Brain tumor classification from MRI scans: a framework of hybrid deep learning model with Bayesian optimization and...

Area of Science:

• Medical Image Analysis
• Artificial Intelligence
• Oncology

Background:

• Brain tumor classification from MRI is challenging due to complex features and data imbalance.
• Current deep learning methods may overlook critical features, impacting diagnostic accuracy.
• Accurate classification is vital for timely and effective patient treatment.

Purpose of the Study:

• To propose an automated deep learning model for brain tumor classification using MRI.
• To develop an optimal information fusion framework to enhance classification performance.
• To address the challenge of imbalanced datasets in brain tumor imaging.

Main Methods:

• A sparse autoencoder network was used for data augmentation to resolve dataset imbalance.
• Two pretrained neural networks were optimized using Bayesian optimization.
• An improved Quantum Theory-based Marine Predator Optimization algorithm (QTbMPA) was employed for feature selection and fusion.

Main Results:

• The proposed framework achieved a high accuracy of 99.80% on an augmented dataset.
• Excellent performance metrics include 99.83% sensitivity and 99.83% precision.
• The QTbMPA algorithm effectively selected and fused relevant deep features for classification.

Conclusions:

• The developed deep learning model and fusion framework significantly improve brain tumor classification accuracy.
• The approach effectively handles imbalanced medical imaging datasets.
• This method offers a promising tool for accurate and automated brain tumor diagnosis.