Partial volume-aware assessment of multiple sclerosis lesions

Area of Science:

• Neuroimaging
• Medical image analysis
• Radiology

Background:

• Accurate white-matter lesion quantification is crucial for multiple sclerosis (MS) diagnosis and monitoring.
• Existing automated segmentation methods struggle with early-stage MS due to low lesion load and small lesion sizes.
• Partial volume effects, where voxels contain a mix of healthy and lesional tissue, complicate accurate segmentation.

Purpose of the Study:

• To develop and evaluate an algorithm for automated MS lesion load assessment (count and volume).
• To specifically address challenges posed by partial volume effects in early-stage MS.
• To improve the accuracy of automated lesion segmentation compared to existing methods.

Main Methods:

• The proposed algorithm utilizes 3D MPRAGE and 3D FLAIR images from routine MS clinical protocols.
• It incorporates a novel approach to handle partial volume effects in voxel analysis.
• The method was evaluated against manual segmentation on 39 early-stage MS patients.

Main Results:

• The algorithm demonstrated superior performance in segmenting small lesions, which are often affected by partial volume effects.
• Achieved higher median Dice Similarity Coefficients (DSC = 0.55) compared to two widely used methods (DSC = 0.50).
• Showcased a higher median detection rate (DR = 61%) than existing automated methods (DR < 45%).

Conclusions:

• The developed algorithm effectively assesses MS lesion load, particularly in early disease stages.
• Accounting for partial volume effects significantly enhances the segmentation of small MS lesions.
• This method offers improved accuracy for automated MS lesion segmentation in clinical practice.