Robust brain MRI image classification with SIBOW-SVM.

Area of Science:

• Neuro-oncology
• Medical Imaging Analysis
• Machine Learning in Healthcare

Background:

• Primary Central Nervous System tumors are aggressive, necessitating early and accurate detection.
• Magnetic Resonance Imaging (MRI) is crucial for brain tumor visualization.
• Manual MRI interpretation is prone to errors, highlighting the need for automated solutions.

Purpose of the Study:

• To develop a novel, accurate, and robust brain tumor classification method.
• To address the limitations of Convolutional Neural Networks (CNNs) in probability estimation.
• To improve high-confidence classification decisions for various brain tumor types.

Main Methods:

• Integration of Bag-of-Features model with SIFT feature extraction.
• Application of weighted Support Vector Machines (SVM) for classification.
• Development of scalable and parallelable algorithms for large datasets.

Main Results:

• SIBOW-SVM effectively extracts hidden image features for tumor differentiation.
• The method achieves accurate label predictions and reliable probability estimations.
• Outperforms state-of-the-art techniques, including CNNs, in accuracy and efficiency.

Conclusions:

• SIBOW-SVM offers a significant advancement in automated brain tumor classification from MRI.
• The method provides superior uncertainty quantification and data robustness.
• Enables practical implementation for massive medical image datasets, aiding clinical decisions.