Neuroscientists Have Been Misreading Brain Scans for Decades | Sabine Hossenfelder

The Gist

Functional magnetic resonance imaging signals fail to correlate with actual brain activity in 40 percent of voxels, while lesion network mapping studies show high non-specificity across diverse neurological disorders.

Quick Overview

Neuroscience faces a methodological crisis because fundamental assumptions underpinning brain imaging and lesion mapping are systematically flawed. Recent studies in Nature Neuroscience demonstrate that blood-oxygen-level-dependent signals frequently misrepresent actual oxygen metabolism and neuronal activity. Furthermore, cross-vendor MRI reliability studies reveal that different scanner manufacturers produce irreproducible functional brain network maps at the individual level.

Key Points: A study published in Nature Neuroscience by Samira Epp and colleagues reveals that 40 percent of fMRI blood-oxygen-level-dependent signals do not correlate with actual oxygen consumption and neuronal activity. The default mode network can show increased oxygen extraction from blood without any corresponding increase in local blood flow or neuronal firing. A famous 2009 study by Craig Bennett detected apparent brain activity in a dead Atlantic salmon using standard fMRI software, highlighting persistent statistical error issues. A second Nature Neuroscience paper by Martijn van den Heuvel and colleagues investigated lesion network mapping and found that supposed disease specific brain circuits look nearly identical across over 100 vastly different neurological and psychiatric conditions. A study in Scientific Reports by Lionel Butry and colleagues demonstrated that scanning human brains with different MRI manufacturers such as Siemens and Philips produces drastically different individual functional connectome results. Scientific publishers continue to incentivize publishing positive methodological papers because journals profit from high volume paper generation, while researchers face career pressures to ignore known data collection errors.

Context: Neuroscience heavily relies on neuroimaging techniques like functional magnetic resonance imaging to study brain function and map disease circuits. For decades, researchers have assumed that localized blood flow changes accurately reflect underlying neural activity, but an accumulation of recent papers exposes deep foundational flaws in data collection and statistical analysis.