Variability in fMRI: an examination of intersession differences.

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Functional Magnetic Resonance Imaging (fMRI)

Background:

• Functional neuroanatomy is typically inferred from single functional magnetic resonance imaging (fMRI) sessions.
• This interpretation assumes negligible response variability between sessions.
• The empirical examination of this assumption is critical for reliable neuroimaging research.

Purpose of the Study:

• To empirically investigate the assumption of negligible session-specific responses in functional magnetic resonance imaging (fMRI).
• To determine the impact of session-by-condition interactions on the interpretation of functional neuroanatomy.
• To assess the validity of single-session fMRI data for subject-level conclusions.

Main Methods:

• Collected 99 fMRI sessions over 2 months from a single subject, repeating motor, visual, and cognitive paradigms.
• Utilized statistical models to analyze session-by-condition interactions and error variance.
• Employed multisession fixed-effects and random-effects analyses to evaluate activation and variance.

Main Results:

• Session-by-condition interactions significantly explained error variance across all paradigms (P < 0.05 corrected).
• Many voxels with significant session-by-condition interactions were not consistently activated across all sessions.
• Most variably activated voxels did not survive random-effects analysis, indicating sensitivity to between-session variance.

Conclusions:

• Accurate inference of subject responses requires statistical models accounting for within- and between-session variance.
• A sufficiently large sample of sessions is necessary for robust fMRI findings.
• Relying on single-session fMRI data risks erroneous conclusions about typical subject responses.