Table 1 EEG during acute clinical assessment to identify stroke versus non-stroke conditions
Reference | Reference standard | Participants | Key exclusions | First EEG start time after onset (h) | EEG procedure | EEG processing | EEG biomarker | Result | Quality score |
|---|---|---|---|---|---|---|---|---|---|
Cohen 1977 [27] | Specialist opinion based on ‘routine clinical assessment form’ | 26 Ischaemic Stroke patients, 26 controls | Previous stroke |  < 72 | 19 electrodes | Offline filter 0.35-35 Hz, 1 min epochs | Absolute spectral power | Stroke participants exhibited significant interhemispheric delta power asymmetry vs non-stroke (p < 0.05) | 2 |
Yan 2011 [28] | Specialist opinionb | 22 Stroke patients, 10 controls | Not Reported |  < 48 | 16 electrodes, eyes closed, resting | Offline visual artifact removal followed by digital filter, 10 s epochs. FFT | BBSI | Higher BBSI in stroke vs non-stroke (diagnostic accuracy = 83% when conscious, 71.43% unconscious) | 2 |
Aminov 2017 [25] | Specialist opinionb | 15 Ischaemic Stroke patients, 4 Haemorrhagic Stroke patients, 19 controls (database) | History of neurological/ psychiatric disorders |  < 72 | Single electrode at FP1, eyes closed | Online filter 0.5-30 Hz, manual artifact removal, 4 s epochs; FFT | Relative spectral power (DAR, DTR) | Less theta power (p = 0.02), more delta (p < 0.01) power, higher DAR (p < 0.01) and DTR (p = 0.01) in stroke participants vs non-stroke | 4 |
Erani 2020 [23] | Specialist opinionb | 43 Ischaemic Stroke patients, 7 Haemorrhagic Stroke patients, 13 TIA patients, 37 Stroke mimics | Not Reported |  < 23 | 17 electrodes, portable, dry electrode system, eyes open, resting | Offline analysis: filtering, noise removal and re-referencing. EEG variables selected using Lasso regression | Relative spectral power (all bands, beta split into low and high) Diagnostic neural network | Deep learning EEG (4 lasso selected electrode pairs) and clinical data model could identify stroke/TIA from mimic (AUC = 0.88, sensitivity = 79%, specificity = 80%) more accurately than combined clinical and EEG (4 electrode pairs) data (AUC = 0.80, sensitivity = 70%, specificity = 80%) and individual EEG (4 electrode pairs) (AUC = 0.78, sensitivity = 65%, specificity = 80%) or clinical (AUC = 0.62, sensitivity = 40%, specificity = 80%) data models. Less high frequencies (alpha and high beta 20.5-28 Hz) and greater low frequencies (low beta 12.5-16 Hz) associated with stroke/TIA | 4 |
Rodriguez 2012 [29] | Admission CT/ Specialist opinion | 29 Ischaemic Stroke patients, 15 Haemorrhagic Stroke patients (all MCA), Unknown no. of controls (database) | Not Reported |  < 72 | Not Reported | Not Reported | Relative spectral power (all bands, DAR, PRI) | Significant increase in slow wave frequencies (< 6.25 Hz) and decrease in alpha/beta in stroke versus control. Significantly greater PRI and DAR in stroke patients vs non-stroke (abstract-no statistics given) | 3 |
Chen 2018 [30] | Specialist opinion informed by CT | 47 Haemorrhagic Stroke patients, 15 controls | Ruptured aneurysm; vascular malformation or stenosis; cerebral trauma; tumour; encephalitis; ischaemic stroke; previous stroke; CNS depressants |  < 59 | Controls eyes closed and awake | Offline filters > 0.3, < / = 30 Hz, artifacts removed. FFT | Relative spectral power delta, alpha, DAR, DTABR), BSI | Lower alpha power, greater delta power and higher DAR and DTABR in stroke patients vs non-stroke (all p < 0.0001). BSI was not significantly different | 4 |
Chan 2019 [31] | Specialist opinion informed by CT | 32 patients (Ischaemic Stroke and control; unclear division) | Haemorrhagic Stroke |  < 72 | 32 electrodes, eyes open and closed, resting but conscious, hyperventilation and photic stimulation | Sampling 250 Hz and 512 Hz, FFT; DWT (Daubechies 4) | Relative spectral power (DAR, DTABR), BSI | Higher BSI, DAR, DTABR and greater delta power in stroke patients vs non-stroke. EEG identified stroke with > 87.5% accuracy | 2 |
Machado 2004 [32] | Specialist opinion based on CT, MRI, medical history and neurological exam | 32 Ischaemic (LMCA) Stroke patients 211 controls | Not Reported |  < 24 | 19 electrodes | Online filters < / = 0.5, > 30, 60 Hz notch filter, sampling 200 Hz, EOG artifact removal, 2.56 s epochs | Tomography | Greater delta and theta and less alpha power in the territory of the stroke (all p < 0.01) compared to the same territory in non-stroke | 3 |
Finnigan 2016 [33] | Specialist opinion, based on CT/MRI within 6 h of onset | 18 Ischaemic (LMCA) Stroke patients, 28 controls | Non-cortical stroke; bilateral stroke; seizures; haemorrhage; previous neurological conditions; previous stroke; encephalitis |  < 24 | 19 electrodes, eyes closed with checking for wakefulness | Sampling 500HZ, offline filter 0.5-40 Hz, 12 dB/octave,, EOG artifact removal, 2 s epochs | Relative spectral power (all bands, DAR, DTABR) | Greater delta (p < .0001, AUC = 0.99, sensitivity = 94%, specificity = 96%) and theta (p < .001, AUC = 0.81, sensitivity = 89%, specificity = 68%), lower alpha (p < .0001, AUC = 0.97, sensitivity = 89%, specificity = 93%) and beta (p < .0001, AUC = 0.9, sensitivity = 83%, specificity = 82%), higher DAR, (p < .0001, AUC = 1.0, sensitivity = 100%, specificity = 100%), DTABR (p < .0001, AUC = 0.99, sensitivity = 100%, specificity = 96%) and QSlowing (p < .0001, AUC = 0.97, sensitivity = 94%, specificity = 96%) in stroke vs non-stroke (p < .001) | 3 |
Rogers 2019 [21] | Specialist opinion based on CT, MRI, echocardiogram, bloods & ultrasound or CTA | 10 Ischaemic Stroke patients, 10 controls | History of neurological/ psychiatric disorders; current haemorrhagic stroke |  < 72 | Single electrode at FP1, Auditory Oddball EP, eyes closed and resting | Offline. filter 0.5-30 Hz, manual artifact removal | Relative spectral power (all bands) | Greater delta (AUC = 0.87, sensitivity = 90%, specificity = 85%) and less theta (AUC = 0.93, sensitivity = 85%, specificity = 90%) power in stroke vs control (both p = 0.03) | 5 |
Gottlibe 2020 [34] | Specialist opinion based on CT/MRI at baseline/admission | 33 Ischaemic Stroke patients, 25 controls | Degenerative neurological conditions; Seizure/epileptiform EEG |  < 48 | 4 electrodes. Awake, alert, sitting position | Sampling 220 Hz, offline computer artifact removal, 10 min overlapping epochs, filter 0.16-76 Hz | r-BSI | Higher r-BSI in stroke vs non-stroke (p = 0.002) | 3 |
Finnigan 2020 [35] a | Specialist opinion, based on CT/MRI within 6 h of onset | 18 Ischaemic Stroke (LMCA) patients, 28 controls | Non-cortical stroke; bilateral stroke; seizures; haemorrhage; previous neurological conditions; previous stroke; encephalitis |  < 24 | Six electrodes, eyes closed with checking for wakefulness | Offline filter 0.5-40 Hz, 12 dB/octave, EOG artifact removal, 2 s epochs | Relative spectral power (DAR) | Higher DAR stroke participants vs non-stroke using two frontal electrodes (F3-F4). AUC = 0.99, sensitivity = 93%, specificity = 94% | 4 |
Murri 1998 [36] | CT within 4 days of onset | 65 Ischaemic Stroke patients, 60 controls | Bilateral stroke; previous stroke; gradual onset; neurological or systemic pathologies |  < 24 | Eyes closed, supine with eye open breaks in a quiet, dimly lit room | Online filter 1-50 Hz, time constant 0.3 s, manual artifact removal, 4 s epochs | Topographic activity | Greater maximum delta power was observed in patients versus control subjects for cortical lesions: frontocentral p < 0.01, AUC = 0.68, sensitivity = 92%, specificity = 45%; Temporal p < 0.01, AUC = 0.85, sensitivity = 88%, specificity = 83%; Parieto-occipital p < 0.01, AUC = 0.75, sensitivity = 79%, specificity = 72%, (diagnostic accuracy extrapolated from true and false positive and negative values). Cortical lesions could be located using the electrode with maximum delta power (Kappa = 0.63 (0.39–0.87)) after striatocapsular lesions excluded. Amongst stroke patients conventional and topographic EEG had 73 and 84% sensitivity respectively for detecting focal lesions | 4 |
Luu 2001 [37] | CT or MRI | 6 Ischaemic Stroke patients, 16 controls | Haemorrhagic Stroke; Non-cortical Stroke; Previous stroke/other brain lesions; state altering or confounding medications; NIHSS < 8 |  < 36 | Variable no of electrodes tested (19–128), eyes open and closed | Online filter 0.1-59 Hz, artifacts removed, 1 s epochs | Topographic activity | Increased slow wave (delta and theta) amplitude 2 standard deviations above mean in stroke related EEG versus control but only in 4/6 (67%) patients | 5 |
Shreve 2019 [22] | CT, MRI and NIHSS | 11 Ischaemic Stroke patients, 3 TIA patients, 10 mimic patients | Haemorrhagic Stroke |  < 43.5 | 256 electrodes but 62 excluded, awake, fixed gaze with bed at 30 degree angle | Offline-only sixth order < 50 Hz filter, independent component analysis artifact removal, 1 s epochs | Relative spectral power (All bands, global power, DAR, DTABR) | No EEG measure significantly distinguished cerebral ischaemia from non-ischaemia | 4 |
Finnigan 2004 [12] | MRI (DWI) 15 h ± 3 h | 11 Ischaemic Stroke patients, 6 controls | Fever, encephalitis, seizures, ICH, non-cortical stroke, confounding neurological condition (e.g. previous stroke) or medication |  < 9 | 64 electrodes, between MRI scans | Online filter .01-100 Hz, artifacts 0.2- 40 Hz, automatic artifact removal, 4 s epochs, sampling 500 Hz, FFT .5-50 Hz | Relative spectral power (aDCI) | Significantly greater mean delta power in patients versus controls (t = 4.68, P = 0.001). Control aDCI was at least 1 order of magnitude lower than the lowest patient aDCI | 3 |
Sheorajapanday 2009 [26] | MRI within 5 days | 21 Ischaemic Stroke patients, 10 controls | Not Reported |  < 72 | 20 electrodes. Eyes closed, alert | Offline filter(s) > 0.3, < / = 30, manual artifact removal, FFT | Relative spectral power (all bands, DAR, DTAR, DTABR), pdBSI | pdBSI distinguished stroke from control patients (p = 0.0003; 1-25 Hz range p = 0.001) and correlated with clinical and radiological status (P’s < 0.001). No significant differences between groups for RAP, RDP, RDTP, DAR, DTAR or DTABR | 3 |
EEG to distinguish stroke from Transient Ischaemic Attack (TIA) | |||||||||
Rogers 2019 [21] | Specialist opinion | 10 Ischaemic Stroke patients, 10 TIA patients | Neurological/ psychiatric disorders, SAH |  < 72 | Single electrode at FP1, Auditory Oddball EP, eyes closed and resting | Offline, filter 0.5-30 Hz, manual artifact removal | Relative spectral power (all bands) | Greater delta (AUC = 0.87, sensitivity = 90%, specificity = 85%) power in stroke vs TIA (p < 0.01). Greater alpha (AUC = 0.81, sensitivity = 80%, specificity = 90%) and beta (AUC = 0.86, sensitivity = 90%, specificity = 80%) power in TIA vs stroke (both p < 0.01) | 5 |
Sheorajapanday 2009 [26] | MRI within 5 days | 21 Ischaemic Stroke patients, 10 TIA patients | Not Reported |  < 72 | 20 electrodes. Eyes closed, alert | Offline filter(s) > 0.3, < / = 30, manual artifact removal, FFT | Relative spectral power (all bands, DAR, DTAR, DTABR), pdBSI | pdBSI distinguished stroke from TIA patients (p = 0.0003; 1-25 Hz range p = 0.001). No significant differences between groups for RAP, RDP, RDTP, DAR, DTAR or DTABR | 3 |