Brain tumor classification from MRI images using a multi-scale channel attention CNN integrated with SVM

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Medicine
• Neuropathology

Background:

• Accurate brain tumor classification from MRI is crucial but challenging.
• Traditional Convolutional Neural Networks (CNNs) exhibit limitations in feature extraction for medical images.
• Suboptimal recognition performance hinders the clinical application of automated methods.

Purpose of the Study:

• To propose a novel and enhanced framework for brain tumor classification using MRI.
• To overcome the feature extraction limitations of traditional CNNs.
• To improve the accuracy and robustness of automated brain tumor detection.

Main Methods:

• Developed a Multi-Scale Channel Attention CNN integrated with Support Vector Machine (MCACNN-SVM).
• Employed multi-scale convolutional kernels for hierarchical spatial feature extraction.
• Integrated a channel attention mechanism for adaptive feature enhancement.
• Utilized a grid-search optimized SVM classifier for improved decision boundaries.
• Implemented cosine annealing with warm restarts for accelerated convergence and generalization.

Main Results:

• The MCACNN-SVM framework demonstrated competitive performance on a brain tumor MRI dataset.
• Achieved high scores in accuracy, precision, recall, and F1-score.
• Exhibited strong robustness and generalization capabilities for practical applications.

Conclusions:

• The proposed MCACNN-SVM framework offers a significant advancement in automated brain tumor classification from MRI.
• The integration of multi-scale features, channel attention, and SVM optimization enhances classification performance.
• The framework shows potential for real-world clinical application due to its robustness and generalization ability.