An advanced hybrid deep learning framework for high-precision brain tumor detection and classification in MRI scans

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Neuro-oncology

Background:

• Manual interpretation of MRI for brain tumors is time-consuming and requires expert analysis.
• Accurate and early detection of brain tumors is critical for effective clinical intervention.

Purpose of the Study:

• To introduce MultiAttenNet, a hybrid deep learning framework for automated brain tumor identification and classification from MRI data.
• To enhance the accuracy and efficiency of brain tumor diagnosis using artificial intelligence.

Main Methods:

• Developed a hybrid deep learning framework (MultiAttenNet) integrating multi-scale CNNs and Transformer-based attention mechanisms.
• Employed a semi-supervised learning paradigm with consistency-based training on limited labeled and abundant unlabeled MRI data.
• Utilized multi-scale feature extraction for robust detection of diverse tumor sizes and structures, and attention modules for precise localization.

Main Results:

• MultiAttenNet achieved high performance metrics: 98.4% accuracy, 96.8% sensitivity, 99.2% specificity, and 1.3% false-positive rate on benchmark datasets.
• The framework demonstrated superior performance compared to existing state-of-the-art methods in glioma segmentation and multi-class tumor classification.
• The semi-supervised approach improved generalization across various clinical scenarios with limited labeled data.

Conclusions:

• MultiAttenNet offers a scalable and efficient solution for real-time clinical brain tumor diagnosis.
• The proposed framework supports reliable automated decision-making in neuro-oncology, potentially improving patient care.
• This AI-driven approach addresses the limitations of manual MRI interpretation for brain tumor detection.