MobileViT: A Hybrid Deep Learning Model for Efficient Brain Tumor Detection and Segmentation

Early and accurate diagnosis of brain tumors is crucial to the right management. Standard Convolutional Neural Networks (CNNs) has proved useful in medical image analysis; however, due to their inherent high computational requirements, they cannot run efficiently on resource constraint systems including the mobile. In this article, we explore the deployment of MobileViT as a condense model that integrates convolutional layers and ViT for forecasting as well as delineation of brain tumors from MRI scans. MobileViT is developed with efficiency as well as performance in mind and they are best fit for use in mobile end devices especially in telemedicine services. MobileViT is evaluated in association with the BraTS 2021 dataset for the classification and segmentation of brain tumors in terms of accuracy, computational complexity, and the ability to learn specific characteristics as well as overall features within an image. The study shows that MobileViT is best suited for achieving the right balance of the model's complexity, performance, and model inference time; it achieves a Die Coefficient of 0.87 in segmentation tasks and 40% less inference time than CNN models. Further, recommendations are proposed to enhance the practicalities of application of telemedicine systems so that accurate diagnosis for patients in other low resource settings can be made in real-time, a measure with big implications for quicker identification or patients problems, and improved patient prognosis.