Comparative Analysis of Deep Learning Models for Brain Tumor Detection Using Transfer Learning

One of the most dangerous, prevalent, and potentially fatal diseases is a brain tumor. If not diagnosed earlier, it might lead to a serious illness that is incurable. The most typical tumor types are gliomas (Gl), meningiomas (Me), and pituitary (Pi) tumors. One of the methods to diagnose this illness is facilitated by computer-assisted imaging. For MRI classification, several different methods have been suggested in the past, ranging from traditional techniques to cutting-edge Deep Learning (DL). Nevertheless, a substantial amount of data and processing power are needed to completely train the deep learning model. To overcome this issue, transfer learning is used to train the models. Transfer Learning (TL) aims to take the information acquired from one activity and use it to enhance learning for a different but related task. This article aims to assess how effective some of the pre-trained models are in classifying brain MRI. A fine-tuned version of the pre-trained models like Xception, DenseNet121, VGG16, VGG19, ResNet50, and InceptionV3 are used, and the performance of each of them is analyzed. Every model is evaluated on a brain MRI dataset from Kaggle. Performance indicators such as specificity, precision, sensitivity, accuracy, and F1- score are accustomed to assess the model’s predictive power. The VGG19 model delivers the best outcomes of all the TL models used in this research, with a 97% accuracy rate. The model achieved an F1-score of 97%, a precision of 97%, a sensitivity of 96%, and a specificity of 99%. ResNet50 is the second-best performing model with an F1-score of 96%, a sensitivity of 97%, an accuracy of 96%, a precision of 96%, and a specificity of 99%. The many categories of brain tumors, such as gliomas, meningiomas, pituitary tumors, and those without a tumor, are properly categorized using this approach.