This study investigates the effectiveness of different deep learning models in detecting brain strokes from imaging data.(Chandaran et al., n.d.) We employed four distinct algorithms—Convolutional Neural Network (CNN), ResNet, MobileNet, and RadiNet—on both normal and augmented image datasets. Our findings highlight a stark contrast in performance: while traditional CNNs achieved an accuracy of 96% on unaltered images, their effectiveness significantly decreased to 55% with augmented images. Conversely, MobileNet demonstrated remarkable robustness, maintaining a 99% accuracy rate across both datasets. ResNet also performed well, particularly with normal images at 98% accuracy. RadiNet, however, showed varied results with 90% accuracy on normal images and a decrease to 79% with augmented data. These results underline the potential and limitations of advanced neural networks in medical imaging, suggesting a pathway towards more accurate and reliable stroke detection methodologies.

KEYWORDS: Stroke Detection, Deep Learning, Convolutional Neural Networks, Image Augmentation, Medical Imaging.