Microwave Imaging-Based Stroke Classification Using a Spatio-Temporal Convolutional Neural Network

Microwave imaging is gaining prominence as a diagnostic method for brain stroke owing to its non-invasive, non-ionizing, reliable, and wearable characteristics. Deep learning has demonstrated significant potential for abnormality detection in comparison to conventional and intricate techniques. However, deep learning systems necessitate considerable quantities of labelled RF data for training, which are frequently constrained in brain stroke diagnostic tasks, including detection, localization, and size classification. The dataset containing S-parameters from CST head model is collected by a microwave set up with two antennas. Then the collected S-parameters are applied as input to Spatial-Temporal Convolutional Neural Network (ST-CNN) model in the form scalogram images. It captures both spatial information and temporal interdependence. This study significantly improves diagnosis and continuous monitoring of brain stroke with an accuracy of 94.6%.