Putative ripples largely index noise originating from region-, state-, and demand-dependent modulation of cortical background activity, with on average 77% of awake ripples in the medial temporal lobe reflecting false positives within the 1/f χ noise floor.
Key Findings
Results
On average, 77% of awake ripples detected in the medial temporal lobe, including the hippocampus, reflect false positives arising from 1/f χ noise.
The false positive rate was estimated using a simulation-based approach applied to intracranial EEG data
The finding held across three separate intracranial EEG studies
The false positives stem from region-, state-, and demand-dependent modulation of cortical background activity
The medial temporal lobe was specifically identified as a region where this problem is prominent during wakefulness
Results
Five commonly used ripple detection algorithms show substantial sensitivity to aperiodic 1/f χ noise across brain states and cortical regions.
Noise sensitivity was established for five common detection algorithms
The evaluation was performed across three intracranial EEG studies
Both sleep and cognitive engagement conditions were examined
The algorithms were tested across multiple cortical regions, not only the hippocampus
Results
Task-related modulations of 1/f χ activity lead to spurious ripple detections during cognitive engagement.
Demand-dependent changes in the aperiodic noise floor were observed during cognitive tasks
These 1/f χ modulations produce artifactual increases in detected ripple activity
The effect was observed in awake, task-engaged participants using intracranial EEG
This suggests that previously reported task-related ripple increases may partly reflect noise-driven false positives
Results
The study identifies specific scenarios where ripple detections are less impacted by 1/f χ noise.
Certain brain states and recording contexts showed reduced false positive rates
The authors used a simulation-based approach to characterize these scenarios
The findings suggest that conditions exist under which ripple detection can be more reliable
These scenarios are distinguished by lower or less variable aperiodic background activity
Methods
A simulation-based approach was developed to estimate the false positive rate in ripple detection attributable to aperiodic noise.
The method models the 1/f χ noise floor specific to each region, brain state, and cognitive demand
The approach was validated across three intracranial EEG datasets
The simulation framework provides a generalizable tool for assessing noise contamination in ripple studies
The authors describe this as offering 'a simulation-based approach to estimate the false positive rate'
Results
Ripple-like activity is modulated in a region-, state-, and demand-dependent manner reflecting changes in cortical background activity rather than true ripple oscillations.
Background aperiodic activity varied systematically across cortical regions and recording conditions
Both sleep and wakefulness conditions were included, revealing state-dependent differences
The neocortex as well as the hippocampus showed susceptibility to noise-driven detections
The findings challenge the interpretation of ripple-like activity observed in the human hippocampus and neocortex during wakefulness and sleep
What This Means
This research suggests that a large proportion of what scientists have been calling 'ripples' in the human brain are actually not true ripples at all, but instead are artifacts of background electrical noise. Sharp-wave ripples are brief bursts of high-frequency brain activity thought to be important for memory and cognition, and have been well-studied in rodents. Researchers have recently reported similar activity in the human brain during both sleep and waking tasks. However, this study, using recordings directly from the brains of human patients (intracranial EEG), found that roughly 77% of the ripple-like events detected in the hippocampus and surrounding regions during wakefulness are false alarms — they match the pattern expected from random neural background noise (called 1/f noise) rather than true ripple events.
The study tested five different commonly used computer algorithms for detecting ripples and found that all of them were susceptible to this noise problem, particularly when brain activity levels shifted due to changes in brain region, sleep versus wake state, or cognitive task demands. Importantly, when people were doing a cognitive task, the background noise level changed in a way that caused even more spurious ripple detections, meaning that previously published findings of task-related ripple increases could partly be explained by noise rather than genuine brain activity.
This research matters because it calls into question a large body of human neuroscience research on memory and cognition that has relied on ripple detection. The authors provide a simulation-based tool to help researchers estimate how many of their detected ripples might be false positives, and they identify specific conditions — certain brain states or recording contexts — where ripple detection is more reliable. This offers a path forward for more accurate study of these important brain signals in humans.