Neural Tracking of Sound Rhythms Correlates With Diagnosis, Severity, and Prognosis of Disorders of Consciousness

Abstract

Effective diagnosis and prognosis of patients with disorders of consciousness (DOC) provides a basis for family counseling, decision-making, and the design of rehabilitation programs. However, effective and objective bedside evaluation is a challenging problem. In this study, we explored electroencephalography (EEG) response tracking sound rhythms as potential neural markers for DOC evaluation. We analyzed the responses to natural speech and tones modulated at 2 and 41 Hz. At the population level, patients with positive outcomes (DOC-P) showed higher cortical synchronization to modulated tones at 41 Hz compared with patients with negative outcomes (DOC-N). At the individual level, phase coherence to modulated tones at 41 Hz was significantly correlated with Coma Recovery Scale-Revised (CRS-R) and Glasgow Outcome Scale-Extended (GOS-E) scores. Furthermore, SVM classifiers, trained using phase coherences in higher frequency bands or combination of the low frequency aSSR and speech tracking responses, performed very well in diagnosis and prognosis of DOC. These findings show that EEG response to auditory rhythms is a potential tool for diagnosis, severity, and prognosis of DOC.

Keywords: EEG; auditory steady state response; disorders of consciousness; machine learning; neural synchronization.

FIGURE 1

Phase coherence spectrum for responses to modulated tones (A) and speech (B) in the group of healthy controls (HC), patients with positive prognosis (DOC-P), and patients with negative prognosis (DOC-N). (A) Coherences to modulated tones at 41 and 41 ± 2 Hz were significant in HC, DOC-P, DOC-N, and there seemed to be no difference between HC and DOC-P. (B) Speech tracking response in delta and theta frequency bands for the group of healthy controls (HC) was significantly different from that of DOC-P group. The colored squares on top denote the frequency bins in which phase coherence is significantly higher compared with chance (P < 0.01, bootstrap, FDR corrected).

FIGURE 2

Topography of phase coherence. Phase coherence was separately normalized for each population by dividing by the 95th percentile of phase coherence across electrodes, to accurately illustrate the spatial distribution, and the values of the 95th percentile are shown on top of each plot.

FIGURE 3

Mean and SEM of the phase coherence between neural responses and acoustic envelope in the healthy controls (HC), patients with positive prognosis (DOC-P), and patients with negative prognosis (DOC-N). (A,D) There were no statistically significant differences in the 41 and 41 ± 2 Hz phase coherences between for HC and DOC-P. However, the 41 and 41 ± 2 Hz phase coherences were significantly higher in HC and DOC-P groups compared with those for DOC-N group. (B,E) The 2 and 4 Hz phase coherences in HC were significantly higher compared with those for DOC-P. (C,F) Phase coherences to natural speech in theta and delta bands were significantly different between HC and DOC-P groups (*P < 0.05, **P < 0.01; n.s., not significant).

FIGURE 4

Correlations between neural synchronization at 41, 41 ± 2 Hz, and clinical evaluations. (A,B) Correlations between the phase coherence at 41 Hz and CRS-R, and GOS-E scores. Phase coherence at 41 Hz was independently correlated with CRS-R and GOS-E scores (r = 0.59, P = 0.011, FDR corrected; r = 0.5, P = 0.027, FDR corrected). (C,D) Correlations between the phase coherence at 41 ± 2 Hz and CRS-R, and GOS-E scores. Phase coherence at 41 ± 2 Hz was correlated with CRS-R score (r = 0.55, P = 0.015, FDR corrected).

FIGURE 5

Correlations between neural synchronization at 2 and 4 Hz, and clinical evaluations. (A–D) Correlations between phase coherence at 2, 4 Hz, and CRS-R, GOS-E scores. Correlations between the phase coherence at 2, 4 Hz, and CRS-R, GOS-E were not statistically significant.

FIGURE 6

Correlations between the phase coherence for the speech tracking response in delta and theta bands and clinical evaluations. (A–D) Correlations between the phase coherence for the speech tracking response in delta, theta bands, and CRS-R, GOS-E score showed no statistical significance.

FIGURE 7

Classification and prognosis prediction in DOC using neural synchronization to temporal modulations at low and high frequencies. (A) Diagnosis classification using phase coherence at 41 and 41 ± 2 Hz performed well (AUC = 87.08%, accuracy = 70%, χ² = 17.052, P = 0.001). Confusion matrix generated by SVM showed 86.67% sensitivity and 87.5% specificity for MCS diagnosis. (B) The SVM classifier trained using phase coherence at 41 and 41 ± 2 Hz showed good performance in prognosis prediction of DOC (AUC = 83.4%, accuracy = 90%, χ2 = 14.072, P = 0.001) and was able to predict positive prognosis of individual patients with high sensitivity (88.24%) and specificity (78.57%). (C,D) The SVM classifier showed poor performance in classification and prognosis prediction of DOC trained using phase coherence at 2 and 4 Hz.

FIGURE 8

Classification and diagnosis prediction using speech tracking response and combination of low frequency aSSR and speech tracking responses. (A,B) Confusion matrix generated by the SVM classifier trained by the phase coherence to the natural speech in delta and theta bands for classification and prognosis prediction of DOC. (C,D) Confusion matrix generated by the SVM classifier trained by the combination of the phase coherence to modulated tones at 2, 4 Hz, and the phase coherence to natural speech in delta, theta bands for classification and prognosis prediction of DOC. The SVM classifiers showed better performance in classification and prognosis prediction of DOC compared with SVM classifiers separately trained with phase coherence at 2 and 4 Hz or the phase coherence to natural speech in delta and theta bands.