Automated brain tumor detection using novel hybrid vision transformer and EfficientNetb4 models with comparative analysis

• Published in: Neural computing & applications Vol. 37; no. 22; pp. 18497 - 18525
• Main Authors: Karthik, S. Selva, Achuthan, V., Ananth, A. Dennis, Venkatesan, R., Rajakumaran, M., Markkandeyan, S.
• Format: Journal Article
• Language: English
• Published: London Springer London 01.08.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Artificial neural networks; Biopsy; Brain cancer; Brain damage; Cancer therapies; Classification; Clustering; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Damage detection; Data augmentation; Data Mining and Knowledge Discovery; Datasets; Deep learning; Glioma; Image processing; Image Processing and Computer Vision; Machine learning; Magnetic resonance imaging; Medical imaging; Original Article; Probability and Statistics in Computer Science; Proprietary; Support vector machines; Tissues; Tumors; MRI classification; EfficientNetB4; CNN (convolution neural network); Vision Transformer (ViT); Brain tumor detection
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-025-11335-x

A brain tumor is an abnormal growth or mass of cells in or around the brain. Early detection of brain tumors is imperative, impacting the quality of life and potential fatality. Prolonged undetected brain tumors can cause irreversible brain damage. Early detection enables medical intervention to prevent severe harm, preserving cognitive function and reducing permanent damage risk. These tumors come in a wide variety of sizes, locations, and other characteristics. When trying to locate cancerous tumors, magnetic resonance imaging (MRI) is a crucial tool. However, detecting brain tumors manually is a difficult and time-consuming task that might lead to inaccuracies. Many researchers investigated a variety of algorithms for detecting and classifying brain tumors that were both accurate and fast. Deep learning (DL) approaches have recently been popular in developing automated systems capable of accurately diagnosing or segmenting various tumors in less time. In this research, we propose several ways to detect brain cancer and tumors using computational intelligence and statistical image processing techniques. We use three different deep learning architecture models along with data augmentation and image processing to categorize brain MRI scan images into cancerous and non-cancerous types. We later conducted a comparative analysis of our models: EfficientNetB4, Vision Transformer (ViT) combined with EfficientNetB4 (a novel hybrid model), and a custom CNN model built from scratch. The experiment results demonstrate that all models achieved high accuracy and very low complexity rate. Specifically, EfficientNetB4 achieving 99.76%, 99.6% in Vision Transformer + EfficientNetB4, and scratch CNN achieved 97.25% accuracy. Our models require very less computational power and have much better accuracy results as compared to other pretrained models.