Quantitative evaluation of statistical inference in resting state functional MRI

Area of Science:

• Neuroimaging
• Statistical Inference
• Computational Neuroscience

Background:

• Resting-state functional MRI (rs-fMRI) connectivity analysis is crucial for understanding brain function.
• Current statistical methods face limitations in sensitivity and specificity due to challenges in characterizing noise and artifact distributions in vivo.
• Empirical validation of advanced statistical techniques in real-world rs-fMRI data remains difficult.

Purpose of the Study:

• To introduce a novel method for empirically validating statistical inference techniques for rs-fMRI connectivity.
• To assess the performance of different inference methods based on their stability against synthetic data reduction.
• To compare the resilience of ordinary versus robust inference methods in the context of rs-fMRI.

Main Methods:

• Leveraging the theoretical framework of SIMulation and EXtrapolation (SIMEX) to estimate bias in statistical estimators.
• Defining and quantifying 'resilience' as the stability of inference methods when empirical data is synthetically reduced.
• Applying the resilience metric to compare ordinary and robust statistical inference methods using rs-fMRI data.

Main Results:

• The SIMEX-based resilience metric provides a quantitative measure of empirical performance for statistical inference methods.
• Demonstrated differences in resilience between ordinary and robust inference methods in rs-fMRI analysis.
• The proposed approach offers a viable strategy for in vivo validation of statistical methods in neuroimaging.

Conclusions:

• Resilience, assessed via a SIMEX-inspired approach, offers a robust metric for evaluating statistical inference in rs-fMRI.
• Robust inference methods exhibit greater resilience compared to ordinary methods when data is diminished.
• This framework facilitates more reliable and sensitive connectivity analyses in rs-fMRI studies.