Brain tumor magnetic resonance images classification based machine learning paradigms.

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Brain tumors are a significant global health concern, with symptoms varying based on location and type.
• Common brain tumors include glioma, meningioma, and pituitary adenoma, originating from different neural or glandular tissues.

Purpose of the Study:

• To develop and evaluate a feature engineering approach for classifying four types of brain MRI scans: glioma, meningioma, pituitary tumor, and no-tumor.
• To compare the performance of five machine learning classifiers for accurate brain tumor detection.

Main Methods:

• Extracted handcrafted features (Histogram of Oriented Gradients, Local Binary Pattern, Grey Level Co-occurrence Matrix) from MRI scans.
• Employed feature fusion techniques to augment the feature vector dimensionality.
• Utilized five machine learning classifiers: Support Vector Machine, K-Nearest Neighbor (KNN), Naive Bayes, Decision Tree, and Ensemble classifier.

Main Results:

• The Fine KNN classifier demonstrated superior performance, achieving 91.1% accuracy and an Area Under the Curve (AUC) of 0.95 for classifying the four MRI categories.
• The proposed feature engineering method, implemented with Fine KNN, reached 91.1% accuracy and 0.96 AUC.

Conclusions:

• The Fine KNN classifier, combined with feature engineering and fusion, provides an effective method for brain tumor classification from MRI.
• This approach is suitable for integration into low-end devices, offering an advantage over computationally intensive deep learning models.