White Matter MS-Lesion Segmentation Using a Geometric Brain Model.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Anatomy

Background:

• Multiple Sclerosis (MS) diagnosis relies on identifying brain lesions via MRI.
• Accurate lesion segmentation is crucial for MS diagnosis and monitoring.
• Current methods often depend on MRI atlases, limiting generalizability.

Purpose of the Study:

• To develop a novel 3D Multiple Sclerosis (MS) lesion segmentation method.
• To create an atlas-independent segmentation approach by modeling topological properties.
• To introduce a new robust metric for evaluating segmentation quality.

Main Methods:

• An adaptive geometric brain model was employed for 3D MS lesion segmentation.
• Topological properties of lesions and brain tissues were modeled to constrain segmentation to white matter.
• The method was evaluated on the MICCAI MS 2008 challenge dataset and an in-house patient cohort.

Main Results:

• The proposed method achieved competitive results on the MICCAI MS 2008 challenge.
• On an in-house dataset, the method outperformed atlas-based approaches in most distance metrics.
• A new, robust segmentation quality evaluation metric was formulated and validated.

Conclusions:

• The developed 3D MS lesion segmentation method is effective and atlas-independent.
• The approach demonstrates strong performance and potential for improving MS diagnosis.
• The novel evaluation metric offers a more reliable assessment of segmentation accuracy.