A Brain Tumor Segmentation Framework Based on Outlier Detection Using One-Class Support Vector Machine.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate brain tumor segmentation is vital for cancer diagnosis and treatment planning.
• Magnetic resonance imaging (MRI) is the primary modality for brain tumor assessment.
• Automated segmentation streamlines workflows and enables radiomics analysis for outcome prediction.

Purpose of the Study:

• To develop an automated segmentation framework for brain tumors in MR images.
• To utilize an outlier-detection approach for delineating tumor and edema regions.
• To evaluate the framework's performance using post-contrast T1-weighted and T2-FLAIR MRI sequences.

Main Methods:

• Proposed an outlier-detection-based framework for automatic brain tumor segmentation.
• Employed independent one-class support vector machines on T1w and T2-FLAIR images to generate outlier masks.
• Refined and fused outlier masks using morphological and logical operators for final tumor segmentation.

Main Results:

• Achieved an average Dice similarity coefficient of 0.84 ± 0.06.
• Obtained an average Hausdorff distance of 1.85 ± 0.48 mm.
• Demonstrated robust performance on an independent test set.

Conclusions:

• The proposed outlier-detection framework enables accurate automatic brain tumor segmentation from MRI.
• This method has the potential to significantly improve image-guided radiation therapy and radiomics-based outcome prediction.
• The framework shows promising results for clinical application in brain tumor management.