At our inaugural fMRI User Meeting, Professor Karsten Specht from the University of Bergen addressed a central challenge in neuroimaging: the repeatability of functional MRI (fMRI).Despite more than 30 years of research, variability remains a major issue, particularly when results are needed at the individual level for clinical decision-making.

Reproducibility is the cornerstone of science. Without it, we cannot trust what we see, he said.

Even small differences in scanner hardware, preprocessing pipelines, or task designs can give different results. Human factors, such as participant compliance, attention, and instructions, add another layer of variability. Physiological influences –including blood pressure, heart-rate variability, caffeine intake, hormones, and circadian rhythms – also shape the BOLD response, making individual-level reliability a challenge.

Variability at the individual level

Prof. Specht illustrated this with a study in which participants were scanned twice while performing identical visual tasks. When instructed to focus on the stimuli, individual activation maps were highly consistent. When not explicitly instructed to attend to the stimuli, results varied dramatically between individuals, even though group-level analyses appeared stable. This demonstrates the critical role of precise, consistent instructions for reliable fMRI, particularly in clinical contexts.  

The need for standardization

To improve repeatability, Prof. Specht stressed the importance of rigorous standardization. Acquisition protocols should be identical across scanners, preprocessing pipelines harmonized, and every detail of task design and stimulus presentation documented. Quality control should be continuous throughout the study.

If we want to compare across centers or timepoints, we must treat fMRI like a clinical diagnostic tool – not just a research instrument, Specht emphasized.

Lessons from language mapping

Language fMRI, commonly used for presurgical planning, exemplifies both the potential and the limitations of fMRI. While motor mapping is generally robust, language networks are more sensitive to task design, attention, and control conditions. Resting-state fMRI, although widely used in research, appears less sensitive than task-based fMRI for clinical use due to its unstructured network activity.

Toward reliable clinical fMRI

Looking ahead, Prof. Specht highlighted the promise of integrating multiple imaging modalities, combining structural, perfusion, and fMRI data. Emerging tools such as BIDS, fMRI Prep, and AI can support consistent analysis, reduce variability, and make results more interpretable. Such approaches bring the field closer to reliable, individualized clinical applications of fMRI.

‍

Watch the presentation here: