Comparison study of clinical 3D MRI brain segmentation evaluation

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Accurate segmentation of brain tissues in MRI is essential for clinical diagnosis and research.
• Existing segmentation methods lack comprehensive evaluation on real-world clinical datasets, particularly for cerebrospinal fluid (CSF).

Purpose of the Study:

• To compare the performance of four distinct brain MRI segmentation methods: gray levels thresholding, 3D level set, fuzzy connectedness, and FSL.
• To quantitatively evaluate segmentation accuracy against manual segmentations on a clinical dataset.

Main Methods:

• Utilized a database of 10 adult subjects' cerebral brain MRI scans.
• Applied and compared gray levels thresholding, 3D level set, fuzzy connectedness, and FSL segmentation algorithms.
• Performed quantitative accuracy assessment by comparing automated segmentations to expert manual segmentations.

Main Results:

• Demonstrated significant performance variations among the evaluated segmentation techniques.
• Identified specific strengths and weaknesses of each method in segmenting white matter, gray matter, and CSF.
• Highlighted the challenges in achieving high accuracy for CSF segmentation across all tested methods.

Conclusions:

• The choice of segmentation method significantly impacts the accuracy of white matter, gray matter, and CSF extraction from brain MRI.
• Further research is needed to develop and validate robust segmentation techniques, especially for CSF, on diverse clinical datasets.