Automated detection of spreading depolarizations
Computer-implemented methods and automated systems for real-time detection of spreading depolarizations in a brain injured patient, based an algorithm of (a) providing a reference data base of spreading depolarization waveform templates generated from EEG recordings of confirmed spreading depolarizations (SD) in a reference brain-injured patient cohort; (b) recording an EEG of the brain injured patient to generate recorded EEG waveforms; (c) detecting a slow potential change present in a recorded EEG waveform by applying a power spectral density estimate to the recorded waveform; (d) comparing a detected SPC to a reference database of SD waveform template to identify a candidate SD; and (e) rejecting a candidate SD as a false positive based on overall signal power and amplitude analysis and identifying a non-rejected candidate SD as a detected SD.
1. A non-transitory computer-readable storage medium containing program instructions for causing a computer to perform a method of real-time detection of spreading depolarizations in a brain injured patient, the method comprising:
(a) providing a reference database of spreading depolarization (SD) waveform templates generated from EEG recordings of confirmed spreading depolarizations in a reference brain-injured patient cohort;
(b) recording, by an EEG device, an EEG of the brain injured patient to generate recorded EEG waveforms;
(c) detecting a slow potential change (SPC) present in the recorded EEG waveform by applying a power spectral density estimate to the recorded EEG waveform;
(d) comparing the detected SPC to the reference database of SD waveform template to identify a candidate SD; and
(e) rejecting the candidate SD as a false positive based on overall signal power and amplitude analysis and identifying a non-rejected candidate SD as a detected SD.
2. The non-transitory computer-readable storage medium according to claim 1 , wherein the method further comprises:
(f) testing a detected SD for depression of high-frequency power and identifying a detected SD as an SD with depression if depression of high-frequency power is observed, wherein high-frequency is frequency greater than 0.5 Hz.
3. The non-transitory computer-readable storage medium according to claim 1 , wherein the power spectral density estimate according to step (c) comprises an autoregressive power spectral density estimate based on selection of a model order of about 20.
4. The non-transitory computer-readable storage medium according to claim 1 , wherein total signal power is between 1 μV2 and 30 μV2 across a 30 second window, and if the total signal power is greater than 30 μV2 then no further analysis is performed within the 30 second window.
5. The non-transitory computer-readable storage medium according to claim 1 , wherein the recorded EEG waveform is identified as a detected SPC according to (c) if, upon application of the AR power spectral density estimate, the recorded EEG waveform exhibits a peak between 0.004 Hz and 0.009 Hz, within a period of 110-250 seconds, said peak having a peak power, a power ratio R1>1.5 and a power ratio R2>5, wherein R1 is a ratio of the peak power to power in a neighboring lower frequency band and R2 is a ratio of the peak power to power in a neighboring higher frequency band.
6. The non-transitory computer-readable storage medium according to claim 1 , wherein comparing according to (d) comprises calculating a coefficient of correlation between the detected SPC and the reference waveform database template and scoring the detected SPC as a candidate SD if r2>0.95.
7. The non-transitory computer-readable storage medium according to claim 6 , wherein the candidate SD is rejected as a false positive if a calculated peak-to-peak value of the candidate SD is measured as less than 500 μV or greater than 35 mV.
8. The non-transitory computer-readable storage medium according to claim 2 , wherein (f) comprises: (i) calculating a mean high-frequency power in a time frame prior to a template marker; (ii) calculating a mean high-frequency power in a time frame subsequent to the template marker; (iii) calculating a ratio of (i) to (ii); and (iv) identifying the detected SD as a detected SD with depression if the ratio is <0.66.
9. The non-transitory computer-readable storage medium according to claim 1 , wherein the EEG is intracranial EEG or scalp EEG.
10. The non-transitory computer-readable storage medium according to claim 2 , wherein he detected SD is further analyzed for propagation, the method comprising: (g) detecting an SD on a neighboring electrode and classifying the detected SD and the detected neighboring SD as propagating from a same event if the detected neighboring SD occurs within eight (8) minutes of the detected SD.
11. The non-transitory computer-readable storage medium according to claim 2 , wherein each detected SD is assigned a confidence classification and one or more notification thresholds are set for notifying a clinician of the detected SD, said thresholds set according to a confidence classification.
12. The non-transitory computer-readable storage medium according to claim 1 , further comprising adding waveforms for the detected SD into the reference database.
13. An automated clinical method for rapid triage of brain-injured patients by detecting spreading depolarizations, the method comprising: effectuating the automated algorithm according to claim 10 and assigning a greater urgency to patients exhibiting a detected SD over patients not exhibiting a detected SD, and a greater urgency to patients exhibiting a propagating SD over patients not exhibiting a propagating SD.
14. The automated clinical method according to claim 13 having a sensitivity for detecting a verified spreading depolarization of greater than 90% and a selectivity for detecting a verified spreading depolarization of greater than 80%.