Electroencephalographic signatures of migraine in small prospective and large retrospective cohorts

Abstract

Migraine is one of the most common neurological disorders in the US. Currently, the diagnosis and management of migraine are based primarily on subjective self-reported measures, which compromises the reliability of clinical diagnosis and the ability to robustly discern candidacy for available therapies and track treatment response. In this study, we used a computational pipeline for the automated, rapid, high-throughput, and objective analysis of encephalography (EEG) data at Cleveland Clinic to identify signatures that correlate with migraine. We performed two independent analyses, a prospective analysis (n = 62 subjects) and a retrospective age-matched analysis on a larger cohort (n = 734) obtained from the sleep registry at Cleveland Clinic. In the prospective analysis, no significant difference between migraine and control groups was detected in the mean power spectral density (PSD) of an all-electrodes montage in the frequency range of 1-32 Hz, whereas a significant PSD increase in single occipital electrodes was found at 12 Hz in migraine patients. We then trained machine learning models on the binary classification of migraine versus control using EEG power features, resulting in high accuracies (82-83%) with occipital electrodes' power at 12 Hz ranking highest in the contribution to the model's performance. Further retrospective analysis also showed a consistent increase in power from occipital electrodes at 12 and 13 Hz in migraine patients. These results demonstrate distinct and localized changes in brain activity measured by EEG that can potentially serve as biomarkers in the diagnosis and personalized therapy for individuals with migraine.

Keywords: Electroencephalography; Machine learning; Migraine.

Fig. 1

No significant difference in PSD of the all-electrodes montage between migraine patients and health controls. (A) Power spectral density of averaged 16 electrodes in the range of 1–30 Hz in control (n = 25) and migraine groups (n = 37). Shaded areas represent standard error of the mean (SEM). (B) Mean power in the frequency bands delta (1–4 Hz), theta (5–9 Hz), alpha (10–13 Hz), and beta (14–32 Hz) for migraine and control patients. Error bars represent SEM.

Fig. 2

Statistical significance of difference in PSD between migraine patients and healthy controls based on individual electrodes. Heat map shows t-test p-values for individual electrodes in migraine (n = 37) and control (n = 25). The highlighted cells (O1, O2 at 12 Hz) are the ones that represented the lowest p values (i.e. highest statistical significance; O2 p = 0.009, O1 p = 0.010). We used uncorrected p-values in this exploratory part of the EEG analysis mainly to highlight regions of interest, which were further validated in subsequent analyses that corrected for multiple comparisons (see Fig. 3).

Fig. 3

Comparison of PSD between migraine patients and healthy controls in both prospective and retrospective cohorts. (Upper panel)—The figure on the left shows PSD for averaged electrodes in the range of 4–20 Hz in age-matched migraine patients and healthy controls in the prospective arm (n = 25). The figure on the right shows the same in age- and ESS-matched migraine patients and healthy controls in the retrospective arm (n = 367). (Lower panel)—Difference in PSD between migraine and control for retrospective (blue) and prospective (green) groups. Z score of PSD difference between migraine and control for retrospective (blue) and prospective (green) groups. Horizontal line reflects adjusted significance threshold.