An Efficient Brain Tumor Segmentation Method Based on Adaptive Moving Self-Organizing Map and Fuzzy K-Mean Clustering

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Brain tumor segmentation in Magnetic Resonance Imaging (MRI) is a complex challenge.
• Machine learning has shown promise in improving tumor detection and segmentation accuracy.
• Existing methods often struggle with precise tumor boundary delineation.

Purpose of the Study:

• To develop an efficient technique for brain tumor detection and segmentation using MRI data.
• To improve the accuracy and efficiency of tumor region extraction.
• To evaluate the performance of the proposed method against established techniques.

Main Methods:

• Utilized the Kaggle Brats-18 brain tumor dataset (1691 images).
• Employed Adaptive Moving Self-Organizing Map (AMSOM) for unsupervised feature learning and classification.
• Applied Fuzzy k-means (FKM) clustering for precise tumor region segmentation.
• Incorporated Wiener filtering for noise removal and Gray Level Co-occurrence Matrix (GLCM) for feature extraction.

Main Results:

• The proposed AMSOM-FKM technique demonstrated superior performance in brain tumor segmentation.
• Achieved over 10% improvement in accuracy, sensitivity, precision, and similarity index compared to Fuzzy-C-means and K-means methods.
• Successfully segmented tumor regions by distinguishing them from surrounding tissues.

Conclusions:

• The AMSOM-FKM method offers an efficient and accurate approach for brain tumor segmentation in MRI.
• This technique holds potential for clinical applications in neuro-oncology.
• Further research can explore its application on diverse brain tumor datasets.