Sleep spindles—brief bursts of rapid brain activity that punctuate non-REM sleep—are emerging as one of the most revealing windows into how the sleeping brain consolidates memories and changes with age. Now, a meta-analysis of electroencephalography (EEG) recordings from domestic dogs suggests that automatically detected canine sleep spindles share several defining features with human spindles. The result strengthens the case for dogs as a model species in sleep and cognition research while offering a new way to test whether algorithms are identifying genuine biological events rather than merely finding rhythmic noise.
Published in Neuroinformatics, the study examined whether an automated detector developed for dog EEG produces events with three “face-validity” characteristics historically associated with real sleep spindles in humans. These indicators are not a definitive biological test, but they provide a practical quality check. Genuine spindles should generally be short, faster spindles should be more common over central and posterior areas of the head, and larger spindles should tend to show a stronger negative chirp—a gradual slowing of their dominant frequency as the event unfolds. All three patterns appeared in most of the datasets and measurements analyzed by researchers Ivaylo Borislavov Iotchev and Anna Kis.
Sleep spindles are transient changes in the electrical rhythm of the brain. During non-REM sleep, the background EEG is dominated by relatively slow activity, generally below about 4 hertz. A spindle occurs when this slow rhythm is temporarily replaced by a faster oscillation, typically within the sigma-frequency range. In the dog detector used in the study, the relevant band was 9 to 16 hertz, although the researchers paid particular attention to “fast” spindles above 13 hertz. Individual spindles can last from roughly half a second to several seconds, but large human datasets suggest that most are much shorter than the commonly cited six-second maximum. Their brevity and changing frequency make them difficult to identify consistently, especially when recordings are collected from different species, brain regions or experimental conditions.
For decades, visual inspection by trained EEG scorers was treated as the gold standard for identifying sleep spindles. Human experts can often agree with one another more closely than several automated detectors agree with each other. Yet visual scoring has important limitations. A spindle may be faint, partially hidden by other rhythms or localized to a region where its electrical signal is diluted by volume conduction—the spread of electrical activity through surrounding tissue before it reaches a scalp electrode. In animals, the problem can be even more pronounced. Some non-human EEG patterns resemble spindles in duration and frequency, while epileptic discharges or so-called pseudo-spindles can create ambiguous waveforms. A detector might therefore identify events that human observers miss, without those events necessarily being false positives.
The algorithm tested in this work was first introduced in 2017 and has been used consistently in the group’s subsequent studies of canine sleep. It analyzes overlapping EEG windows lasting 0.5 seconds. A window is initially marked as containing spindle activity when its greatest frequency power falls in the sigma range and its root-mean-square amplitude exceeds one standard deviation above the average for that recording. A second filtering step removes candidate events with amplitude or frequency values more than two standard deviations above or below the mean of the initially selected activity. Both stages are applied only to periods classified through visual inspection as non-REM sleep. This approach does not attempt to reproduce the exact shape of a spindle by eye; instead, it identifies brief, relatively powerful bursts within a biologically plausible frequency range.
The researchers analyzed afternoon nap recordings lasting two to three hours from five previously collected datasets. These datasets represented studies of command learning, age and sex, and targeted memory reactivation in dogs. In total, the analysis included 20 sampling events, each defined by a combination of dog population, experimental condition and EEG electrode. The recordings primarily used electrodes positioned along the midline of the skull, including Fz near the front of the head and Cz near the center, with the occipital bone serving as a reference. A dog appearing in three datasets was accounted for when assessing overlap, and datasets with substantial subject duplication were excluded in favor of the larger sample.
Because the detector identifies overlapping windows rather than the precise beginning and end of every spindle, the researchers estimated average duration rather than measuring each event individually. They counted the number of windows assigned to spindle activity and the number of separate spindles, treating windows separated by more than half a second as distinct events. The resulting calculation incorporated both the 0.5-second window length and the 0.375-second spacing between the centers of neighboring overlapping windows. This analysis indicated that automatically detected dog spindles were characteristically brief, with most events falling within the short time range expected for genuine spindles and well below the six-second upper limit sometimes used in descriptions of the phenomenon.
The second test examined spindle topography, or how events are distributed across recording locations. In humans, rats and dogs, fast spindles are generally more frequent over central and posterior parts of the head than over anterior regions. The current analysis reproduced this pattern across the eligible canine datasets, extending an earlier observation that had been reported in dogs but had not yet received a broad replication. Since all of the recordings were non-invasive scalp measurements, the result does not address how spindle frequency might vary with recording depth inside the brain. It does, however, suggest that the detector is sensitive to a spatial organization expected from thalamocortical sleep rhythms rather than simply responding to any high-frequency activity equally across the scalp.
The third criterion involved chirp, a change in the dominant frequency during a single spindle. A negative chirp means that the oscillation slows from the beginning of the event toward its end. Previous human work has associated stronger negative chirps with larger spindle amplitudes, providing a relationship that may help distinguish spindles from other short-lived oscillations. To estimate chirp, the researchers compared the dominant frequency on the left side of each spindle-containing window with the dominant frequency on the right, using a narrow region around the window’s center. The method was intentionally simple and was not based on more elaborate wavelet analysis, but the predicted positive association between amplitude and negative chirp appeared in the canine data.
Replication was more dependable when results were compared across complete datasets than when they were examined separately across every sampling event. The latter included different electrodes and experimental conditions, sometimes involving the same animals, and consequently introduced more opportunities for variation. Nevertheless, when evidence across tests was combined using Fisher’s method, the overall probabilities consistently favored compliance with all three face-validity criteria. That pattern does not prove that every detected event is a true spindle, nor does it establish that canine and human spindles are identical. Instead, it shows that the population of events identified by the algorithm behaves in ways expected of sleep spindles across multiple independent collections of recordings.
The findings are especially relevant because the same automated detections have previously been linked in dogs to age, sex, reproductive status and cognitive performance—relationships that broadly parallel observations in human sleep research. Those earlier associations provided evidence of predictive or external validity: the measurements behaved as expected when compared with traits outside the EEG itself. The new analysis adds a different layer of support by showing that the detected events also possess recognizable physiological characteristics. Together, the results suggest that automatic analysis can complement, rather than simply replace, visual scoring, particularly in animal studies where human inspection may be less reliable.
The work also highlights the limits of current sleep-spindle research. The duration measure is an estimate created after the original detection algorithm was designed, and the chirp calculation is deliberately crude. The study does not use visual confirmation of every event, nor does it directly record activity from the thalamus, a deep brain structure central to the generation and coordination of sleep spindles. Further studies will need to test the detector against alternative algorithms, larger and more diverse canine populations, higher-density EEG and recordings collected under different sleep conditions. Even with those caveats, the convergence of short duration, characteristic scalp distribution and amplitude-linked slowing gives researchers a stronger basis for interpreting dog sleep spindles as meaningful neural events—and for using sleeping dogs to investigate how the brain preserves learning and cognition during the night.
Subject of Research: Automatically detected sleep spindles in domestic dogs and their similarity to human sleep-spindle characteristics.
Article Title: Meta-Analysis of Face-Validity Indicators in Automatically Detected Dog Sleep Spindles
Article References: Iotchev, I. B., & Kis, A. “Meta-Analysis of Face-Validity Indicators in Automatically Detected Dog Sleep Spindles.” Neuroinformatics, volume 24, article 53 (2026). Published 15 August 2026.
Image Credits: AI Generated
DOI: 10.1007/s12021-026-09810-4
Keywords: sleep spindles, dogs, canine EEG, sleep research, non-REM sleep, automated detection, neuroinformatics, memory consolidation, cognitive aging, chirp, EEG topography
Tags: automated EEG detection in dogsbiological validation of sleep spindle detectioncanine models in sleep and cognition researchcomparison of dog and human sleep spindlesdog sleep spindlesEEG analysis of sleep spindlesEEG signal analysis in sleep studiesface-validity measures for sleep spindlesnon-REM sleep brain activity in dogsrole of sleep spindles in memory consolidationsleep research using domestic dogssleep spindle characteristics and features
