In the sphere of diagnosing stroke, a life-threatening condition that stands as the second leading cause of death globally, the intricacy of the brain—comprising the cerebrum, cerebellum, and brain stem, underscores the urgent need for early detection and treatment to stave off further cerebral damage and boost patient recovery. In this study, using 2501 CT images of brain strokes and a range of pretrained convolutional neural networks, including MobileNetV2, DenseNet169, and ResNet101, the MobileNetV2 model outperformed others in accuracy, precision, F1-score, and kappa, with scores of 95.80%–96.62%, 95.82%, 95.81%, and 91.11%, respectively. MobileNetV2's robust performance, along with its adaptability and computational efficiency, make it an ideal fit for Internet of Things applications. The comparison of MobileNetV2 with models such as VGG-19, characterized by higher computational demands and slower processing, highlights the critical role of MobileNetV2 in advancing swift and precise stroke diagnoses and enhancing patient prognoses.