CLASSIFICATION OF BRAIN TUMORS WITH DEEP LEARNING MODELS

Abstract

This study aims to present a comparative analysis of existing (state-of-the-art) deep learning models to identify early detection of brain tumor disease using MRI (Magnetic Resonance Imaging) images. For this purpose, GoogleNet, Mobilenetv2, InceptionV3, and Efficientnet-b0 deep learning models were coded on the Matlab platform and used to detect and classify brain tumor disease. Classification has been carried out on the common Glioma, Meningioma, and Pituitary brain tumors. The dataset includes 7022 brain MRI images in four different classes, which are shared publicly on the Kaggle platform. The dataset was pre-processed and the models were fine-tuned, and appropriate parameter values were used. When the statistical analysis results of the deep learning models we compared were evaluated, the results of Efficientnet-b0 (%99.54), InceptionV3 (%99.47), Mobilenetv2 (%98.93), and GoogleNet (%98.25) were obtained, in the order of success. The study results are predicted to be useful in offering suggestions to medical doctors and researchers in the relevant field in their decision-making processes. In particular, it offers some advantages regarding early diagnosis of the disease, shortening the diagnosis time, and minimizing human-induced errors.