An Intelligent and Lightweight Approach Based on MobilenetV2 Architecture for Identifying Brain Tumors

Integration of machine learning approaches has the potential to alleviate human error and reduce the time required to diagnose brain tumors by assisting radiologists. The main focus of the existing studies is on developing a model that is as accurate as possible to perform such a task. On the other hand, a model's computational cost and image processing speed are not extensively examined. However, they are significant parameters for the model deployment in real-time. This study aims to close the gap by introducing MobileNetV2-0.5 as a lightweight, fast, and effective approach for identifying brain tumors using real-time Magnetic Resonance Imaging (MRI) images. The results indicated that the proposed approach successfully identified the tumors by 98.78% and detected the non-tumor cases by 99.75%. The computational cost and the processing speed have improved by around 50% compared to the original MobileNetV2 architecture. A similar improvement has also been observed when comparing the proposed approach with the models existing in the literature. Based on the results of the analysis, it is concluded that the proposed MobileNetV2-0.5 has the potential to identify brain tumors in real-time by deploying the model through embedded devices.