T1-weighted MRI-based brain tumor classification using hybrid deep learning models.

Area of Science:

• Neuroimaging and Artificial Intelligence
• Medical Image Analysis and Deep Learning

Background:

• Brain health is crucial for cognitive function, with Magnetic Resonance Imaging (MRI) vital for diagnosis.
• Deep learning models are increasingly used for high-performance image processing in healthcare.
• Accurate brain tumor classification is essential for effective treatment planning.

Purpose of the Study:

• To classify brain tumors (glioma, meningioma, pituitary) using the U-Net architecture on MRI scans.
• To evaluate the performance of various convolutional neural networks (CNNs) with transfer learning.
• To assess the diagnostic potential of U-Net for neuroimaging.

Main Methods:

• Application of the U-Net segmentation architecture to brain MRI scans for tumor classification.
• Utilizing transfer learning with CNNs like Inception-V3, EfficientNetB4, and VGG19.
• Performance evaluation using metrics including accuracy, F-score, recall, and precision.

Main Results:

• U-Net achieved superior performance with 98.56% accuracy, 99% F-score, 99.8% AUC, and 99% recall/precision.
• Cross-dataset validation demonstrated robust performance with 96.01% accuracy on an external cohort.
• U-Net proved effective for brain tumor segmentation and classification.

Conclusions:

• U-Net demonstrates high effectiveness for accurate brain tumor segmentation and classification in neuroimaging.
• The study highlights the potential of U-Net and transfer learning to improve diagnostic accuracy.
• Findings support enhanced clinical decision-making and improved patient care in neurooncology.