Clustering of Brain Tumor Based on Analysis of MRI Images Using Robust Principal Component Analysis (ROBPCA)...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Accurate brain tumor detection is crucial for diagnosis and treatment planning.
• Existing methods may require manual intervention or struggle with complex image features.

Purpose of the Study:

• To develop and validate an automated method for brain tumor location detection using unsupervised clustering.
• To assess the efficacy of a robust Principal Component Analysis (PCA) algorithm for segmenting brain MRI images.

Main Methods:

• Unsupervised robust PCA algorithm applied to cluster brain MRI pixels into four leverages.
• Algorithm validated on MRI images from five brain diseases: glioma, Huntington's, meningioma, Pick's, and Alzheimer's.
• Utilized ten images per disease for performance evaluation.

Main Results:

• The algorithm successfully identified tumor locations with high sensitivity, accurately discerning lesions.
• Approximately 2% of data in suboptimal image regions were identified as indicative of tumors.
• The method demonstrated superior performance for glioma images compared to other diseases.

Conclusions:

• The unsupervised robust PCA method shows significant potential for automated brain tumor detection in various neurological conditions.
• The algorithm's ability to accurately locate lesions, as confirmed by ROC curve analysis, supports its clinical utility.
• Further refinement may enhance performance across all tested brain diseases.