An advanced hybrid deep learning framework for high-precision brain tumor detection and classification in MRI scans.
Authors
Affiliations (5)
Affiliations (5)
- School of Computing Science and Engineering, Galgotias University, Greater Noida, Uttar Pradesh, 203201, India.
- Department of Communication Technology and Network, Faculty of Computer Science and Information Technology, University Putra Malaysia (UPM), Serdang, Selangor, 43400, Malaysia.
- Department of Mathematics, Faculty of Mathematics and Natural Sciences, University of Jember, Jember, Indonesia.
- Manipal Institute of Technology Bengaluru, Manipal Academy of Higher Education, Manipal, India. [email protected].
- Department of Computer Science and Engineering, KG Reddy College of Engineering and Technology, Hyderabad, Telangana, 501504, India.
Abstract
Early and accurate identification of brain tumors from magnetic resonance imaging (MRI) is essential for timely clinical intervention; however, manual interpretation remains time-consuming and dependent on expert analysis. The study proposes MultiAttenNet, a hybrid deep learning framework that integrates multi-scale convolutional neural networks (CNNs) for hierarchical feature extraction and Transformer-based attention mechanisms for global contextual learning, within a semi-supervised learning paradigm. The multi-scale feature extraction improves robustness in detecting tumors of varying sizes and irregular structures, while the adaptive attention module dynamically emphasizes diagnostically relevant regions to enhance localization and reduce false positives. A consistency-based semi-supervised learning scheme enables effective training using limited labeled data alongside unlabeled samples, improving generalization across diverse clinical scenarios. The framework is evaluated using the BraTS 2023 for glioma segmentation and publicly available datasets such as the Figshare Brain Tumor Dataset for multi-class tumor classification (glioma, meningioma, and pituitary tumors), MultiAttenNet achieves accuracy, sensitivity, specificity, and false-positive rates of 98.4, 96.8, 99.2 and 1.3%, respectively, outperforming existing state-of-the-art approaches. The proposed framework provides a scalable and efficient solution for real-time clinical brain tumor diagnosis and supports reliable automated decision-making in neuro-oncology.