Multiple comparison correction methods for whole-body magnetic resonance imaging.

Area of Science:

• Medical Imaging
• Statistical Analysis
• Bioinformatics

Background:

• Voxel-level hypothesis testing in medical imaging faces challenges due to test multiplicity.
• Existing correction methods are primarily evaluated on neuroimaging and synthetic data.
• New approaches like Imiomics require robust multiplicity correction for reliable inference.

Purpose of the Study:

• To evaluate existing familywise error rate (FWER) correction methods for whole-body MRI and correlation analyses within the Imiomics framework.
• To develop novel multiplicity correction techniques tailored for Imiomics data and analysis types.
• To enhance the reliability of statistical inference in complex imaging datasets.

Main Methods:

• Evaluation of common FWER controlling procedures on whole-body correlation analyses.
• Standard nominal error rate estimation using synthetic datasets.
• Prior-knowledge based stringency analysis incorporating anatomical priors.
• Comparison with anatomy-based method extensions.

Main Results:

• Nonparametric statistical methods demonstrated superior performance for whole-body MRI correlation analyses.
• Anatomical prior-informed extensions achieved comparable statistical power while maintaining FWER closer to the target rate.
• The proposed prior-knowledge based evaluation framework validated the effectiveness of the new methods.

Conclusions:

• Permutation-based statistical approaches are suitable for Imiomics but can be enhanced by incorporating image structure information.
• Anatomical priors significantly improve the performance of multiplicity correction methods in Imiomics.
• These enhanced methods are crucial for advancing inference in large-scale imaging studies and emerging applications.