Comparative Analysis of Fine-Tuned MobileNet versions to Detect Brain Tumors from MRI images

A prominent area of research in the realm of medical imaging is the categorization of brain tumors into multiple classes. Researchers have devised a variety of strategies to improve categorization accuracy. Brain tumor diagnosis and classification using standard medical data analysis is a difficult and time-consuming job. Biomedical investigation demonstrates that systematic categorization with human involvement may result in erroneous diagnosis and prognosis. One of the most remarkable approaches for dealing with this issue is to utilize deep learning algorithms. This study examines an intra-version comparison of MobileNet models with the goal of identifying brain tumors using MRI images. This research aims to evaluate the effectiveness of each version of MobileNet —MobileNetV1, MobileNetV2, and MobileNetV3, to pinpoint the three distinct forms of brain tumors which are pituitary, glioma, and meningioma. Multiple evaluation metrics, including sensitivity, specificity, F1-score, accuracy, and negative likelihood ratio have been used to evaluate the effectiveness of these models. According to a comparative analysis of these models, MobileNetV2 tends to perform better than other versions achieving a train and test accuracy of 99.73% and 97.86%.