Enhancing Study Design and Analysis of MR Imaging Markers Through Measurement Error Modeling

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Statistical Modeling

Background:

• Measurement error in medical imaging studies can reduce statistical power and bias results.
• This compromises the reliability and accuracy of neuroimaging research.

Purpose of the Study:

• To introduce a dual data collection design to quantify measurement error in imaging.
• To apply regression calibration to correct for error-prone imaging markers.
• To improve biomarker-outcome estimation, statistical power, and sample size planning in neuroimaging studies.

Main Methods:

• Utilized a prospective reliability dataset and a retrospective main dataset.
• Employed regression calibration with MR fingerprinting (MRF) T1 and T1-weighted (T1w) MPRAGE sequences.
• Assessed reliability coefficients and applied corrections in an epilepsy cohort, with simulations for multi-site scenarios.

Main Results:

• MRF T1 markers showed higher reliability (λ=0.887-0.941) than T1w SI markers (λ=0.246-0.554).
• Regression calibration significantly increased effect sizes, particularly for T1w SI mean (333.22%).
• Combined regression calibration and Combat effectively reduced bias in multi-site simulations with larger site effects.

Conclusions:

• The dual data acquisition and regression calibration approach enhances the reliability and generalizability of multi-site neuroimaging studies.
• This method effectively restores attenuated imaging biomarker associations and improves statistical power.
• The findings provide a framework for optimizing sample size and study design in neuroimaging research.