TAPAS: A Thresholding Approach for Probability Map Automatic Segmentation in Multiple Sclerosis.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Multiple Sclerosis Research

Background:

• Total brain white matter lesion (WML) volume is a key MRI outcome measure in multiple sclerosis (MS) studies.
• Manual delineation is the gold standard for WML volume estimation, but automatic methods are widely used.
• Existing automatic methods often use manually selected thresholds, introducing human error and bias.

Purpose of the Study:

• To propose and validate an automated thresholding algorithm, TAPAS, for subject-specific estimation of WML volumes.
• To address the limitations of fixed or manually selected thresholds in automatic WML segmentation.
• To improve the accuracy and reliability of WML volume quantification in MS.

Main Methods:

• Developed the Thresholding Approach for Probability Map Automatic Segmentation in Multiple Sclerosis (TAPAS) algorithm.
• Utilized multimodal MRI to generate probability maps for T2-hyperintense WMLs.
• Optimized subject-specific thresholds by maximizing the Sørensen-Dice similarity coefficient (DSC) and modeled using a generalized additive model.

Main Results:

• TAPAS demonstrated an average reduction in subject-level absolute error of 0.1 mL per 1 mL increase in manual volume in the JHH dataset.
• Bland-Altman analysis showed that TAPAS mitigated volumetric bias associated with group-level thresholding.
• Similar absolute error estimates were observed in the BWH dataset, with no systematic biases identified for TAPAS or group-level thresholding.

Conclusions:

• TAPAS is the first validated, fully automated method for subject-specific threshold prediction in brain lesion segmentation.
• The algorithm enhances the accuracy and reduces bias in automated WML volume quantification for MS research.
• This automated approach offers a more reliable and reproducible method for assessing WML burden in multiple sclerosis.