CART-ANOVA-Based Transfer Learning Approach for Seven Distinct Tumor Classification Schemes with Generalization...

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Convolutional Neural Networks (CNNs) are crucial for brain tumor detection via deep transfer learning.
• Key challenges include hyperparameter optimization and model generalization.
• Existing methods like grid or random search may not fully capture hyperparameter interactions.

Purpose of the Study:

• To introduce a novel CART-ANOVA hyperparameter tuning framework for brain tumor classification.
• To enhance the accuracy, robustness, and generalization of deep learning models.
• To integrate statistical significance testing with hyperparameter optimization.

Main Methods:

• A ResNet18-based knowledge transfer learning (KTL) model was utilized.
• Hyperparameters were optimized using the proposed CART-ANOVA framework.
• Model performance was validated on independent datasets and compared against other CNN models.

Main Results:

• Exceptional testing accuracy achieved: 99.65% (4-class) and 98.05% (7-class) on dataset 1.
• High generalization maintained: 98.77% (4-class) and 96.77% (7-class) on dataset 2.
• The framework surpassed other pre-trained CNN models in performance.

Conclusions:

• The CART-ANOVA framework significantly improves brain tumor classification accuracy, robustness, and generalization.
• This approach offers enhanced diagnostic precision for AI-driven healthcare.
• The findings support advancements in medical imaging and treatment strategies.