Brain tumor classification utilizing pixel distribution and spatial dependencies higher-order statistical...

Area of Science:

• Medical Imaging Analysis
• Machine Learning in Healthcare
• Computational Neuroscience

Background:

• Accurate brain tumor diagnosis is critical for effective treatment and improved patient outcomes.
• Manual analysis of Magnetic Resonance Imaging (MRI) data is time-consuming and challenging.
• There is a need for automated, reliable machine learning (ML) methods for brain tumor detection.

Purpose of the Study:

• To propose a novel, comprehensive ML approach for identifying and classifying abnormal brain MR images.
• To accurately diagnose three common brain tumors: glioma, meningioma, and pituitary tumor.
• To enhance the diagnostic accuracy and efficiency of brain tumor detection using advanced ML techniques.

Main Methods:

• Feature extraction using 1st-order, 2nd-order, and Discrete Wavelet Transform (DWT) statistics.
• Handling missing data with KNNImputer and feature selection/dimensionality reduction using ExtratreesClassifier and PCA.
• Training and evaluating seven ML models (RF, GB, CB, SVM, LGBM, DT, LR) with k-fold cross-validation.
• Utilizing Explainable AI (XAI) for transparent model evaluation and insights.

Main Results:

• The proposed comprehensive approach achieved the highest accuracy, precision, recall, F1 score, MCC, Kappa, AUC-ROC, and R2.
• The method demonstrated the lowest loss among the seven evaluated machine learning models.
• The model's effectiveness was proven on the Figshare MRI brain image dataset.

Conclusions:

• The developed ML approach offers a highly effective and accurate method for brain tumor classification from MRI data.
• The study highlights the potential of advanced feature extraction, selection, and ML models in medical diagnostics.
• The findings support the applicability of this approach in various analytical tasks using publicly available datasets.