Statistical inference for same data meta-analysis in neuroimaging multiverse analyzes

Area of Science:

• Neuroimaging analysis
• Computational neuroscience
• Statistical modeling

Background:

• Task-functional magnetic resonance imaging (fMRI) research utilizes numerous analysis tools.
• Varied analytical approaches can inflate false positive rates and hinder reproducibility in neuroimaging.
• Multiverse analyses explore pipeline variations but create challenges in interpreting multiple outputs from a single dataset.

Purpose of the Study:

• To develop and validate methods for meta-analysis in the context of multiverse analyses, specifically addressing inter-pipeline dependence.
• To introduce "same data meta-analysis" (SDMA) as a solution for extracting consensus inferences from multiple analytical outputs derived from a single dataset.
• To provide reliable tools for interpreting findings from complex neuroimaging studies.

Main Methods:

• Developed a suite of same data meta-analysis (SDMA) methods to account for inter-pipeline dependence in multiverse outputs.
• Assessed the validity of SDMA methods through simulations.
• Tested SDMA models on real-world multiverse analysis outputs from the NARPS and HCP Young Adult studies.

Main Results:

• The proposed SDMA models demonstrate validity in the presence of inter-pipeline dependence.
• Simulations confirmed the effectiveness of the developed SDMA methods.
• Real-world application on NARPS and HCP Young Adult datasets showed the practical utility of SDMA.

Conclusions:

• Same data meta-analysis (SDMA) provides a robust framework for analyzing dependent results from multiverse neuroimaging studies.
• The validated SDMA methods offer researchers reliable options for drawing consensus inferences from complex analytical pipelines.
• This work enhances the interpretability and reproducibility of neuroimaging findings generated through multiverse analyses.