*3.1. Results of CPR Process*

All eight experiments were performed successfully. Once VF was triggered, mean arterial pressure (MAP) decreased to almost 0 mmHg, while approximately 20% of baseline MAP remained as residual pressure in the vessel. CBF dropped rapidly to almost 0% of baseline values during untreated VF. When BLS sessions began, hemodynamic parameters started to recover. Recovery rates of hemodynamic parameters over the BLS and ACLS sessions are presented specifically in Supporting File S1.

Sustained ROSC was achieved in five animals. Among them, one animal was defibrillated after the last BLS session. Another four animals were defibrillated during the course of the ACLS sessions. No animals experienced VF again during the monitoring sessions. Three animals were not resuscitated until the tenth ACLS session was completed. BLS cycles were performed a total of 32 times, and ACLS cycles were performed 48 times, and data after those sessions were included for analysis.

## *3.2. EEG Changes with the Recovery of CBF*

The EEG waveforms between an ROSC (Test 6) and a non-ROSC (Test 5) case were compared (Figure 3). Before VF, the amplitude of EEG with irregular morphology exceeding ±20 µV was observed. Since cerebral oxygenation decreased due to VF, the amplitude started to decrease in 10–15 s and almost entered the isoelectric state (±5 µV) at the end of untreated VF.

‐ ‐ μ **Figure 3.** Comparison of EEG over time between return of spontaneous circulation (ROSC) and non-ROSC cases: (**a**) EEG waveforms obtained from animals with successful defibrillation after fourth BLS session and sustained ROSC confirmed after follow-up monitoring for 20 min (Test 6); (**b**) EEG waveforms obtained from animals in which ROSC was not achieved until the end of experiment (Test 5). Dashed lines denote the level of ±5 µV, the limits of the isoelectric state.

‐ The recovery of the EEG was different, depending on the recovery of the CBF. In Test 6, which showed a better recovery, the recovery rate reached almost 40% during the last two BLS sessions. Concurrently, an increased background activity with higher amplitude and increased higher frequency components was observed. EEG activity during the monitoring session appeared similar to the baseline values during the pre-VF period. This means that the cerebral circulation was restored successfully, whereas the recovery rates in Test 5 exceeded 30% during the first BLS session but decreased consistently during the rest of the CPR sessions. The EEG decreased in amplitude and entered the suppression status and increased lower frequency components during the second BLS session. The cerebral resuscitation was poor, with the low CBF recovery rates of below 10%. Nearly flat patterns resulting from electrocerebral inactivity appeared, and EEG did not recover until the end of the ACLS sessions.

‐ Table 2 shows the Pearson correlation coefficients between EEG parameters and the recovery rates of CBF. Among them, time-domain magnitude and two entropy indices, log energy entropy [25] and Rényi entropy [26], showed a correlation coefficient of approximately 0.78. Figure 4 demonstrates the scatter plots for these three parameters.


**Table 2.** Pearson correlation coefficients between EEG parameters and the recovery rates of CBF.

**Figure 4.** Scatter plots between EEG parameters and the recovery of CBF. Correlation coefficients were denoted above the plots: (**a**) magnitude; (**b**) log energy entropy; (**c**) Rényi entropy.

*3.3. Changes in EEG Parameters Depending on Four CBF Groups*

‐ Figure 5 illustrates the results of one-way ANOVA tests for three parameters. For magnitude, the lowest quartile (group 1) showed significant differences to other groups, with *p* < 0.05. However, significant difference was not confirmed among the other three groups. Similar patterns were observed in following two entropy indices. Table 3 demonstrates the results of the post hoc test based on the Dunnett T3 method. ‐

‐ **Figure 5.** Results of one-way ANOVA: (**a**) magnitude; (**b**) log energy entropy; (**c**) Rényi entropy. Asterisk (\*) denotes statistical significance at the *p* < 0.001 level. Error bars indicate the upper and lower extreme values of the data.

‐ **Table 3.** Results of multiple comparisons between groups in three EEG parameters.


Differences were obtained by Group I minus Group II.

− − −

− − −

− − −

− − −

#### *3.4. EEG Parameters Depending on the Di*ff*erent CBF Recovery Groups*

Figure 6 illustrates the ROC curves for the three EEG parameters. All possible cut-off values are plotted with a combination of true positive rate (sensitivity) and false positive rate (1 − specificity). The optimal cut-off points are also denoted. Table 4 presents the results of ROC curve analysis including area under the curve (AUC), true positive rate, false positive rate, and cut-off values. The AUC values of all three parameters were over 0.88. ‐ − ‐ ‐

− − −

− − −

− − −

 ‐ **Figure 6.** Receiver operating characteristic (ROC) curves (blue) for three EEG parameters: (**a**) magnitude; (**b**) log energy entropy; (**c**) Rényi entropy. Red dots indicate the optimal cut-off points, and the diagonal lines (green) indicate random chance.


**Table 4.** Results of the ROC curve analysis for EEG parameters.

Abbreviation AUC: Area under the curve.

#### **4. Discussion**

‐ This study investigated the relationship between the EEG and CBF, to evaluate the feasibility of non-invasive EEG parameters as potential predictors of the recovery of CBF in the CA swine model. The current CPR protocol consists of an ECG rhythm check, chest compression (CC), defibrillation, and medication [22], while CBF or EEG measurement and analysis have rarely been performed during CPR. Monitoring cerebral circulation could provide beneficial information to improve patients' prognosis and quality of life [7,8]. EEG was considered as one of the possible markers because it could reflect the level of cerebral circulation [27]. Post-resuscitation care could be seriously disrupted with a sparsity of EEG activity [28]. If the EEG could reflect the CBF and be measurable in the OHCA setting, CPR with a feedback of non-invasive EEG parameters could guide EMTs to achieve a higher CBF recovery, for example, by guiding leg elevation or the Trendelenburg position [29], which is expected to improve brain perfusion and neurologic outcomes of CA patients after CPR. It is noteworthy that the present study used only single-channel EEG signals from forehead sites where the installation of EEG sensors is convenient.

Several studies have attempted to apply the BIS monitor during CPR. However, unwanted artefacts due to CCs contaminated the original EEG, and generated unreliable outputs [30,31]. The BIS monitor is not adequate to use for the short intervals between CCs, because it is based on the moving-average function over 60 s [24]. Prolonged no- or low-flow periods can deteriorate brain function of CA patients [19,32]. Thus, this study focused on data measured during short pauses between the defibrillation attempts. We observed that the EEG background activity increased and became more irregular with the CBF recovery. The frequency distribution of EEG was also affected. As the CBF recovered, the higher frequency components including alpha (8–13 Hz) and beta (13–30 Hz) increased, whereas the lower frequency components including delta (<4 Hz) and theta (4–8 Hz) decreased. These changes affected the functional dynamics associated with varying amplitudes and multi-frequency responses, including the level of complexity and the amount of energy, diversity, and randomness [33], which could be indicated by the increase of log energy entropy and Rényi entropy, as shown in Figure 4. Entropy parameters have been applied to EEG signals, especially in anesthesia or epileptic seizure studies [34–36]. A previous study analyzing epileptic EEG signals reported that log energy entropy of the modulated EEG signals obtained from the epileptogenic area had relatively lower values [35]. Consistently, another study showed that the complexity derived by Rényi entropy was also higher in healthy signals [36]. These parameters might also have a potential to identify sufficient cerebral circulation for satisfying the metabolic requirements of brain cells of CA patients [27].

This study has several limitations. First, the experimental model was finalized assuming a witnessed OHCA. EEG parameters, such as log energy entropy and Rényi entropy, might be able to reflect the cerebral resuscitation only with very short no- or low-flow duration (<1 min). The association between the CBF and EEG recovery is probably less pronounced with a longer untreated VF. Further research should be performed to validate this method with a longer VF period for at least 5 min. Second, this study was performed only with the limited sample size of eight animals. Feature analysis with larger datasets should be performed to confirm our findings. To generalize our findings to real OHCA patients, moreover, future clinical studies should be guaranteed with the experimental setup optimized for human anatomy.
