Effect-Size Estimation Using Semiparametric Hierarchical Mixture Models in Disease-Association Studies with...

Area of Science:

• Neuroimaging analysis
• Statistical modeling
• Biomedical research

Background:

• Evaluating the clinical significance of brain regions in disease association studies requires accurate estimation of effect sizes.
• Existing methods for neuroimaging data analysis may suffer from selection bias in effect size estimates.

Purpose of the Study:

• To propose a novel model-based framework for voxel-based inferences in neuroimaging data that accounts for spatial dependency.
• To improve the estimation of effect sizes for individual brain areas in disease association studies.
• To reduce selection bias in effect size estimates.

Main Methods:

• Developed a hierarchical mixture model incorporating a hidden Markov random field structure to handle spatial dependencies between voxels.
• Utilized a nonparametric approach for the effect size distribution to allow flexible estimation.
• Validated the proposed method through simulation experiments.

Main Results:

• The proposed framework significantly reduces selection bias in effect size estimates compared to naive methods.
• Demonstrated improved accuracy in identifying and quantifying associations between brain areas and disease.
• Successfully applied the method to neuroimaging data from an Alzheimer's disease study.

Conclusions:

• The proposed model-based framework offers a robust approach for voxel-based inferences in neuroimaging, enhancing the reliability of disease association studies.
• Accurate effect size estimation is crucial for understanding the biological and clinical significance of neuroimaging findings.
• This method has potential applications in various neurological disease research, including Alzheimer's disease.