A multi-time-point modality-agnostic patch-based method for lesion filling in multiple sclerosis.

Area of Science:

• Medical Imaging
• Neuroscience
• Computer Graphics

Background:

• Multiple sclerosis (MS) lesions complicate brain image analysis, causing segmentation issues and biased morphometric estimates.
• Current lesion filling methods, treating lesions as normal-appearing white matter, introduce errors due to segmentation inaccuracies and adjacent structures.
• Artefacts arise from current techniques, impacting the reliability of quantitative MRI studies in MS.

Purpose of the Study:

• To develop a novel, advanced lesion filling strategy for multiple sclerosis brain imaging.
• To minimize bias, artefacts, and spurious edges in image analysis caused by MS lesions.
• To preserve anatomical integrity and signal-to-noise ratios during lesion imputation.

Main Methods:

• A five-dimensional (5D), patch-based Non-Local Means algorithm inspired by in-painting techniques was employed.
• The method utilizes multi-modality and multi-time-point data for lesion texture plausibility.
• The algorithm was designed to be modality-agnostic and applicable across multiple time points.

Main Results:

• The proposed lesion filling strategy demonstrated reduced bias and fewer artefacts compared to existing methods.
• The technique successfully preserved anatomical structures and signal-to-noise characteristics.
• It effectively handled lesions adjacent to grey matter or cerebrospinal fluid without significant blurring or rasterization.

Conclusions:

• The novel 5D Non-Local Means algorithm offers a superior approach to lesion filling in multiple sclerosis imaging.
• This method enhances the accuracy of tissue segmentation and morphometric analysis in MS research.
• The technique provides a robust solution for lesion imputation, preserving image quality and anatomical details.