNN-predicted dielectric image, containing abundant prior information from MR images, significantly reduces the non-linearity and ill-posedness of the inverse scattering problem. We demonstrate the application of the proposed method to recover human brain dielectric images at 4 and 2 mm resolution with single-frequency and multifrequency microwave measurements. The reconstructed brain dielectric images with the proposed method show significant improvements in image quality compared with images reconstructed with no assistance from MR and CNN.