Pooling Morphometric Estimates: A Statistical Equivalence Approach.

Area of Science:

• Neuroimaging
• Biostatistics
• Medical Imaging Analysis

Background:

• Multicenter studies often face challenges with pooled MRI data due to hardware or image-processing variations.
• Classical inference testing is unsuitable for proving data similarity under changing conditions.

Purpose of the Study:

• To introduce and apply statistical equivalence testing for assessing the poolability of MRI data acquired under varying conditions.
• To demonstrate the utility of equivalence testing in multicenter neuroimaging research.

Main Methods:

• Employed the two one-sided tests (TOST) approach for statistical equivalence testing.
• Applied TOST to datasets including cortical thickness, automated and manual hippocampal volumetry, and corpus callosum area.
• Conducted power analyses to determine sample size requirements for robust equivalence testing.

Main Results:

• Cortical thickness measurements were equivalent across over 61% of the cortex when using different head coils.
• Automated hippocampal volume estimates were statistically equivalent between two identical coils.
• Manual hippocampal volumetry by two readers was not statistically equivalent, while automated corpus callosum area estimates with reader correction were equivalent.

Conclusions:

• Statistical equivalence testing, specifically TOST, is a valid method for determining if MRI morphometric measures from variable conditions can be pooled.
• The choice of equivalence margin significantly impacts sample size calculations for power analyses.
• This technique is crucial for maintaining data integrity and enabling robust analysis in large-scale, longitudinal, or multicenter neuroimaging studies.