Quantifying brain tissue volume in multiple sclerosis with automated lesion segmentation and filling

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Neurology

Background:

• Multiple Sclerosis (MS) lesions affect brain tissue segmentation accuracy.
• Lesion filling techniques can mitigate these effects.
• Fully automated pipelines for lesion segmentation and filling in tissue volume analysis are underexplored.

Purpose of the Study:

• To evaluate the error introduced by automated lesion segmentation and filling in brain tissue segmentation.
• To compare different automated and manual pipeline combinations for lesion processing.
• To assess the accuracy of automated pipelines for brain tissue volume analysis in MS patients.

Main Methods:

• Utilized LST and SLS toolkits for lesion segmentation and filling on 70 clinically isolated syndrome patient images.
• Implemented pipelines with automated/manual lesion segmentation and lesion filling/masking.
• Segmented brain tissues (gray matter and white matter) using SPM8 and compared results to expert-filled annotations.

Main Results:

• Fully automated pipelines significantly reduced % error in gray matter (GM) and white matter (WM) volumes for MS patients.
• Automated lesion segmentation and filling performed similarly to expert lesion masking.
• Misclassified lesion voxels were the primary error source, but filling reduced error more than masking.

Conclusions:

• Automated lesion segmentation and filling pipelines enable accurate brain tissue volume measurements without manual intervention.
• LST and SLS toolkits support reliable analysis, reducing time, cost, and inter/intra-observer variability.
• This automation is crucial for efficient and consistent MS neuroimaging analysis.