Reconstruction and Classification of Brain Strokes Using Deep Learning-Based Microwave Imaging

Prompt identification of the type of brain stroke is a pivotal measure for medical practitioners in commencing therapeutic interventions for patients afflicted with stroke. In this paper, our purpose is to diagnose the type of stroke using high-quality images. This paper presents a novel methodology for image reconstruction using U-Net, followed by the classification of brain stroke type through a deep convolutional neural network. To achieve the intended objective, the initial step involves the utilization of the distorted Born iterative technique, employing a suitable initial guess, and a proper generalized Tikhonov regularization approach. The quality of the reconstructed images is high but not enough to correctly classify the type of brain stroke. To further enhance their quality, a U-Net is employed, which is trained using the reconstructed images obtained in the first step. The quality of the images results from U-Net is very high, but in a few situations, the brain stroke cannot be identified in the reconstructed images. So we propose a new deep convolutional neural network to classify the type of brain stroke, utilizing images reconstructed by U-Net. The simulations are conducted in two dimensions and across three levels of signal-to-noise ratio for the scattered field, specifically at 30, 20, and 10 dB. The results obtained indicate a significantly high rate of detection in identifying the type of strokes in all signal-to-noise ratios.