Mouse EEG Atlas Pipeline Reaches 0.93 mm Localization for Fmr1 Study

By Olivia Hart

A 30-channel mouse EEG pipeline localized superficial electrical sources to a mean error of 0.93 millimeters and deeper sources to 5.08 mm, according to an open-access NeuroImage paper published 15 October 2026 (article 122218) by Edmondson, Pedapati and seventeen co-authors across Cincinnati Children's Hospital, the University of Cincinnati, the University of California Riverside, and collaborating sites.

The team built an atlas-based electrical source imaging (ESI) workflow, a method that maps scalp EEG signals back to their underlying brain generators and is standard in human neurophysiology, and adapted it to mice. Mouse brains are roughly 3000 times smaller than human brains, a scale that makes sparse-array source separation difficult. The investigators benchmarked eighteen combinations of source space, head model and inverse operator against simulated dipoles (mathematical stand-ins for point-source neural activity) under conductivity and noise sweeps.

Key benchmark findings

• Best configuration: ROI (region-of-interest) source space, ellipsoid head model, sLORETA (standardized low-resolution electromagnetic tomography, a distributed inverse method yielding a blurred but unbiased localization map).
• Superficial localization error: 0.93 mm.
• Deep localization error: 5.08 mm.
• Parcel-attribution recall: 70 percent superficial; 2 percent beyond 5 mm depth.

What does this change for mouse EEG laboratories?

For preclinical groups that record scalp EEG in mice, the result means a 30-electrode array can resolve millimetric regions rather than merely detect global spectral changes. Researchers can move beyond electrode-level statistics toward parcel-level claims about which cortical or subcortical structure generated an oscillation. The pipeline uses Allen CCFv3, the Allen Institute's three-dimensional mouse brain atlas, for anatomical labeling. The 70-to-2 percent recall drop across depth sets the bound any mouse ESI study must respect when assigning activity to deep structures.

Application to an Fmr1 knockout cohort

The authors applied the pipeline to resting-state EEG from 18 male Fmr1 knockouts and 17 wild-type littermates. Fmr1 encodes fragile X messenger ribonucleoprotein 1; its deletion is the standard model for Fragile X syndrome, a neurodevelopmental disorder linked to intellectual disability and autism spectrum features.

Gamma power was elevated in knockouts across all eighteen pipeline configurations. Low gamma reached a standardized mean difference (SMD, a unit-free effect size) of 1.02 at q = 0.004, where q denotes a false-discovery-rate-adjusted p-value. High gamma reached an SMD of 0.67 at q = 0.004. The elevated activity localized to six parcels, three of which appeared in both gamma bands. Regional differences extended across delta, theta and high gamma relative power, and across both aperiodic parameters (the slope and offset of the 1/f background spectrum, indices of non-oscillatory neural activity).

A parallel electrode-level analysis recovered neither the delta pattern nor any named anatomical region, underscoring the gain from source localization. The gamma findings persisted after adjustment for acquisition cohort, reaching p = 0.001 for low gamma and p = 0.002 for high gamma.

What the pipeline cannot yet say

The authors flag that validation rests on simulated dipoles, which bound what the workflow can recover but cannot confirm a labeled parcel as the true generator in a living animal. Replication across recording systems, sex, age, and animals naïve to experimental procedures remains pending.

The team proposes formal reporting criteria for validating mouse source-imaging pipelines, an early step toward routine adoption in preclinical neuroscience. No FDA-cleared device is involved; the work is investigational and animal-only, and the next studies will test whether the same localization bounds hold across laboratories and in female and aged cohorts.