Segmentation of multiple sclerosis lesions in intensity corrected multispectral MRI

Area of Science:

• Medical Imaging
• Neuroscience
• Computer Vision

Background:

• Accurate segmentation of brain tissues in Magnetic Resonance (MR) images is crucial for diagnosing neurological conditions.
• Existing stochastic relaxation methods struggle with segmenting small lesions in multiple sclerosis (MS) patients, especially in thick MR slices.
• Lesion segmentation in MS requires precise identification of white matter abnormalities.

Purpose of the Study:

• To improve the accuracy of brain tissue and lesion segmentation in MR images of multiple sclerosis patients.
• To extend a stochastic relaxation method by incorporating pre- and post-processing steps for enhanced lesion detection.
• To develop a robust automated or semi-automated segmentation technique for MS lesions.

Main Methods:

• Implemented a 3-D stochastic relaxation method with partial volume analysis for brain voxel segmentation.
• Applied homomorphic filtering for image enhancement to correct MR image non-homogeneities.
• Utilized morphological processing and thresholding to create a white matter/lesion mask, followed by re-segmentation.

Main Results:

• The enhanced method successfully segmented white matter and multiple sclerosis lesions in multispectral MRI scans.
• The segmentation results demonstrated favorable comparison with manual segmentations performed by radiology professionals.
• The pre- and post-processing steps significantly improved the accuracy of lesion segmentation.

Conclusions:

• The extended stochastic relaxation method provides a more accurate approach for segmenting brain tissues and multiple sclerosis lesions in MRI.
• The integration of image enhancement and targeted masking improves the detection of small lesions.
• This technique offers a promising tool for the analysis of multiple sclerosis in neuroimaging.