Round Randomized Learning Vector Quantization for Brain Tumor Imaging.

Area of Science:

• Medical Imaging
• Machine Learning
• Artificial Intelligence

Background:

• Accurate classification of brain magnetic resonance imaging (MRI) as normal or abnormal is crucial but challenging.
• Learning Vector Quantization (LVQ) is a potential technique for medical image classification.
• Existing LVQ methods face limitations in winner selection and class distribution handling.

Purpose of the Study:

• To enhance the performance and accuracy of the LVQ technique for brain tumor detection in MRIs.
• To introduce improvements in the winner code vector selection process within the LVQ algorithm.
• To address class imbalance issues in competitive learning classifiers through a novel multiresampling technique.

Main Methods:

• Modified LVQ by incorporating a round-off function with Euclidean distance for improved winner selection.
• Proposed a multiresampling technique using random selection via preclassification to achieve better class distribution.
• Utilized brain tumor MRI datasets from Universiti Kebangsaan Malaysia Medical Center and UCI benchmark datasets for testing.

Main Results:

• The enhanced LVQ methods demonstrated promising results in brain tumor classification.
• Comparative studies indicated improved detection accuracy compared to standard LVQ.
• The proposed techniques showed competitive performance against other classifiers like Multilayer Perceptron and Radial Basis Function.

Conclusions:

• The developed enhancements to LVQ significantly improve brain tumor detection accuracy in MRIs.
• The combination of modified winner selection and multiresampling effectively addresses key limitations of traditional LVQ.
• The proposed approach offers a more robust and accurate solution for medical image classification tasks.