Detection of Spreading Depolarization with Noninvasive Electroencephalography

Spreading depolarizations (SDs) are waves of near-complete breakdown of the transmembrane ion gradients that underlie migraine aura and occur in patients with moderate-to-severe traumatic, ischemic or hemorrhagic brain injury. SDs are involved in primary and secondary brain injury. They are associated with poor neurological outcome and represent a modifiable therapeutic target. However, despite decades of investigation, SD detection still requires invasive subdural electrocorticography (ECoG) using an implanted subdural electrode strip, restricting monitoring to a small subset of neurosurgical patients.

We developed WAVEFRONT, a noninvasive scalp electroencephalography (EEG) biomarker of SD based on spatiotemporal propagation characteristics of delta-band EEG activity. The biomarker was trained and validated using concurrent, multi-day scalp EEG and invasive subdural electrocorticography recordings from patients with traumatic brain injury (TBI), subarachnoid hemorrhage (SAH), or malignant hemispheric stroke (MHS) across three institutions. We assessed out-of-sample performance across institutions, diagnosis, and surgical approaches, as well as calibration, temporal synchrony, and spatial specificity relative to gold-standard SDs annotated on ECoG. Clinical utility was further evaluated in 607 patients comatose after cardiac arrest from the multicenter International Cardiac Arrest REsearch (I-CARE) consortium, a population in whom invasive monitoring is not performed. Outcome analyses excluded patients who died (Cerebral Performance Category [CPC] 5) because death may reflect decisions to withdraw life-sustaining treatment in addition to neurological injury.

A total of 1,119 gold-standard SDs were annotated in 32 patients (TBI, n=17; SAH, n=12; MHS, n=3; range 0-187 SDs per patient). Across exhaustive cross-validation and institution-, diagnosis-, and surgical approach-based hold-out analyses, WAVEFRONT models trained on one set of patients consistently maintained high performance when evaluated in previously unseen out-of-group cohorts. At the optimal operating point, WAVEFRONT achieved a mean cross-validated true positive rate of 97.3±1% with a false positive rate of 6.0±0.5%, calculated over 2-min sliding windows with 30-s overlaps. The cumulative duration of WAVEFRONT detections within 30-hour windows strongly correlated with the number of SDs measured by invasive ECoG (R ² =0.7411). Scalp-detected depressions were spatially consistent with the location of the invasive recording strip, and visual inspection demonstrated close temporal correspondence with invasive detections, including during SD clusters. External validation in the independent I-CARE cohort using WAVEFRONT trained on the invasive-monitoring cohort demonstrated significantly greater SD burden on day 3 of post-arrest EEG monitoring among patients with poor neurological outcome (CPC 3-4, n=29) than those with good outcome (CPC 1-2, n=225; p=0.007).

Eight decades after Leão's description of SD in 1944, we propose and validate a noninvasive scalp EEG biomarker of SD. Consistent performance across institution-, diagnosis-, and surgical approach-based hold-out analyses, together with burden calibration, spatial specificity, and temporal synchrony with ECoG detections, supports the validity of the WAVEFRONT framework. The association of biomarker burden with neurologic outcome after cardiac arrest suggests that noninvasive monitoring may enable investigation of spreading depolarizations in populations where invasive monitoring is not feasible. Future studies should evaluate prospective performance, treatment responsiveness, and disease-specific associations of the SD biomarker with clinical outcome.