NeuroSight: A Deep‐Learning Integrated Efficient Approach to Brain Tumor Detection

Abstract

ABSTRACT Brain tumors pose a significant health risk and require immediate attention. Despite progress, accurately classifying these tumors remains challenging due to their location, shape, and size variability. This has led to exploring deep learning and machine learning in biomedical imaging, particularly in processing and analyzing Magnetic Resonance Imaging (MRI) data. This study compared a newly developed Convolutional Neural Network model to pre‐trained models using transfer learning, focusing on a comprehensive comparison involving VGG‐16, ResNet‐50, AlexNet, and Inception‐v3. VGG‐16 model outperformed all other models with 95.52% test accuracy, 99.87% training accuracy, and 0.2348 validation loss. ResNet‐50 model got 93.31% test accuracy, 98.78% training accuracy, and 0.6327 validation loss. The CNN model has a 0.2960 validation loss, 92.59% test accuracy, and 98.11% training accuracy. The worst model seemed to be Inception‐v3, with 89.40% test accuracy, 97.89% training accuracy, and 0.4418 validation loss. This approach facilitates deep‐learning researchers in identifying and categorizing brain cancers by comparing recent papers and assessing deep‐learning methodologies.

Keywords

• biomedical field
• brain tumor detection
• convolutional neural network
• recurrent neural network
• transformer learning
• visual geometry group