Predicting stroke severity with a 3-min recording from the Muse portable EEG system for rapid diagnosis of stroke

Figure 1. Pairwise Derived Brain Symmetry Index by group and frequency. Pairwise derived brain symmetry index by group across all frequencies (1–30 Hz) at electrodes TP9 and TP10 (A), pdBSI by severity across all frequencies (B), pdBSI by severity for each frequency bin (C). Standard error was calculated for each bar. *Significance at p < 0.050. Brain activity is more symmetrical after stroke. Average pdBSI does not readily distinguish between stroke severities. Data is presented as mean ± SEM.

Figure 2. Pairwise Derived Brain Symmetry Index across frequency by stroke severity at electrodes TP9 and TP10, DAR across severities, and DTABR across severities. **Significance at p = 0.01. In stroke patients, pdBSI increases at low frequencies (A). However, above ~ 10 Hz, pdBSI decreases in stroke patients. When stroke patients are grouped by severity (B), it is difficult to differentiate severity by pdBSI at higher frequencies, largely due to high variability in the severe stroke group. DAR increases as stroke severity increases in moderate and severe strokes (C). We do not observe any clear hemispheric differences in DAR. DTABR increases with increasing severity (D). No clear hemispheric differences have been observed, but in small and moderate stroke there is a trend for increased slowing on the injured hemisphere. Data are presented as mean ± SEM.

Figure 3. Background fit of data from fooof analysis at each electrode by severity. Non-oscillatory activity was not different between severities, indicating that band-ratios are not affected by background brain activity levels.

Figure 4. RMS for accelerometer (A) and gyroscope (B) across three planes of movement. RMS of accelerometer data increased with increasing stroke severity in the X-plane and decreased with stroke severity in the Z-plane (A). No significant group differences in the gyroscope data (B). Data are presented as mean ± SEM.

Figure 5. Classification trees for determining (A) stroke severity, (B) presence of stroke, (C) higher-risk stroke. Predicting stroke severity was the least accurate model and predicting more severe strokes was the most accurate model.