Subject-Level Segmentation Precision Weights for Volumetric Studies Involving Label Fusion

Area of Science:

• Neuroimaging
• Brain Imaging Analysis
• Computational Neuroscience

Background:

• Volumetric data in neuroimaging are crucial for understanding brain changes in aging and disease.
• Current methods for extracting region of interest (ROI) volumes often use multi-atlas segmentation but neglect segmentation precision.
• This oversight limits the ability to accurately compare tissue volumes and identify disease-related associations.

Purpose of the Study:

• To develop a novel method for estimating segmentation variance in ROI volume measurements.
• To improve the statistical power of detecting group differences in brain volumes by incorporating segmentation precision.
• To enhance the analysis of neuroimaging data for identifying associations between tissue volume and neurological conditions.

Main Methods:

• Proposed a new method to estimate the variance of measured ROI volume for each subject.
• Utilized multi-atlas segmentation procedures to derive volume estimates.
• Applied weighting based on estimated segmentation variance to improve statistical power.

Main Results:

• Demonstrated the effectiveness of the novel method using real neuroimaging data.
• Showed that weighting by segmentation variance significantly improved the power to detect differences in hippocampal volume.
• Successfully identified a mean difference in hippocampal volume between control and mild cognitive impairment/Alzheimer's disease groups.

Conclusions:

• The proposed method enhances the reliability of volumetric measurements in neuroimaging.
• Accounting for segmentation precision is vital for accurate comparisons of brain tissue volumes.
• This approach offers a more sensitive tool for detecting subtle brain changes in conditions like Alzheimer's disease.