Evaluation of the intra- and inter-method agreement of brain MRI segmentation software packages: A comparison between...

Area of Science:

• Neuroimaging
• Brain MRI Segmentation
• Computational Neuroscience

Background:

• Inconsistent results in neuroimaging studies can arise from a lack of agreement between different analysis methods.
• Automatic segmentation of brain structures is crucial for quantitative MRI analysis.

Purpose of the Study:

• To evaluate the intra-method repeatability and inter-method reproducibility of FreeSurfer (FS) and Statistical Parametric Mapping (SPM) for brain MRI segmentation.
• To assess the impact of these segmentation methods on group-level statistical comparisons, specifically sex differences in brain volume.

Main Methods:

• Segmentation of gray matter, white matter, and subcortical structures was performed using FS and SPM on test-retest MRI data from the Kirby-21 and OASIS datasets.
• Intra-method repeatability and inter-method reproducibility were assessed using Pearson's correlation (r), Bland-Altman plots, and the Dice index.
• Statistical comparisons of male versus female brain volumes were conducted to evaluate method-dependent effects.

Main Results:

• Both FS and SPM demonstrated high intra-method repeatability (r = 0.95-0.99).
• Inter-method reproducibility showed moderate to high correlation (r = 0.72-0.98) and overlap (Dice index = 0.76-0.83) between FS and SPM.
• SPM systematically yielded larger gray matter volumes and smaller white matter/subcortical volumes compared to FS, leading to inconsistencies in sex difference findings between datasets.

Conclusions:

• Inter-method reproducibility is a critical factor that must be assessed before interpreting neuroimaging study results.
• Differences in segmentation algorithms can influence quantitative volumetric measures and subsequent statistical outcomes.
• Careful consideration of software choice and validation is necessary for robust and reliable neuroimaging research.