Performance optimisation of deep learning models using majority voting algorithm for brain tumour classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Brain tumor grading is crucial for treatment but current biopsy methods are invasive and risky.
• There is an urgent need for non-invasive techniques for accurate brain tumor grading.

Purpose of the Study:

• To develop and evaluate a non-invasive brain tumor grading method using magnetic resonance imaging (MRI) data.
• To leverage deep learning (DL) and machine learning (ML) techniques for improved tumor classification.

Main Methods:

• Four datasets were used to train and test five DL models (AlexNet, VGG16, ResNet18, GoogleNet, ResNet50) and five ML models (SVM, KNN, Naive Bayes, Decision Tree, LDA).
• A majority voting (MajVot)-based ensemble algorithm was proposed to enhance the performance of combined DL and ML models.
• The methodology involved five-fold cross-validation for robust model evaluation.

Main Results:

• The DL-based MajVot algorithm demonstrated an average accuracy improvement of 1.04%–2.67% across four datasets compared to individual DL models.
• A significant 10.12% average accuracy improvement was observed for the DL method over ML methods.
• The MajVot algorithm also showed improved performance on synthetic face image classification tasks.

Conclusions:

• The proposed MajVot algorithm shows promise for non-invasive brain tumor classification.
• This ensemble approach effectively combines the strengths of multiple DL and ML models for enhanced diagnostic accuracy.