 
Abstract— Hypoxic ischemic encephalopathy (HIE) is a 
severe consequence of cardiac arrest (CA) representing a 
substantial diagnostic challenge. We have recently designed a 
novel method for the assessment of HIE after CA. The method 
is based on estimating the severity of the brain injury by 
analyzing changes in the electroencephalogram (EEG) slow 
wave activity while the patient is exposed to an anesthetic drug 
propofol in a controlled manner. In this paper, Hilbert-Huang 
Transform (HHT) was used to analyze EEG slow wave activity 
during anesthesia in ten post-CA patients. The recordings were 
made in the intensive care unit 36-48 hours after the CA in an 
experiment, during which the propofol infusion rate was 
incrementally decreased to determine the drug-induced 
changes in the EEG at different anesthetic levels. HHT was 
shown to successfully capture the changes in the slow wave 
activity to the behavior of intrinsic mode functions (IMFs). 
While, in patients with good neurological outcome defined after 
a six-month control period, propofol induced a significant 
increase in the amplitude of IMFs representing the slow wave 
activity, the patients with poor neurological outcome were 
unable to produce such a response. Consequently, the proposed 
method offer substantial prognostic potential by providing a 
novel approach for early estimation of HIE after CA. 
I. INTRODUCTION 
ARDIAC arrest (CA) is a condition in which the 
circulation of blood is ceased due to the failure of the 
 
This work was financially supported by Medical Research Center Oulu 
(MRC Oulu), grant 40273/14 from Tekes - Finnish Funding Agency for 
Innovation, Orion Research Foundation, Instrumentarium science 
foundation, Emil Aaltonen foundation, Oulu University Scholarship 
Foundation, Finnish Foundation for Cardiovascular Research, and Finnish 
Science Foundation for Economics and Technology. 
J. Kortelainen is with the Physiological Signal Analysis Team, Center 
for Machine Vision and Signal Analysis, University of Oulu, Finland and 
Department of Clinical Neurophysiology, Medical Research Center Oulu, 
University of Oulu and Oulu University Hospital, Finland (e-mail: 
jukortel@ee.oulu.fi). 
E. Väyrynen and T. Seppänen are with the Physiological Signal Analysis 
Team, Center for Machine Vision and Signal Analysis, University of Oulu, 
Finland. 
U. Huuskonen is with the Department of Clinical Neurophysiology, 
Medical Research Center Oulu, University of Oulu and Oulu University 
Hospital, Finland. 
J. Laurila, J. Koskenkari and T. Ala-Kokko are with the Unit of Surgery, 
Anaesthesia and Intensive Care, Medical Faculty, University of Oulu, 
Finland and Division of Intensive Care Medicine, Medical Research Center 
Oulu, University of Oulu and Oulu University Hospital, Finland. 
J. T. Backman is with the Department of Clinical Pharmacology, 
University of Helsinki and Helsinki University Hospital, Finland. 
S. Alahuhta is with the Unit of Surgery, Anaesthesia and Intensive Care, 
Medical Faculty, University of Oulu, Finland and Department of 
Anaesthesiology, Medical Research Center Oulu, University of Oulu and 
Oulu University Hospital, Finland. 
heart to contract effectively. Approximately 360,000 out-of-
hospital CAs occur in the USA alone each year [1]. While 
half of the patients admitted to hospital and intensive care 
unit (ICU) survive [2], a major part of the survivors suffer 
from permanent brain damage called hypoxic ischemic 
encephalopathy (HIE) due to global cerebral ischemia during 
the CA [3]. Consequently, neurological complications 
represent a substantial source of costs and loss of quality of 
life being the leading cause of disability after CA. 
Incorporating the brain in any cardiopulmonary resuscitation 
and follow-up treatment strategies has thus been demanded 
in the recent guidelines of American Heart Association [4]. 
 At the moment, there is no reliable technological solution 
for early estimation of HIE after CA. For example, regular 
imaging modalities such as computed tomography or 
magnetic resonance imaging do not reliably estimate the 
diffuse brain injury resulting from hypoxia. 
Electrophysiological measurements as well as specific serum 
proteins in addition to clinical examination provide today’s 
golden standard for the assessment of HIE, giving, however, 
only a rough estimate of the damage [5],[6]. The lack of 
reliable estimation of HIE complicates the clinical decision 
making as well as the development of new treatments of this 
condition. 
 We have recently designed a novel method for the 
assessment of HIE after CA. The method is based on 
estimating the severity of the injury by analyzing changes in 
the electroencephalogram (EEG) while the patient is 
exposed to an anesthetic drug propofol in a controlled 
manner. According to our recent findings [7], hypoxic brain 
injury fundamentally disturbs the neural system responsible 
for the generation of slow waves, the most important EEG 
signatures of non-rapid eye movement sleep [8] also seen 
during general anesthesia [9]. By measuring the EEG 
response to propofol, this disturbance can be revealed and 
used as a prognostic tool for irreversible brain injury. 
 In this paper, Hilbert-Huang Transform (HHT) [10], an 
adaptive data-driven method designed for the analysis of 
non-stationary signals, is used to assess EEG slow wave (< 1 
Hz) activity during anesthesia in post-CA patients. We have 
previously shown the method to be able to extract stable 
signal components representing slow wave activity from 
EEGs of neurologically healthy patients during propofol 
anesthesia [11]. Here we investigate whether these signal 
components can be extracted from the data collected in the 
ICU during controlled propofol exposure and utilized in the 
Using Hilbert-Huang Transform to Assess EEG Slow Wave Activity 
During Anesthesia in Post-Cardiac Arrest Patients 
Jukka Kortelainen, Member, IEEE, Eero Väyrynen, Usko Huuskonen, Jouko Laurila, Juha 
Koskenkari, Janne T. Backman, Seppo Alahuhta, Tapio Seppänen, and Tero Ala-Kokko 
C 
978-1-4577-0220-4/16/$31.00 ©2016 IEEE 1850
  
 
estimation of the severity of HIE. The paper is organized so 
that Section II describes the data collection procedure and 
the EEG analysis. The results are given in Section III. In 
Section IV, the conclusions of the study are presented. 
II. MATERIALS AND METHODS 
A. Experimental Protocol 
The study was approved by The Regional Ethics 
Committee of the Northern Ostrobothnia Hospital District. 
Ten comatose patients resuscitated from out-of-hospital CA 
participated in the study. Patients with earlier disease 
affecting the central nervous system were excluded. The 
patients had received hypothermia treatment as a 
neuroprotective measure before the experiment, which was 
carried out 36–48 h after the CA while the patients were still 
sedated with propofol. 
 In the experiment, EEG was recorded with 19-channel 
electrode cap according to the 10/20 international system 
using Nicolet nEEG Modular Neurodiagnostic System with 
a v32 Amplifier. A sampling frequency of 500 Hz and 
bandwidth of 0.053 – 125 Hz was used and the signals were 
referenced to common average. During the experiment, the 
propofol infusion rate was incrementally decreased 
 
Fig. 1. The propofol infusion rate, blood concentration and EEG during the experimental protocol. Above, the propofol infusion rate (black line) and 
blood concentration (green line) is given in relation to the EEG (blue data) during the entire experimental protocol for one patient with good outcome. 
The infusion rate was decreased step-wise every 30 min and the propofol concentration was determined on every step just before the next decrease as 
well as in the end of the experiment (green circles). Below, short EEG signal samples and the corresponding IMFs (3-11) at four different phases (A-D) 
of the experiment are presented. 
1851
  
 
 
Fig. 2. Instantaneous amplitude (higher row) and frequency (lower row) 
of intrinsic mode functions (IMFs) 5, 7, 9, and 11 for one patient with 
good outcome. The IMFs are extracted from the EEG recorded during 
the entire experiment. As in Fig. 1, time point zero represents the first 
decrease in the propofol infusion rate. 
 
Fig. 3. The mean instantaneous frequencies of different intrinsic mode 
functions (IMFs) calculated for the 30-s signal samples. The values are 
given in logarithmic scale and each box shows the median, range, and 
25th and 75th percentiles. 
following a predefined protocol (Fig. 1) to determine the 
drug-induced changes in the EEG at different anesthetic 
levels. The decrease was started from the highest acceptable 
infusion rate during the intensive care (4 mg  kg–1  h–1) 
and continued step-wise every 30 min until the drug 
administration was finally switched off. Same infusion rates 
(4, 3, 2, 1, 0.5, and 0 mg  kg–1  h–1) were used for all 
patients. To determine the propofol concentration, a blood 
sample was taken before each change in the infusion rate as 
well as in the end of the experiment, i.e. 30 min after turning 
propofol off (Fig. 1). The severity of the HIE hypoxic-
ischemic brain injury was determined by evaluating the 
neurological recovery of the patient six months after the CA 
using the Cerebral Performance Category (CPC). Based on 
this, the patients were assigned to either good (CPC 1–2) or 
poor (CPC 3–5) outcome group. 
B. EEG Analysis 
The signal processing was carried out with Matlab 
technical computing language, version 2011b (The 
MathWorks, Inc., Natick, MA). In the study, only the data 
from channel Fp1 were used. This frontal channel was 
selected as our earlier study showed pronounced EEG 
responses to the anesthetic in the prefrontal area [7]. For the 
EEG analysis, 5-min signal samples were picked just before 
each decrease of the drug infusion rate and in the end of the 
experiment corresponding to the collection of the drug 
concentration blood samples. From each 5-min signal 
sample a 30-s representative artifact-free sequence was 
selected for further analysis. The signals were then filtered 
using a low-pass FIR filter with a cutoff frequency of 48 Hz 
to reduce high-frequency noise. 
To investigate the dynamical changes in the signal 
characteristics, HHT was applied to the EEG data. First, the 
method was applied to the entire EEG recording of one 
patient to illustrate the signal changes as a function of time. 
Then, the method was applied to the 30-s signal sequences 
representing different propofol infusion rates to examine the 
differences between the good and poor outcome groups in 
the whole dataset. The HHT algorithm contains two 
principal parts:  
1. Empirical Mode Decomposition (EMD) 
2. Hilbert Spectral Analysis (HSA) 
In EMD, the signal is iteratively decomposed into 
components called intrinsic mode functions (IMFs) each 
component containing lower frequency oscillations than the 
preceding one. In HSA, Hilbert transform is used to compute 
the instantaneous frequency and amplitude of different 
IMFs. A more detailed description of the HHT applied is 
given in [11]. 
III. RESULTS 
In Fig. 1, the propofol infusion rates, blood concentrations 
and EEG during the entire experimental protocol are 
presented for one patient with good outcome. In addition, 
short EEG signal samples and the corresponding IMFs at 
four different phases of the experiment are presented. 
Lowering the infusion rate decreased the amount of drug in 
the blood which consequently is seen in the EEG as a 
reduced anesthetic effect. In signal characteristics, this 
means a shift of power from lower frequencies (IMFs 8-11) 
to higher frequencies (IMFs 3-7). When applied to this entire 
EEG recording, EMD extracted a total of 23 IMFs. Fig. 2 
presents the instantaneous frequencies and amplitudes of 
IMFs 5, 7, 9, and 11 calculated for the same recording. The 
shift of activity from lower frequencies to higher frequencies 
1852
  
 
 
Fig. 4. The effect of propofol on the amplitude of intrinsic mode 
function (IMF) 5 in patients with good and poor neurological outcome. 
The amplitudes are given at different propofol infusion rates for 
patients with good (N = 6) and poor (N = 4) neurological outcome. The 
values represent mean instantaneous amplitudes calculated for the 30-s 
signal samples given relative to the individual value at propofol 
infusion rate 0 mg  kg–1  h–1. Data are means; error bars show SD. 
Asterisks indicate statistically significant difference between groups 
(*p < 0.05, unpaired two-sample t-test). 
is clearly seen in the behavior of IMFs of which 5 and 7 
substantially increase their amplitude whereas the amplitude 
of IMF 11, representing slow wave activity, is halved. 
The mean instantaneous frequencies of different IMFs 
calculated for the 30-s signal samples at different anesthetic 
levels are given in Fig. 3. For these shorter signal sequences, 
9-12 IMFs were extracted by EMD depending on the 
sample. As shown in the figure, the slow wave activity was 
captured mostly by IMF 5. Fig. 4 presents the effect of 
propofol on the amplitude of IMF 5 in patients with good 
and poor neurological outcome. In the patients with good 
neurological outcome (N = 6), the amplitude of IMF 5 
representing the slow wave activity was found to be strongly 
affected by propofol being substantially larger at high 
infusion rates. On the contrary, the patients with poor 
neurological outcome (N = 4) were unable to generate 
propofol-induced slow waves measured by the amplitude of 
IMF 5. Consequently, propofol-induced change in the 
amplitude of IMF 5 was statistically significantly different 
between the two groups at high infusion rates (3 and 4 mg  
kg
–1
  h–1). 
IV. CONCLUSIONS 
As a conclusion, HHT was used to analyze EEG slow 
wave activity during anesthesia in ten post-CA patients. The 
recordings were made in the intensive care unit 36-48 hours 
after the CA in an experiment, during which the propofol 
infusion rate was incrementally decreased to determine the 
drug-induced changes in the EEG at different anesthetic 
levels. HHT was shown to successfully capture the changes 
in the slow wave activity to the behavior of IMFs. While, in 
patients with good neurological outcome defined after a six-
month control period, propofol induced a significant 
increase in the amplitude of IMFs representing the slow 
wave activity, the patients with poor neurological outcome 
were unable to produce such a response. Consequently, the 
proposed method offer substantial prognostic potential by 
providing a novel approach for early estimation of HIE after 
CA. 
REFERENCES 
[1]  A. S. Go, D. Mozaffarian, V. L. Roger et al., "Executive summary: 
heart disease and stroke statistics--2014 update: a report from the 
American Heart Association," Circulation, vol. 129, pp. 399-410, 
2014.  
[2]  M. S. Eisenberg and T. J. Mengert, "Cardiac resuscitation," N. Engl. J. 
Med., vol. 344, pp. 1304-1313, 2001.  
[3]  G. B. Young, "Clinical practice. Neurologic prognosis after cardiac 
arrest," N. Engl. J. Med., vol. 361, pp. 605-611, 2009.  
[4]  C. W. Callaway, M. W. Donnino, E. L. Fink et al., "Part 8: Post-
Cardiac Arrest Care: 2015 American Heart Association Guidelines 
Update for Cardiopulmonary Resuscitation and Emergency 
Cardiovascular Care," Circulation, vol. 132, pp. S465-82, 2015.  
[5]  E. G. Zandbergen, R. J. de Haan, C. P. Stoutenbeek, J. H. Koelman, 
and A. Hijdra, "Systematic review of early prediction of poor outcome 
in anoxic-ischaemic coma," Lancet, vol. 352, pp. 1808-1812, 1998.  
[6]  M. Tiainen, R. O. Roine, V. Pettila, and O. Takkunen, "Serum neuron-
specific enolase and S-100B protein in cardiac arrest patients treated 
with hypothermia," Stroke, vol. 34, pp. 2881-2886, 2003.  
[7]  J. Kortelainen, E. Väyrynen, U. Huuskonen et al., "Propofol-induced 
slow waves – A pharmacological test for brain dysfunction," 
Anesthesiology, submitted for publication.  
[8]  V. Crunelli, F. David, M. L. Lorincz, and S. W. Hughes, "The 
thalamocortical network as a single slow wave-generating unit," Curr. 
Opin. Neurobiol., vol. 31C, pp. 72-80, 2014.  
[9]  P. L. Purdon, E. T. Pierce, E. A. Mukamel et al., 
"Electroencephalogram signatures of loss and recovery of 
consciousness from propofol," Proc. Natl. Acad. Sci. U. S. A., vol. 
110, pp. E1142-51, 2013.  
[10]  Huang NE, Shen Z, Long S et al., "The empirical mode decomposition 
and the Hilbert spectrum for nonlinear and non-stationary time series 
analysis," Proc. R. Soc., vol. 454, 1998.  
[11]  J. Kortelainen and E. Vayrynen, "Assessing EEG slow wave activity 
during anesthesia using Hilbert-Huang Transform," Conf. Proc. IEEE 
Eng. Med. Biol. Soc., vol. 2015, pp. 117-120, 2015. 
1853