Quantitative analysis of relative volume of low apparent diffusion coefficient value can predict neurologic outcome after cardiac arrest

Accepted Manuscript Title: Quantitative Analysis of Relative Volume of Low Apparent Diffusion Coefficient Value Can Predict Neurologic Outcome after Cardiac Arrest Authors: Hyung Ki Moon, Jinhee Jang, Kyu Nam Park, Soo Hyun Kim, Byung Kook Lee, Sang Hoon Oh, Kyung Woon Jeung, Seung Pill Choi, In Soo Cho, Chun Song Youn PII: DOI: Reference:

S0300-9572(18)30088-1 https://doi.org/10.1016/j.resuscitation.2018.02.020 RESUS 7507

To appear in:

Resuscitation

Received date: Revised date: Accepted date:

15-9-2017 14-2-2018 18-2-2018

Please cite this article as: Moon Hyung Ki, Jang Jinhee, Park Kyu Nam, Kim Soo Hyun, Lee Byung Kook, Oh Sang Hoon, Jeung Kyung Woon, Choi Seung Pill, Cho In Soo, Youn Chun Song.Quantitative Analysis of Relative Volume of Low Apparent Diffusion Coefficient Value Can Predict Neurologic Outcome after Cardiac Arrest.Resuscitation https://doi.org/10.1016/j.resuscitation.2018.02.020 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Quantitative Analysis of Relative Volume of Low Apparent Diffusion Coefficient Value Can Predict Neurologic Outcome after Cardiac Arrest

Hyung Ki Moon, M.D.1, Jinhee Jang, M.D.2, Kyu Nam Park, M.D. 1, Soo Hyun Kim, M.D. 1,

IP T

Byung Kook Lee, M.D. 3, Sang Hoon Oh, M.D. 1, Kyung Woon Jeung, M.D. 3, Seung Pill

1

SC R

Choi, M.D. 4, In Soo Cho, M.D. 5, Chun Song Youn, M.D.1*

Department of Emergency Medicine, Seoul St. Mary Hospital, College of Medicine, College

Department of Radiology, Seoul St. Mary Hospital, College of Medicine, The Catholic

A

2

N

U

of Medicine, The Catholic University of Korea, Seoul 137-701, South Korea

Department of Emergency Medicine, Chonnam National University Medical School,

Gwangju, Korea

Department of Emergency Medicine, Yeouido St. Mary’s Hospital, College of Medicine, The

PT

4

ED

3

M

University of Korea, Seoul 137-701, South Korea

Catholic University of Korea, 222 Banpo-daero, Seocho-gu, Seoul 137-701, Republic of

CC E

Korea 5

Department of Emergency Medicine, Hanil General Hospital, Korea Electric Power Medical

A

Corporation, Seoul, Korea

*Corresponding author: Chun Song Youn, M.D., Associate Professor, Department of Emergency Medicine, Seoul St. Mary’s Hospital, College of Medicine, The Catholic

University of Korea, 222, Banpo-daero, Seocho-Gu, Seoul, 06591, Republic of Korea. Tel: 82-2-2258-1988 Fax: 82-2-2258-1997 E-mail: [email protected]

IP T

Word count: 3504

SC R

Introduction: Predicting neurologic outcomes after cardiac arrest (CA) is challenging. This study tested the hypothesis that a quantitative analysis of diffusion weighted imaging (DWI)

U

using the FMRIB Software Library (FSL) can predict neurologic outcomes after CA and can

N

clarify the optimal apparent diffusion coefficient (ADC) thresholds for predicting poor

A

neurologic outcomes.

M

Methods: Out-of-hospital CA patients treated with targeted temperature management

ED

(TTM) who underwent DWI were included in this study. Voxel-based analysis was performed to calculate the mean ADC value. ADC thresholds (750, 700, 650, 600, 550, 500, 450 and

PT

400) and brain volumes below each threshold were also analyzed for their correlation with outcomes. The patients were divided into early (within 48 h after return of spontaneous

CC E

circulation (ROSC)) and late group (between 48 h and 7 days after ROSC) according to the DWI scan time. The primary outcome was a poor neurologic outcome at 6 months after CA,

A

defined as a cerebral performance category (CPC) of 3-5. Results: One hundred ten DWIs were analyzed. The mean ADC values were 789.0 (761.5-

826.5) ×10-6mm2/s for the good neurologic outcome group and 715.2 (663.1-778.4) ×106

mm2/s for the poor neurologic outcome group (p<0.001). All the ADC thresholds could

differentiate patients with good versus poor outcomes. The ADC threshold of 400×10-6mm2/s

had the highest odds ratio (4.648 in the early group and 11.283 in the late group) after adjusting for initial rhythm and anoxic time. To achieve 100% specificity using an ADC threshold of 400×10-6mm2/s, the sensitivity was 64% (cutoff value; > 2.5% ADC threshold of 400×10-6mm2/s) in the early group and 79.2% (cutoff value; > 1.66% ADC threshold of

IP T

400×10-6mm2/s) in the late group. Conclusions: Voxel-based analysis using FSL software can predict neurologic outcomes after

SC R

CA. The ADC threshold of 400×10-6mm2/s had the highest OR for predicting a poor

U

neurologic outcome.

M

A

N

Keywords: cardiac arrest; hypothermia, induced; magnetic resonance imaging; prognosis

ED

1. Introduction

Cardiac arrest (CA) is an extreme medical emergency with high mortality and significant

PT

morbidity among survivors [1]. Despite recent advances in critical care and targeted temperature management (TTM), the survival rate after cardiac arrest is approximately 10%,

CC E

and post–cardiac arrest brain injury is a major cause of disability and death [1-5]. The most common cause of death after cardiac arrest is the withdrawal of life-sustaining treatment (WLST) for perceived poor neurological prognosis [6-8]. Moreover, the use of sedative drugs

A

and paralytic agents during TTM could interfere with accurate prognostication [9]. Therefore, approaches that minimize erroneous prognostication after CA are critically needed. There are many advances in the prediction of prognosis after CA in terms of timing of neuro-prognostication and combining of variable modalities to help improve prognostic accuracy [10]. However, the optimal prognostic tool for use after CA remains unknown.

Several modalities for predicting neurologic outcomes after cardiac arrest have been investigated, including clinical examination, neuron-specific enolase (NSE), somatosensory evoked potential (SSEP), electroencephalography (EEG), brain computed tomography and diffusion weighted imaging (DWI) [11-16]. Among these, DWI detected early ischemic brain

IP T

injury objectively, and apparent diffusion coefficient (ADC) values provided valuable information regarding the severity and extent of brain injury [17-20]. ADC is a measure of

SC R

the magnitude of diffusion of water molecules within tissue and is calculated using magnetic resonance imaging (MRI) with DWI [21].

DWI can be analyzed with quantitative or qualitative techniques. Qualitative analysis has

U

the disadvantage of providing subjective information, and one multicenter retrospective study

N

showed a relatively high false-positive rate (FPR) of 14% for poor neurologic outcomes [22].

A

A previously reported quantitative analysis of DWI used a specific program that was not

M

available to treating clinicians [17-18]. This method was an obstacle that prevented treating clinicians from using DWI in their clinical practice to predict neurologic outcomes after CA.

ED

Even though treating clinicians may have difficulty using the FMRIB Software Library (FSL)

PT

because it requires several steps to analyze MRI, nevertheless FSL is a comprehensive library of analysis tools for functional, structural and diffusion magnetic resonance imaging (MRI)

CC E

data [23]. Although FSL is not approved by FDA for clinical use, it has been used widely in the research field of neuroimaging. No studies have applied FSL for the prediction of outcomes after cardiac arrest [24].

A

This study tested the hypothesis that quantitative analysis of DWI using FSL software could

predict neurologic outcomes after CA and could clarify the optimal ADC thresholds for predicting poor neurologic outcome. Voxel-based analysis was performed in patients treated with TTM after CA. These measurements were related to neurologic outcome at 6 months after CA.

2. Materials and methods 2.1. Patients and settings The current study was based on data prospectively collected from a cardiac arrest registry at a single tertiary educational hospital in Seoul, Korea, between January 2009 and December

IP T

2014. The following inclusion criteria were used: ≥18 y of age; treatment with TTM after

successful resuscitation from CA; and DWI scan for neurological prognostication. The

SC R

exclusion criteria included the following: cardiac arrest due to trauma, drug intoxication or intracranial hemorrhage; previous history of neurologic disease; DWI with absence of

U

diffusion protocol; DWI with poor image quality; and DWI performed more than 7 days after

N

ROSC. If a patient underwent MRI twice or more, each imaging session was considered an

A

independent case that needed to satisfy exclusion criteria.

M

The patients were divided into two groups according to the timing of the DWI scan. The

ED

early group was defined by DWI scans performed within 48 h after ROSC, and the late group was defined by DWI scans performed between 48 h and 7 days after ROSC.

PT

This study was approved by the Review Board of Seoul St. Mary’s Hospital, and consent

CC E

was waived because of the retrospective nature of this study.

A

2.2. Post-cardiac arrest care protocol All comatose patients who were resuscitated from CA were eligible for TTM, according to

our previously established post-cardiac arrest care protocol [25]. TTM was started in the emergency department using an endovascular cooling device (CoolGard Thermal Regulation System, Alsius Corporation, Irvine, CA, USA) or ArticSun (Bard Medical, Louisville, CO,

USA) to obtain a target temperature of 33°C. The target temperature was maintained for at least 24 h. After 24 h at 33°C, the patients were slowly rewarmed to 37°C at a rate of 0.25°C/h. The patients were sedated using fentanyl and midazolam and were paralyzed using rocuronium during the induction and maintenance periods, if necessary. We encouraged

IP T

maintaining arterial blood levels as follows: SaO2 94–96%, and PaCO2 35–45 mmHg. We encouraged infusing fluids and using vasopressors or inotropes to achieve a mean arterial

SC R

pressure ≥70 mmHg and a urine output ≥ 0.5 mL/kg/h. In September 2010, our hospital added single-channel amplitude-integrated electroencephalography (aEEG) recordings for the

first 72 h after resuscitation from cardiac arrest as standard monitoring for all comatose post-

U

cardiac arrest patients [14]. If there was evidence of electrographic seizure or a clinical

2.3. Outcome measure

ED

M

A

levetiracetam or phenobarbital were administered.

N

diagnosis of seizure, anti-epileptic medications such as valproic acid, clonazepam,

PT

The primary outcome of this study was neurologic outcome at 6 months after CA, with a Cerebral Performance Category of 3 to 5 used to define poor neurologic outcome [26].

CC E

Neurologic outcome was assessed via telephone interview. The five categories of the CPC are CPC 1, conscious and alert with good cerebral performance; CPC 2, conscious and alert with moderate cerebral disability; CPC 3, conscious with severe cerebral disability; CPC 4,

A

comatose or in a persistent vegetative state; and CPC 5, brain dead. The patients were divided into two groups according to the neurologic status at 6 months after CA (the good outcome group vs. the poor outcome group).

2.4. MRI protocol and analysis All cases with appropriate brain DWI MRIs obtained within 7 days after the ROSC were analyzed. Three MRI machines were used in our hospital. Approximately 25 contiguous DWI sections per patient were acquired using either a 1.5T (Signa HD, GE Medical Systems,

IP T

Milwaukee, WI or Achieva, Philips Healthcare, Best, Netherlands) or 3.0 scanner (Magnetom Verio, Siemens Medical Solutions, Erlargen, Germany) according to clinical circumstances,

SC R

due to lack of standardized MRI protocol. The imaging protocols varied for each case and

ADC value could be influenced. Therefore, all included MRIs were performed on a 3.0T MR unit with a 16-channel head and neck coil. Acquisition parameters are summarized in

N

U

supplementary Table A.

A

Briefly, the standard of b=1000 s/mm2 was used for all DWIs. Apparent diffusion

M

coefficient (ADC) maps were created from the monoexponential calculation of DWI with a commercial software and workstation system (Leonardo MR Workplace; Siemens Medical

ED

Solutions, Erlangen, Germany). The images were processed and analyzed on a PC using a program (FMRIB Software Library, Release 5.0 (c) 2012, The University of Oxford) that can

PT

extract brain tissue from the skull and optic structures, and can measure many parameters,

CC E

including the mean absolute diffusion coefficient (ADC) of the extracted brain image or the absolute volume of voxels that have a pre-defined ADC range [Figure 1]. The images were retrieved in DICOM (Digital Imaging and Communications in Medicine) format from picture

A

archiving and communication system (PACS) servers at the hospital and were converted to NITFI format using the program MRIcron (http://www.nitrc.org/projects/mricron). Voxels with ADC values under 50−6mm2/sec or above 1200−6mm2/sec were extracted from the analysis to exclude artifacts or cerebrospinal fluid. The ADC thresholds ranged from 400 to 750×10−6mm2/sec and were analyzed at 50×10−6mm2/sec intervals. The % voxels of low

ADC value (PV), meaning the percentage of voxels below different ADC thresholds, was calculated. The MRI analyses were performed by two independent researchers who were not familiar with the patients.

IP T

2.5. Statistical methods

SC R

The differences between the outcome groups were assessed using the t-test or Wilcoxon rank sum test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables. Summary statistics are presented as n (%) for categorical variables and

U

as the mean ± standard deviation (SD) or median (IQR) for continuous variables. Our main

N

exposures of interest were PVs and mean ADC. Univariate logistic regression analysis was

A

performed to select covariates [Supplementary Table 2]. Among variables with p value less

M

than 0.1 in the univariate logistic regression analysis, the variables with smallest p value and

ED

clinical importance were finally chosen as covariates. The shockable rhythm and time from collapse to ROSC were selected as covariates. We used multivariate logistic regression to

PT

adjust for the effect of demographic factors and MR diffusion on prognosis. The results of the multivariate logistic regression are reported as odds ratios (ORs) and 95% CIs. We estimated

CC E

receiver operating characteristic (ROC) curves to calculate the area under the curve (AUC). All results were analyzed using SAS Version 9.4 (SAS Institute, Cary, NC, USA). A p-value

A

of less than 0.05 was considered statistically significant.

3. Results 3.1. Patient demographics

Two hundred twenty-seven patients were treated with TTM after cardiac arrest during the study period. Of these, 142 underwent DWI after CA. Thirty-seven patients underwent DWI twice; consequently, 179 brain DWIs were acquired. Of these, 52 DWIs were excluded due to the absence of diffusion protocol, being performed more than 7 days after CA or having

IP T

unacceptable image quality. In addition, 17 DWIs were excluded because the DWI was not obtained using a 3.0T scanner. Finally, a total of 110 DWIs from 96 patients were included in

SC R

this study.

Thirty-one patients had a good outcome at 6 months after CA. The respective mean ages of the good and poor outcome groups were 47 ± 14 years and 54 ± 16 years. Other clinical

U

parameters are shown in Table 1. Table 2 shows the demographic characteristics of subjects

N

according to DWI scan time. The mean times from ROSC to DWI scan were 17 (SD 14) h for

A

the early group and 77 (SD 23) h for the late group. In our study, 14 cases of repeated MRI

ED

M

were registered. Subgroup analysis of these patients is presented in supplementary Table 3.

PT

3.2. Quantitative analysis of % volume of low ADC value Voxel-based analysis using FSL was performed to calculate the mean ADC value and PV (%

CC E

voxels with ADC values below the predefined ADC thresholds) ranging from PV400 to PV750 at 50×10−6mm2/sec intervals.

A

3.2.1. Early group

Forty-four DWIs were analyzed in the early group (DWIs performed within 48 h after CA).

The mean ADC value was significantly different between the good and poor outcome groups (767.3 10−6mm2/sec, IQR 753-785 10−6mm2/sec vs 749 10−6mm2/sec, IQR 685.4-770.9, p=0.026). PV750 did not significantly differ between the outcome groups, but PVs between

700 and 400 were significantly different between the outcome groups [Table 3]. 3.2.2. Late group Sixty-six DWIs were analyzed in the late group (DWIs performed between 48 h and 7 days after CA). The mean ADC value was significantly different between the good and poor

IP T

outcome groups (822.2 10−6mm2/sec, IQR 801.7-848 10−6mm2/sec vs 691.9 10−6mm2/sec, IQR 657.9-785.4, p<0.001). All PVs between 750 and 400 were significantly different

SC R

between the outcome groups [Table 3].

N

U

3.3. Logistic regression analysis with % volume of low ADC value

A

3.3.1. Early group

M

The mean ADC and PVs were adjusted for shockable rhythm and time from collapse to

ED

ROSC. The mean ADC was not associated with a poor neurologic outcome at 6 months after CA, according to multivariate logistic regression analysis (adjusted OR (aOR) 0.989, 95% CI

PT

0.968-1.010). PV750, 700 and 650 were also not associated with poor neurologic outcomes. The PVs between 600 and 400 were associated with poor neurologic outcomes, and PV400

CC E

had the highest OR (aOR 4.648, 95% CI 1.366-15.814) [Table 4].

A

3.3.2. Late group

The mean ADC was associated with poor neurologic outcomes at 6 months after CA when adjusted for shockable rhythm and time from collapse to ROSC (aOR 0.979 95% CI 0.9650.993). In addition, all PVs were associated with poor neurologic outcomes. PV400 had the highest OR for poor neurologic outcomes at 6 months after CA (aOR 11.283, 95% CI 1.136-

112.037) [Table 4].

3.4. Sensitivity analysis with the proportional volume of different cutoff ADC values

IP T

3.4.1. Early group A sensitivity analysis was performed using two different cutoff values for predicting poor

SC R

neurologic outcomes based on a maximal Youden’s index and cutoff values that achieved

100% specificity. The area under the curve (AUC) was also calculated. PV500 had the highest AUC (0.907, 95% CI 0.781-0.974) for predicting poor neurologic outcomes at 6

U

months after CA. The PV500 of 4.59% had 84% (95% CI 63.9-95.5) sensitivity and 89.5%

3.4.2. Late group

ED

M

A

72% (95% CI 50.6-87.9) sensitivity [Table 5].

N

(95% CI 63.9-95.5) specificity. The PV500 of 6.25%, which achieved 100% specificity, had

PT

The AUC of the mean ADC was 0.823 (95% CI 0.709-0.906), and a mean ADC below 778

CC E

had 75% (95% CI 60.4-86.4) sensitivity and 94.4% (95% CI 72.7-99.9) specificity. PV400 had the highest AUC (0.906, 95% CI 0.809-0.964) for predicting poor neurologic outcomes at 6 months after CA. PV400 of 1.66% had 79.2% (95% CI 65-89.5) sensitivity and 100% (95%

A

CI 81.5-100) specificity [Table 5].

3.5 ADC values of repeatedly measured patients Fourteen patients underwent MRI twice. Six had good neurologic outcome and 8 had poor

outcome. The mean ADC value was similar between the good and poor outcome groups when performed within 48 h after ROSC (Early group, 756 in good outcome and 760 in poor outcome, p value of 0.867). In addition, the mean ADC value was higher in the good outcome group than that of poor outcome group when performed after 48 h after ROSC but without

IP T

statistical significance (Late group, 827 with good outcome and 795 with poor outcome, p value of 0.127). The PV 400 was not significantly different between the good and poor

SC R

outcome groups when performed within 48 h after ROSC (Early group, 1.16% in the good

outcome group and 2.04 in the poor outcome group, p value of 0.072). However, the PV 400 was significantly lower in the good outcome group than in it was in the poor outcome group

U

when performed after 48 h after ROSC (Late group, 0.5% for good outcome and 2.23% for

M

A

N

poor outcome, p value of 0.019) [Figure 2, Supplementary Table 3].

ED

Discussion

The main findings of this study were that 1) voxel-based analysis using FSL software can

PT

predict neurologic outcomes after CA and 2) an ADC threshold of 400×10-6mm2/s had the highest aOR for predicting neurologic outcomes in both the early and late groups. To achieve

CC E

100% specificity using an ADC threshold of 400×10-6mm2/s, the sensitivity was 64% (cutoff value; > 2.5% ADC threshold of 400×10-6mm2/s) in the early group and 79.2% (cutoff value;

A

> 1.66% ADC threshold of 400×10-6mm2/s) in the late group. This study has several strengths. We only enrolled patients who were treated with TTM,

currently considered the standard care after CA. Although the treating physician was not blinded to the DWIs, the withdrawal of life-sustaining treatment, a major cause of selffulfilling prophecy, is legally prohibited in Korea. Finally, the ADC thresholds were adjusted

for demographic factors to determine the best ADC threshold for predicting poor neurologic outcomes after CA. Both quantitative and qualitative analysis have been used to predict neurologic outcome after CA. However, there is no universally accepted standard DWI analysis technique.

IP T

Recently, Hirsch et al. developed a qualitative brain MRI scoring system based on 21 predefined brain regions observed on fluid-attenuated inversion recovery (FLAIR), and DWI

SC R

sequences to identify the severity of the abnormality and to provide prognostic information for PCAS patients [27]. However, this score has not been validated, and its use in real clinical

practice is neither easy nor objective. Ryoo et al. reported that positive diffusion-weighted

U

image findings, a simpler qualitative technique using large multicenter registry data, were

N

associated with poor neurologic outcomes after CA [22]. However, the FPR of this technique

A

was 14%. Moreover, previously reported quantitative analyses require a specific program or

M

technician that is not available to primary physician in real clinical practice [17, 18].

ED

Therefore, we hypothesized that FSL software could predict neurologic outcomes after CA because FSL software can provide more objective information regarding the extent of brain

PT

injury than qualitative analysis.

CC E

A previous study by Wijman et al. found that the percentage of brain volume with an ADC value less than 400 to 450 ×10-6mm2/s best differentiated between Glasgow Outcome Scores of 4 or 5 and 3 [17]. Additionally, they determined that the ideal window for prognostication

A

using DWI was between 49 h and 108 h after CA. This result is consistent with the results of our study. In a multicenter quantitative analysis study, Hirsch et al. hypothesized that an ADC value of less than 650×10-6mm2/s > 10% of brain volume could predict neurologic outcomes after CA [28]. An ADC value of less than 650×10-6mm2/s > 10% of brain volume had 91% specificity and 72% sensitivity for predicting a poor outcome. In addition, an ADC value of

less than 650×10-6mm2/s > 22% of brain volume was needed to achieve 100% specificity for a poor outcome. However, this result was not repeated in our study. In our study, PV 650 was less predictive than PV 400. Moreover, an ADC value of less than 650×10-6mm2/s > 61.1% of brain volume was needed to achieve 100% specificity for predicting a poor neurologic

IP T

outcome in the pooled cohort and the late group. Therefore, the optimal ADC thresholds for predicting a poor neurologic outcome after CA should be further investigated. Kim et al.

SC R

proposed a new method of voxel-based analysis for predicting neurologic outcome after CA [20]. They suggested that the relative volume of the most dominant cluster of low-ADC voxels was well-correlated with neurologic outcome.

U

The ideal time window of DWI for predicting neurologic outcomes after CA should be

N

further tested. Ischemic injury becomes clearer over time on DWI [16]. The guidelines

A

recommend obtaining brain MRIs 2 days after ROSC [10, 29]. However, there is also a report

M

suggesting that very early DWI, even prior to TTM, is associated with outcomes after CA

ED

[30]. In our study, 44 DWIs were obtained within 48 h after CA. PV 400 had the highest OR for predicting a poor neurologic outcome after CA, and an ADC value of less than 400×10mm2/s > 2.5% of brain volume had 100% specificity and 64% sensitivity. This does not

PT

6

mean that early DWI, especially before 48 h after CA, does not help to predict outcomes.

CC E

However, late MRI had higher aOR, higher AUC and higher sensitivity for predicting neurologic outcome than did early MRI. Therefore, our findings support current guidelines

A

for performing MRI after 48 h after CA. Fourteen patients performed MRI twice in our study. The changing pattern of mean ADC and PV 400 were different between good and poor outcome groups. However, the clinical value of repeated DWI should be further tested. One major concern of quantitative analysis of brain DWI is inter-imager ADC variability. Sasaki et al. reported that ADC values significantly varied, depending on coil systems,

imagers, vendors, and field strengths used for MR imaging among healthy volunteers [31]. The average inter-imager differences ranged from 29×10-6mm2/s to 56×10-6mm2/s. Tsujita et al. also reported that inter-imager differences in ADC values were more remarkable in gray matter [32]. However, the differences were less obvious when performed with 3.0T imagers.

IP T

The relative ADC values may be more appropriate than absolute ADC values for predicting neurologic outcome after CA. Therefore, we included only DWIs performed by 3.0T MRI

SC R

with the same diffusion protocol to minimize inter-imager bias.

This study has several limitations. First, this is a single-center retrospective study, and some patients did not undergo DWI. This could indicate selection bias. Therefore, the results

U

of this study may not be generalizable to all cardiac arrest patients. Second, of the 179 DWIs,

N

69 were excluded from the final analysis for various reasons. This could also indicate

A

selection bias. Third, as described in introduction, FSL is not an authorized program (e.g.,

M

approved by the FDA) for clinical use, although there many studies in the neuroimaging

ED

analysis field. Therefore, if available, it would be desirable to have a clinically FDAapproved software for DWI quantification. Finally, there may be some inter-machine or inter-

PT

protocol (MRI or FSL program) bias. We did not include patients who underwent scanning with 1.5T MRI. When scanning with 1.5T MRI, the results may differ from that of our study.

CC E

Finally, we did not perform anatomical analyses. For example, the cortex and deep gray nuclei may show different results. Because we only took scalar quantity data from voxels, a

A

loss of location information occurred that could weaken the predictive power of our results. However, at present, anatomical structure discrimination by computer programs without human assistance is not satisfactory. Nonetheless, the results of our study suggest that the objective and quantitative aspects of DWI analysis can have predictive value and could be complemented by further development of the software. Future prospective study is needed to

confirm our results. In conclusion, voxel-based quantitative analysis using FSL software can predict neurologic outcomes after CA. The ADC threshold of 400×10-6mm2/s had the highest OR for predicting

IP T

poor neurologic outcomes in both the early and late groups.

SC R

Study funding: Dr. Hyung Ki Moon: No targeted funding reported. Dr. Jinhee Jang: No targeted funding reported.

N

Dr. Soo Hyun Kim: No targeted funding reported.

U

Dr. Kyu Nam Park: No targeted funding reported.

A

Dr. Byung Kook Lee: No targeted funding reported.

M

Dr. Sang Hoon Oh: No targeted funding reported.

ED

Dr. Kyung Woon Jeung: No targeted funding reported. Dr. Seung Pill Choi: No targeted funding reported.

PT

Dr. In Soo Cho: No targeted funding reported.

CC E

Dr. Chun Song Youn: No targeted funding reported.

Conflict of interest:

A

Dr. Hyung Ki Moon: reports no disclosures. Dr. Jinhee Jang: reports no disclosures. Dr. Kyu Nam Park: reports no disclosures. Dr. Soo Hyun Kim: reports no disclosures. Dr. Byung Kook Lee: reports no disclosures.

Dr. Sang Hoon Oh: reports no disclosures. Dr. Kyung Woon Jeung: reports no disclosures. Dr. Seung Pill Choi: reports no disclosures. Dr. In Soo Cho: reports no disclosures.

A

CC E

PT

ED

M

A

N

U

SC R

IP T

Dr. Chun Song Youn: reports no disclosures.

References 1. Mozaffarian D, Benjamin EJ, Go AS, et al.; American Heart Association Statistics Committee; Stroke Statistics Subcommittee. Heart Disease and Stroke Statistics-2016 Update: A Report From the American Heart Association. Circulation. 2016;133:e38-360.

IP T

2. Lee BK, Park KN, Kang GH, et al.; Korean Hypothermia Network Investigators. Outcome

and current status of therapeutic hypothermia after out-of-hospital cardiac arrest in Korea

SC R

using data from the Korea Hypothermia Network registry. Clin Exp Emerg Med. 2014;1:1927.

U

3. Hypothermia after Cardiac Arrest Study Group. Mild therapeutic hypothermia to improve

N

the neurologic outcome after cardiac arrest. N Engl J Med 2002;346:549–56.

A

4. Bernard SA, Gray TW, Buist MD, et al. Treatment of comatose survivors of out-of-hospital

M

cardiac arrest with induced hypothermia. N Engl J Med 2002;346:557–63.

ED

5. Nielsen N, Wetterslev J, Cronberg T, et al., TTM Trial Investigators. Targeted temperature

PT

management at 33◦C versus 36◦C after cardiac arrest. N Engl J Med 2013;369:2197–206. 6. Laver S, Farrow C, Turner D, et al. Mode of death after admission to an intensive care unit

CC E

following cardiac arrest. Intensive Care Med 2004;30:2126–8. 7. Mulder M, Gibbs HG, Smith SW, et al. Awakening and withdrawal of life-sustaining

A

treatment in cardiac arrest survivors treated with therapeutic hypothermia. Crit Care Med 2014;42:2493–9. 8. Elmer J, Torres C, Aufderheide TP, et al. Association of early withdrawal of life-sustaining therapy for perceived neurological prognosis with mortality after cardiac arrest. Resuscitation 2016;102:127–35.

9. Samaniego EA, Mlynash M, Caulfield AF, Eyngorn I, Wijman CA. Sedation confounds outcome prediction in cardiac arrest survivors treated with hypothermia. Neurocrit Care. 2011;15:113-9. 10. Callaway CW, Donnino MW, Fink EL, et al. Part 8: Post-Cardiac Arrest Care: 2015

Emergency Cardiovascular Care. Circulation. 2015;132:S465-82.

IP T

American Heart Association Guidelines Update for Cardiopulmonary Resuscitation and

SC R

11. Bouwes A, Binnekade JM, Kuiper MA, et al. Prognosis of coma after therapeutic hypothermia: a prospective cohort study. Ann Neurol 2012;71:206–12.

U

12. Stammet P, Collignon O, Hassager C, et al.; TTM-Trial Investigators. Neuron-Specific

N

Enolase as a Predictor of Death or Poor Neurological Outcome After Out-of-Hospital Cardiac

A

Arrest and Targeted Temperature Management at 33°C and 36°C. J Am Coll Cardiol.

M

2015;65:2104-14.

ED

13. Choi SP, Park KN, Wee JH, et al. Can somatosensory and visual evoked potentials predict neurological outcome during targeted temperature management in post cardiac arrest

Oh

SH,

Park

KN,

Shon

YM,

et

al.

Continuous

amplitude-Integrated

CC E

14.

PT

patients? Resuscitation. 2017;119:70-5.

electroencephalographic monitoring is a useful prognostic tool for hypothermia-treated cardiac arrest patients. Circulation. 2015 Sep 22;132(12):1094-103.

A

15. Kim SH, Choi SP, Park KN, Youn CS, Oh SH, Choi SM. Early brain computed tomography findings are associated with outcome in patients treated with therapeutic hypothermia after out-of-hospital cardiac arrest. Scand J Trauma Resusc Emerg Med 2013;21:57.

16. Youn CS, Park KN, Kim JY, et al. Repeated diffusion weighted imaging in comatose cardiac arrest patients with therapeutic hypothermia. Resuscitation2015;96:1–8. 17. Wijman CA, Mlynash M, Caulfield AF, et al. Prognostic value of brain diffusionweighted imaging after cardiac arrest. Ann Neurol 2009;65:394–402.

IP T

18. Wu O, Sorensen AG, Benner T, Singhal AB, Furie KL, Greer DM. Comatose patients with cardiac arrest: predicting clinical outcome with diffusion-weighted MR imaging. Radiology

SC R

2009;252:173–81.

19. Choi SP, Park KN, Park HK, et al. Diffusion-weighted magnetic resonance imaging for

U

predicting the clinical outcome of comatose survivors after cardiac arrest: a cohort study. Crit

N

Care 2010;14:R17.

A

20. Kim J, Kim K, Suh GJ, et al. Prognostication of cardiac arrest survivors using low

M

apparent diffusion coefficient cluster volume. Resuscitation 2016;100:18–24.

ED

21. Sener RN. Diffusion MRI: apparent diffusion coefficient (ADC) values in the normal brain and a classification of brain disorders based on ADC values. Comput Med Imaging

PT

Graph. 2001;25: 299-326.

CC E

22. Ryoo SM, Jeon SB, Sohn CH, et al.; Korean Hypothermia Network Investigators. Predicting outcome with diffusion-weighted imaging in cardiac arrest patients receiving

A

hypothermia therapy: Multicenter retrospective cohort study. Crit Care Med. 2015;43:2370-7. 23. Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM. FSL. Neuroimage. 2012;62:782-90. 24. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage. 2004;23:S208-19.

25. Youn CS, Kim SH, Oh SH, et al: Successful implementation of comprehensive packages of post cardiac arrest care after out-of-hospital cardiac arrest: a single institution experience in South Korea. Ther Hypothermia Temp Manage 2013;3:17–23. 26. Rittenberger JC, Raina K, Holm MB, Kim YJ, Callaway CW. Association between

IP T

cerebral performance category, modified ranking scale, and discharge disposition after cardiac arrest. Resuscitation 2011;82:1036–40.

SC R

27. Hirsch KG, Mlynash M, Jansen S, et al. Prognostic value of a qualitative brain MRI scoring system after cardiac arrest. J Neuroimaging. 2015;25:430-7.

U

28. Hirsch KG, Mlynash M, Eyngorn I, et al. Multi-Center Study of Diffusion-Weighted

N

Imaging in Coma After Cardiac Arrest. Neurocrit Care. 2016;24:82-9.

A

29. Nolan JP, Soar J, Cariou A, et al.; European Resuscitation Council; European Society of

M

Intensive Care Medicine. European Resuscitation Council and European Society of Intensive

2015;41:2039-56.

ED

Care Medicine 2015 guidelines for post-resuscitation care. Intensive Care Med.

PT

30. Park JS, Lee SW, Kim H, et al. Efficacy of diffusion-weighted magnetic resonance

CC E

imaging performed before therapeutic hypothermia in predicting clinical outcome in comatose cardiopulmonary arrest survivors. Resuscitation. 2015;88:132-7. 31. Sasaki M, Yamada K , Watanabe Y, et al.; Acute Stroke Imaging Standardization Group-

A

Japan (ASIST-Japan) Investigators. Variability in absolute apparent diffusion coefficient values across different platforms may be substantial: a multivendor, multi-institutional comparison study. Radiology. 2008;249:624-30.

32. Tsujita N, Kai N, Fujita Y, et al. Interimager variability in ADC measurement of the

A

CC E

PT

ED

M

A

N

U

SC R

IP T

human brain. Magn Reson Med Sci. 2014;13:81-7.

Figure legends Figure 1. Brain DWI color ADC maps obtained from comatose patients. Each case shows different patterns of hypoxic-ischemic brain injury. (A) CPC 1 at 6 months after CA: mean ADC value of 827 10−6mm2/sec, PV400 of 0.22%, DWI scan time of 59 h from ROSC. (B)

IP T

CPC 4 at 6 months after CA: mean ADC value of 549 10−6mm2/sec, PV400 of 28.09%, DWI scan time of 57 h from ROSC. (C) CPC 5 at 6 months after CA: mean ADC value of 785

SC R

10−6mm2/sec, PV400 of 3.12%, DWI scan time of 7 h from ROSC (D) CPC 5 at 6 months

after CA: mean ADC value of 679 10−6mm2/sec, PV400 of 9.18%, 71 h from ROSC. (C) and (D) Images acquired from the same patient. All images were obtained from the FSL eyes

A

CC E

PT

ED

M

A

N

U

program.

Figure 2. Comparison of mean ADC and PV400 between good and poor outcome group of

A

CC E

PT

ED

M

A

N

U

SC R

IP T

repeatedly measured patients.

Table 1. Demographic characteristics of the subjects according to neurologic outcomes at 6 months after cardiac arrest Good

Poor

N=96

N=31

N=65

Age, mean, yr (SD)

52 (16)

47 (14)

54 (16)

Male, N. (%)

66 (68)

23 (74)

43 (66)

HTN, N. (%)

25 (26)

6 (19)

19 (30)

DM, N. (%)

16 (17)

2 (7)

Witnessed arrest, N. (%)

69 (73)

26 (84)

Bystander CPR, N. (%)

57 (60)

22 (71)

Shockable rhythm, N. (%)

30 (32)

Cardiac cause of arrest, N. (%)

62 (64)

0.023 0.427

0.303

SC R

History

P

IP T

Total

0.064

44 (68)

0.095

35 (55)

0.110

18 (58)

12 (19)

< 0.001

27 (87)

35 (54)

0.001

A

N

U

14 (22)

21 (14)

37 (19)

< 0.001

LOS, median (IQR), day

16 (9-28)

14 (5-27)

0.134

16 (7-28)

ED

Abbreviation;

M

Time from collapse to ROSC, 31 (19) mean, min (SD)

SD=standard

deviation;

HTN=hypertension;

DM=diabetes

mellitus;

PT

CPR=cardiopulmonary resuscitation; ROSC=return of spontaneous circulation; LOS=length

A

CC E

of stay; IQR=interquartile range

Table 2. Demographic characteristics of the subjects according to DWI scan time Early

Late

N=110

N=44

N=66

Age, mean, yr (SD)

51 (16)

50 (15)

66 (16)

0.495

Male, N. (%)

78 (71)

34 (77)

44 (67)

0.230

HTN, N. (%)

29 (26)

12 (27)

17 (26)

DM, N. (%)

18 (16)

8 (18)

10 (15)

Witnessed arrest, N. (%)

78 (71)

32 (73)

Bystander CPR, N. (%)

62 (56)

18 (41)

Shockable rhythm, N. (%)

34 (31)

14 (32)

Cardiac cause of arrest, N. (%)

72 (66)

0.860 0.674 0.732

44 (67)

0.008

20 (30)

0.866

32 (73)

40 (61)

0.190

28 (18)

31 (19)

0.562

16 (6-25)

17 (7-38)

0.133

37 (34)

19 (43)

18 (27)

0.084

73 (66)

25 (57)

48 (73)

Time from ROSC to DWI scan, hrs 53 (36) (SD)

17 (14)

77 (23)

U

46 (70)

N

SC R

History

P

IP T

Total

M

A

Time from collapse to ROSC, 30 (19) mean, min (SD) LOS, median (IQR), day

16 (7-28)

ED

6-month prognosis, N. (%)

CC E

Poor

PT

Good

Abbreviation;

SD=standard

deviation;

HTN=hypertension;

<0.001

DM=diabetes

mellitus;

CPR=cardiopulmonary resuscitation; ROSC=return of spontaneous circulation; LOS=length

A

of stay; IQR=interquartile range

Table 3. Comparison of median [IQR] DWI parameters between patients with good and poor outcomes dichotomized by whether DWI was performed early or late

Late

N= 110

N= 44

N= 66

Poor

p

Good

Poor

p

Good

Poor

Mean ADC

789.0

715.2

767.3

749.0

691.9

(663.1778.4)

(753785)

(685.4770.9)

0.0 26

822.2

(761.5826.5)

<0. 001

(801.7848.0)

(657.9785.4)

PV 750

44.8

58.0

51.5

55.1

61.6

(45.669.6)

(47.156.5)

(49.469.4)

0.1 10

31.8

(30.152.5)

<0. 001

(25.338.6)

(41.269.7)

27.3

47.8

34.1

41.2

51.1

(32.160.4)

(29.740.7)

(35.059.7)

0.0 17

18.8

(17.636.4)

<0. 001

(14.925.8)

(28.560.6)

16.2

36.8

19.8

28.6

42.2

(21.450.5)

(16.625.1)

(22.647.1)

<0. 001

9.7

(9.121.5)

<0. 001

(8.415.5)

(17.350.5)

8.3

26.1

10.5

18.6

33.1

(13.240.8)

(8.213.4)

(14.034.8)

<0. 001

5.7

(4.811.6)

<0. 001

(4.48.6)

(11.342.0)

<0. 001

5.6

11.8

3.2

24.5

(3.96.7)

(8.825.5)

<0. 001

(2.14.6)

(8.034.0)

<0. 001

3.2

7.0

2.0

16.8

(2.03.9)

(5.517.2)

<0. 001

(1.12.6)

(5.125.0)

<0. 001

1.7

4.7

1.3

11.0

(1.12.5)

(3.310.0)

<0. 001

(0.61.6)

(3.017.7)

<0. 001

1.0

3.1

0.8

6.8

(0.6-

(1.9-

<0. 001

(0.4-

(2.0-

PV 600

PV 550

4.2

(8.731.5)

CC E

(2.66.5)

21.1

A

PV 500

PV 450

PV 400

2.4

14.2

(1.43.5)

(5.223.9)

1.4

8.4

(0.72.0)

(3.317.0)

0.9

3.9

(0.4-

(2.0-

U

N

A

M

PV 650

ED

PV 700

p <0. 00 1

SC R

Good

IP T

Early

PT

All

<0. 00 1 <0. 00 1 <0. 00 1 <0. 00 1 <0. 00 1 <0. 00 1 <0. 00 1 <0. 00 1

10.5)

1.7)

5.5)

1.1)

11.1)

A

CC E

PT

ED

M

A

N

U

SC R

IP T

1.3)

Table 4. Adjusted OR for predicting poor outcome at 6 months after cardiac arrest All

Early

Late

N=110

N=44

N=66

95% CI

Adjuste d OR

95% CI

Adjusted OR

95% CI

Mean ADC

0.99

(0.98, 0.99)

0.989

0.9681.010

0.979

0.9650.993

PV 750

1.05

(1.02, 1.08)

1.028

0.9661.094

1.089

1.0271.156

PV 700

1.06

(1.03, 1.10)

1.061

0.9471.155

PV 650

1.09

(1.05, 1.14)

1.168

0.9881.380

PV 600

1.14

(1.07, 1.22)

1.425

1.0661.905

PV 550

1.22

(1.11, 1.35)

1.737

PV 500

1.40

(1.17, 1.67)

PV 450

1.97

(1.38, 2.80)

PV 400

3.91

1.0321.179

1.124

1.0371.218

SC R 1.103

1.0451.287

1.1462.631

1.234

1.0571.442

2.268

1.2384.158

1.511

1.0392.196

3.139

1.3197.471

3.969

0.89617.577

4.648

1.36615.814

11.283

1.136112.037

A

N

U

1.160

M

ED

PT

(2.00, 7.67)

CC E

Adjusted for shockable rhythm and time from collapse to ROSC

A

IP T

Adjuste d OR

Table 5. Sensitivity and specificity of mean ADC and PV 400 for predicting neurologic outcome at 6 months after cardiac arrest All Cutoff

AUC

Sensitivit 95% CI y

Specificit 95% CI y

Youden

<=736

0.781

58.9

46.8-70.3 97.3

85.8-99.9

0

<=627

(0.692, 0,854)

19.2

10.9-30.1 100.0

90.5-100.0

>1.72

0.895

80.8

69.9-89.1 94.6

81.8-99.3

>2.5

(0.836, 0.953)

67.1

55.1-77.7 100.0

90.5-100.0

Youden 0

726

0.698

Sensitivit 95% CI y

A

AUC

N

Early group Cutoff

SC R

proportion

U

Volume (400)

IP T

Mean ADC

Youden

M

Mean ADC

Specificit 95% CI y

44.0

24.4-65.1 100.0

82.4-100.0

Youden

>1.72

0.891

84.0

63.9-95.5 89.5

66.9-98.7

>2.50

(0.792, 0.989)

64.0

42.5-82.0 100.0

82.4-100.0

CC E

0

proportion

PT

Volume (400)

ED

(0.541, 0.827)

Late group Sensitivit 95% CI y

Specificit 95% CI y

A

AUC

Mean ADC Youden

<=778

0.823

75.0

60.4-86.4 94.4

72.7-99.9

0

<=627

(0.709, 0.906)

20.8

10.5-35.0 100.0

81.5-100.0

Volume (400)

proportion >1.66

Youden

0.906

79.2

65.0-89.5 100.0

81.5-100.0

A

CC E

PT

ED

M

A

N

U

SC R

IP T

(0.809, 0.964)