NEURO IMAGING REDEFINED: A DEEP LEARNING DRIVEN MICROWAVE IMAGING MODEL FOR STROKE LOCALIZATION AND SEGMENTATION

Abstract

Stroke is a major neurological disorder in which early detection, localization, and segmentation are essential for timely treatment and reducing permanent brain damage. Conventional imaging techniques such as Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) provide accurate diagnosis but are expensive, timeconsuming, and require specialized infrastructure. This work proposes a Deep Learning-Driven Microwave Imaging Model for Stroke Localization and Segmentation using non-invasive, non-ionizing, portable, and cost-effective microwave imaging. The proposed framework consists of microwave signal acquisition, preprocessing, image reconstruction, CNN-based feature extraction, and Attention-Based U-Net for precise lesion localization and pixel-level segmentation. Transfer learning is incorporated to improve model generalization, while the Adam optimizer and Dice Loss support effective training and segmentation. The system is evaluated using Accuracy, Precision, Recall, F1-score, Dice Coefficient, and Intersection over Union (IoU). The proposed framework aims to provide rapid, reliable, affordable, and accessible stroke diagnosis, particularly in emergency and resource-limited healthcare environments. Keywords: Stroke Detection, Microwave Imaging, Deep Learning, CNN, Attention U-Net, Lesion Segmentations.