Abstract Stroke is one of the most prominent health problems that significantly impacts individuals and societies. To improve prompt diagnosis and treatment of this disease, medical imaging technologies are essential tools that doctors rely on to identify the type and location of stroke accurately. In this study, a novel microwave imaging system based on microstrip antennas is proposed for stroke detection. The proposed system exploits differences in the dielectric characteristics of brain tissue to distinguish between ischemic and hemorrhagic strokes. An antenna is designed to operate within the X band and is positioned near the brain. The antenna has a directional radiation pattern, and the peak gain is 7 dBi. A voxel-based human head phantom is used in CST Studio Suite to simulate realistic brain tissue. An imaging system is constructed comprising six antennas distributed around the head model. Four of these antennas are arranged horizontally around the circumference of the head, while the other two are positioned on the top of the head, ensuring three-dimensional coverage encompassing the sides and upper region of the brain. S-parameter data are generated from simulated scenarios involving 702 samples, which included normal brains as well as ischemic and hemorrhagic strokes placed at various locations within the brain. These samples are obtained across 13 different human voxel-based phantom models and analyzed using multiple machine learning algorithms. The Random Forest algorithm achieved the best classification accuracy of 97.29%, demonstrating the effectiveness of integrating microwave imaging with machine learning for non-invasive, cost-effective, and time-efficient stroke diagnosis. In addition to Random Forest, both XGBoost and LightGBM demonstrated prominent performance, further highlighting tree-based ensemble learning methods’ superior capability and robustness for accurately classifying stroke types in microwave imaging applications.
Mohammed et al. (Thu,) studied this question.