Deep and Reinforcement Learning Approach to Enhanced Microwave Imaging for Tumor Identification

In this work, a framework integrating Deep and Reinforcement Learning with microwave imaging techniques is presented to improve tumor identification. The proposed approach implements microwave imaging reconstruction by using the Quadratic Born Iterative Method (BIM) and Complex-Valued Neural Networks (CVNNs). Through the adoption of Deep QNetworks (DQNs) and Reinforcement Learning, an agent is implemented to learn and perfomr accurate tumor identification within breast images. This is confirmed by the achieved Intersection over Union (IoU) values greater than 0.9 in the discussed validation tests, thus revealing successfull applications in the framework of medical imaging.