Brain Tumor Classification in MRI Images Using Combined Transfer Learning and Convolutional Neural Networks

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Early and accurate detection of brain tumors is crucial for effective patient treatment and management.
• Magnetic Resonance Imaging (MRI) is a primary modality for visualizing brain structures and identifying abnormalities.
• Deep learning offers potential for automating and improving the accuracy of diagnostic processes in medical imaging.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework for accurate brain tumor classification using MRI data.
• To compare the performance of the proposed Custom Convolutional Neural Network (CNN) against established pre-trained models and ensemble methods.
• To assess the efficiency and effectiveness of a task-specific deep learning architecture for automated brain tumor detection.

Main Methods:

• A novel Custom CNN architecture was designed and implemented for brain tumor classification.
• The Custom CNN was evaluated independently against six pre-trained models: InceptionV3, EfficientNetV2L, ResNet152V2, Xception, VGG16, and MobileNetV2.
• Three separate ensemble models were constructed to investigate the impact of model combination on classification performance.
• Experiments were conducted using the Kaggle-Multiclass brain MRI dataset.

Main Results:

• The proposed Custom CNN achieved the highest classification accuracy at 99.54%.
• The Custom CNN demonstrated superior performance compared to individual pre-trained models and ensemble approaches due to its domain-specific architecture and feature learning.
• Among pre-trained models, EfficientNetV2L (99.47%) and InceptionV3 (99.39%) showed competitive results.
• The best ensemble model reached 99.47% accuracy, but the Custom CNN achieved better performance with greater computational efficiency (0.57M parameters).

Conclusions:

• The developed Custom CNN offers a highly effective and computationally efficient solution for automated brain tumor classification from MRI scans.
• The proposed model surpasses the performance of widely used pre-trained deep learning models and ensemble methods without added complexity.
• These findings suggest significant potential for the Custom CNN in clinical settings for early and accurate brain tumor detection, pending external validation.