Microwave Diagnosis of Bone Fractures: an Artificial Intelligence-Based Approach

This paper evaluates the suitability of a Deep Neural Network (DNN) for diagnosing bone fractures through non-invasive radio frequency wave propagation. The DNN is trained using S-parameter profiles instead of X-ray images to avoid labeling and data collection challenges. The resulting network can classify diverse fracture types (normal, transverse, oblique, and comminuted) and simultaneously determine the size of cracks. Using a portable device, the proposed system provides fast preliminary diagnoses in emergency settings where radiologists are unavailable. The DNN was trained using human body models with varying tissue diameters to simulate different anatomical regions. Numerical results demonstrate successful training without overfitting, and experiments on sheep femur bones in liquid phantom validate the accuracy of fracture classification without harmful X-rays.