Objective Bayesian fMRI analysis-a pilot study in different clinical environments.

Area of Science:

• Neuroimaging
• Neurosurgery
• Statistical Analysis

Background:

• Functional MRI (fMRI) is crucial for neurosurgical planning, identifying eloquent brain areas via BOLD signal analysis.
• Frequentist statistics, while controlling false positives, risk harmful false negatives in surgical contexts.
• Bayesian statistics offer an alternative but lack objective effect size thresholds for clinical use.

Purpose of the Study:

• To implement a Bayesian analysis framework for neurosurgical planning fMRI.
• To develop and validate an automated effect-size threshold selection method for Bayesian fMRI analysis.
• To compare Bayesian and frequentist approaches for fMRI data acquired at different scanner strengths (3T and 1.5T).

Main Methods:

• A novel Bayesian framework was developed, incorporating an automated effect-size threshold for posterior probability maps.
• The method was calibrated using frequentist results and expert-defined thresholds.
• Bayesian and frequentist analyses were compared using fMRI data from healthy volunteers and a brain tumor patient.

Main Results:

• The Bayesian approach identified all four activation categories (activated, deactivated, non-activated, low confidence) in 3T data.
• Activation extent and foci in Bayesian analysis aligned with frequentist results at 3T.
• At 1.5T, Bayesian analysis showed superior sensitivity for brain activation detection compared to frequentist methods, despite smaller spatial extents.

Conclusions:

• Automated effect-size thresholding in Bayesian fMRI analysis enhances sensitivity and certainty for neurosurgical guidance.
• This approach may mitigate risks associated with false negatives in identifying critical brain regions.
• Bayesian fMRI analysis holds promise for improving intra-operative and pre-operative neurosurgical planning.