A Computationally Efficient and Improved Brain Tumor Recognition System by MRI-Segmentation Integrated Classification Network.
Authors
Affiliations (3)
Affiliations (3)
- College of Engineering, United Arab Emirates University, Asharej, Al Ain, Abu Dhabi, UAE.
- College of Engineering, United Arab Emirates University, Asharej, Al Ain, Abu Dhabi, UAE. [email protected].
- College of Information Technology, United Arab Emirates University, Asharej, Al Ain, Abu Dhabi, UAE.
Abstract
Brain tumors present a major global health concern, and a precise diagnosis is essential for proper treatment. Many existing MRI-based machine learning approaches focus solely on segmentation or classification, rather than addressing both tasks together. To bridge this gap, a unified deep learning model is designed to perform tumor segmentation and multiclass classification within a single architecture. By integrating a segmentation backbone with a dedicated classification head, the framework simultaneously captures anatomical details and tumor-specific features. On three public datasets, the proposed model achieved up to 99.6% classification accuracy and 0.935 Dice score, with average performance of 96.9% accuracy and 0.966 F1-score on Dataset 1, 99.4% accuracy and 0.984 F1-score on Dataset 2, and 98.2% accuracy and 0.982 F1-score on Dataset 3. Thus, evaluated on these publicly available brain MRI datasets, the proposed network outperforms CNN-based baselines and recent attention-based models, delivering improved tumor localization and classification accuracy within an integrated segmentation-classification framework, while maintaining computational efficiency. These results highlight its strong promise for supporting clinical decision-making in brain tumor diagnosis and treatment planning.