Authors :
Michael Jensen, PhD – Mayo Clinic
Harvey Huang, BS – Mayo Clinic
Nicholas Gregg, MD – Mayo clinic, Rochester, Minnesota
Klaus-Robert Mueller, PhD – Technical University of Berlin
Dora Hermes Miller, Ph.D. – Mayo Clinic
Presenting Author: Kai Miller, MD, PhD – Mayo Clinic
Rationale:
Precise delineation of effective networks is critical for epilepsy surgery. Single-pulse electrical stimulation (SPES) evokes brain-stimulation evoked potentials (BSEPs) that can reveal these pathways, but thousands of heterogeneous traces are difficult to interpret. We combined several recently introduced, data-driven tools—canonical response parameterization (CRP) and basis profile curve (BPC) identification—to quantify BSEP structure and spatial origin in a large pediatric cohort. These were synthesized in an analytic framework that combines modified common-average re-referencing (mCAR), bipolar re-referencing, and wavelet-based spectral analysis to localize the tissue sources that generate BSEPs.Methods:
Twenty patients (6–18 y) undergoing stereo-EEG for drug-resistant epilepsy received 6 mA biphasic SPES. Signals were separately re-referenced with modified common average and bipolar montages, then high-pass/notch filtered. For every pair, CRP automatically extracted (i) a reproducible response shape, (ii) its duration (τR), and (iii) single-trial magnitudes (α), providing robust statistics despite variable morphology and noise. At each recording site we clustered α-weighted shapes with the BPC algorithm to uncover recurring temporal motifs linked to specific anatomical inputs. Broadband spectral change and referencing heuristics were then used to label each interaction as arising within, near, or far from the active tissue source.Results:
CRP revealed a rich diversity of canonical shapes beyond the traditional N1/N2, including prolonged depolarizing plateaus and fast polyphasic bursts. BPC clustering reduces this heterogeneity to several motifs per recording site within each subject, each mapping to discrete white-matter pathways or gyral inputs. Source-domain classification allows for clear delineation between BSEP sources that are were generated within local cortex, in near-field gray or juxtacortical white matter, or propagated from distant dipoles. For example, thalamus-to-sensorimotor stimulation produced far-field BSEPs that would have been mis-attributed to peri-rolandic connectivity without careful attention to the effect of referencing and the difference between evoked potentials and broadband changes.Conclusions:
By fusing CRP’s single-trial parametrization with BPC’s network-level clustering, we deliver an unsupervised, scalable pipeline that converts raw SPES data into anatomically interpretable maps. The framework distinguishes true synaptic connectivity from volume-conducted field spread, providing a practical advance for both circuit research and surgical planning. Findings from this largest pediatric SPES cohort underscore the promise of quantitative BSEP phenotyping to personalize therapy and deepen our understanding of human brain networks.Funding:
MAJ: NIHF31NS135898; KRM: BMBF for BIFOLD (01IS18037A); KJM: NIH U01-NS128612;