A Multidimensional Particle Swarm Optimization-Based Algorithm for Brain MRI Tumor Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Image segmentation is crucial for medical image analysis, particularly for identifying pathologies like brain tumors.
• Traditional segmentation methods often require manual parameter tuning and predefined segment counts, limiting their adaptability.
• Existing approaches may struggle with complex data like multi-modal MRI, necessitating more robust techniques.

Purpose of the Study:

• To develop an automated, clustering-based brain tumor segmentation algorithm using a multidimensional Particle Swarm Optimization (PSO).
• To enhance segmentation precision by integrating PSO with a Random Forest Classifier (RFC).
• To reduce the dependency on large, labeled datasets for training accurate tumor segmentation models.

Main Methods:

• Implementation of a multidimensional PSO variant for unsupervised clustering-based image segmentation.
• Incorporation of grayscale intensity and spatial information from multi-modal MRI data.
• Integration of initial segmentations with a Random Forest Classifier (RFC) for refined results.
• Validation using the RSNA-ASNR-MICCAI brain tumor segmentation (BraTS) challenge dataset.

Main Results:

• The proposed algorithm automatically determines optimal segmentation granularity without predefined segment numbers.
• The method effectively isolates brain tumors by leveraging multi-modal MRI data and spatial information.
• Integration with RFC significantly enhanced segmentation precision.
• Achieved robust results on the BraTS dataset with reduced need for extensive labeled training data.

Conclusions:

• The developed PSO-based approach offers an efficient and accurate method for brain tumor segmentation.
• Automatic selection of segmentation granularity and multi-modal data integration improve clinical relevance.
• This method presents a promising alternative for scenarios with limited labeled training data in medical image analysis.