The independent prognostic value of brain diffusion tensor imaging in comatose patients after cardiac arrest

Abstract

Background/Objective

Predicting neurological outcomes in comatose cardiac arrest survivors remains challenging. Diffusion tensor imaging (DTI) offers potential as an objective biomarker of white matter injury, but its prognostic value needs further validation. We aimed to investigate the predictive value of DTI-derived metrics for six-month neurological outcomes in comatose cardiac arrest patients.

Methods

This prospective study enrolled 28 comatose cardiac arrest patients (mean age 54.36 ± 3.01 years; 71% male) and 28 age-/sex-matched healthy controls (HCs). All participants underwent 3T brain MRI (median 4 days post-arrest). DTI parameters (fractional anisotropy [FA], mean diffusivity [MD], axial diffusivity [AD], radial diffusivity [RD]) were analyzed using Tract-based spatial statistics (TBSS) and ROI approaches based on white matter atlas. Neurological outcome was assessed at six months using the modified Rankin Scale (good outcome: mRS 0–2; poor outcome: mRS 3–5). Statistical analyses included voxel-wise comparisons and ROC curve analysis for predictive performance.

Results

Compared to HCs, patients showed widespread reductions in FA, MD, AD, and RD (TFCE-corrected p < 0.05). Patients with poor outcomes (n = 18) exhibited significantly lower DTI metrics than those with good outcomes (n = 10) across most white matter tracts. The combination of whole-brain FA and RD demonstrated exceptional prognostic accuracy for good outcome (AUC = 0.984; 95% CI 0.925–1.000; sensitivity 92%, specificity 97.7%), significantly outperforming clinical variables and individual DTI parameters. ROI analysis identified specific tracts (e.g., right cingulum hippocampus, right uncinate fasciculus) with high predictive values. Ventricular fibrillation as initial rhythm was more frequent in the group with good outcomes.

Conclusions

DTI metrics, particularly the combination of FA and RD, provided outstanding early prediction of good six-month neurological outcomes after cardiac arrest, surpassing traditional biomarkers. These findings supported integrating DTI into multimodal prognostic models to guide clinical decisions and prevent premature withdrawal of life-sustaining therapy.

Introduction

Cardiac arrest is associated with high mortality and cerebral dysfunction due to prolonged whole-body ischemia [1, 2]. Despite advances in resuscitation and post-cardiac arrest care, predicting neurological outcomes in comatose patients remains a significant challenge. Most studies focus more on predictors of poor outcomes after cardiac arrest. However, because some patients with potential for good recovery experience premature withdrawal of life sustaining treatment (WLST), predicting good outcomes is equally crucial [3, 4]. Current prognostic assessment tools, including clinical examinations, electrophysiological tests, and serum biomarkers such as neuron-specific enolase (NSE), vary in their sensitivity and specificity, while being subject to multiple confounding factors [5, 6]. These limitations underscore the need for objective, early-stage biomarkers to guide clinical decision-making and improve prognostic accuracy.

Ischemic hypoxic injury and ischemia-reperfusion induces cerebral edema in comatose patients after resuscitation from cardiac arrest [7, 8]. Within 1 hour after bilateral carotid artery occlusion in cats, MRI shows diffuse cellular swelling, including neurons and glial cells. The two main pathological mechanisms were cytotoxic edema, from cellular homeostasis failure with osmotic gradient-driven fluid accumulation, and vasogenic edema, due to blood-brain barrier break down from endothelial dysfunction and tight junction disruption. Two types of edema can exist independently or simultaneously in the same disease stage [7].

Magnetic resonance imaging (MRI) has emerged as a promising modality for evaluating brain injury after cardiac arrest. Our previous research has shown that rs-fMRI displays complex changes in gray matter signal abnormalities in resuscitated patients, as rs-fMRI is sensitive to both cortical and subcortical neuronal activities [9]. Diffusion weighted imaging detects the diffusion of water molecules and is sensitive to detecting cytotoxic edema, thereby reducing the diffusion rate, providing microstructural white matter integrity and insights into axonal and myelin damage [10, 11]. Prior studies have demonstrated associations between the apparent diffusion coefficient or fractional anisotropy (FA) and neurological outcomes in cardiac arrest and brain injury patients [12–16]. However, the prognostic value of DTI in cardiac arrest patients remains underexplored. Furthermore, existing studies often lack rigorous comparisons between patients and healthy controls (HCs) or fail to account for region-specific white matter alterations.

This prospective study aimed to investigate the predictive utility of DTI-derived parameters, for assessing six-month neurological outcomes in comatose cardiac arrest patients. We hypothesized that microstructural white matter changes, quantified by DTI, would correlate with functional recovery and outperform traditional prognostic markers. By comparing whole brain and region of interest (ROI) analyses with clinical variables, this work aimed to advance personalized prognostication and inform targeted therapeutic strategies.

Methods

Study design and participants

This is a prospective study to identify prognostic biomarkers for neurological recovery of comatose patients after cardiac arrest. Between January 2021 and December 2023, we consecutively recruited cardiac arrest patients admitted to the emergency intensive care unit (EICU) of Beijing Chaoyang Hospital. Exclusion criteria were: age < 18 years, pre-existing structural brain lesions (e.g., stroke, cerebral hemorrhage), pregnancy, contraindications to MRI (e.g., metallic implants), severe anemia (hemoglobin < 120 g/L), and lack of informed consent from legal surrogates. The clinical data were the age, sex, witness, initial rhythm, etiology, time to return of spontaneous circulation (ROSC), and time from cardiac arrest to MRI. The outcome was assessed six months after the evaluation using a structural script, the modified Rankin Scale (mRS) score, through a phone interview with the patient or their close relative. The mRS score was divided into good (mRS 0–2) and poor outcome (mRS 3–5). All patients underwent brain MRI and blood gas analyses at same day, and serum neuron-specific enolase (NSE) levels on day 3. All patients received a standardized targeted temperature management (TTM) protocol involving 24-h maintenance at 33 °C and subsequent rewarmed at 0.1 °C/h to 37 °C, facilitated by an automated surface cooling device. More detailed treatment protocols in this study were performed as described previously [9]. At the same time, age- and sex-matched healthy controls (HCs) were included from the community to observe the changing trends of the white matter microstructure in the patients. The exclusion criteria included that the history of neurological or psychiatric disorders (e.g., stroke, dementia, major depression), brain space occupying lesions, cardiovascular disease, and claustrophobia.

This study was approved by the Ethics Committee Boards of Beijing Chaoyang Hospital, Capital Medical University and written informed consent was obtained from all patients’ next of kin at the time of inclusion.

Image acquisition

After successful rewarming to achieve and maintain normothermia, along with 24 h maintenance of hemodynamic stability, brain MRI performed on a 3.0 T scanner (MAGNETOM Prisma, Siemens Healthcare) with a 64-channel head coil. The 3D T1-weighted magnetization prepared rapid acquisition gradient echo (MPRAGE) imaging parameters were: repetition time (TR) = 2300 ms, echo time (TE) = 2.98 ms, inversion time (TI) = 900 ms, slice thickness = 1 mm, flip angle = 9°, field of view (FOV) = 256 × 256 mm², number of slices = 192. The T2 imaging parameters were: TR = 4500 ms, TE = 94 ms, slice thickness = 5 mm, number of slices = 20. T2-weighted fluid attenuated inversion recovery (FLAIR) imaging parameters were: TR = 8000 ms, TE = 81 ms, slice thickness = 5 mm, number of slices = 20. DTI was acquired in the posterior-anterior (P-A) phase-encoding direction, one non-diffusion-weighted volume (b = 0 s/mm²) and 45 diffusion-weighted volumes with a b-value of 1000 s/mm² were obtained. To address susceptibility-induced distortions, an additional reverse phase-encoding dataset was acquired in the opposite (A-P) direction, comprising one b = 0 s/mm² volume and six diffusion-weighted volumes at b = 1000 s/mm². This bidirectional acquisition protocol was implemented to enable subsequent eddy current and geometric distortion correction during postprocessing. All HCs underwent the identical brain MRI protocol on the same scanner using the same head coil as the patient cohort.

Data processing

After converting DICOM files to NIFTI images, DTI image pre-processing was performed using the FMRIB Software Library (FSL, version 6.0.7; http://www.fmrib.ox.ac.uk/fsl) [17]. The diffusion-weighted images with opposing phase-encoding directions (P-A and A-P) were combined and processed to correct for susceptibility-induced distortions and eddy currents using the topup and eddy tools. Then we performed the brain extraction and tensor calculation to generate fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) maps [18]. Subsequently, DTI data were skeletonized using the Tract-Based Spatial Statistics analysis performed using the standard pipeline as implemented in FSL: FA maps were used as input and non-linear aligned to FMRIB58_FA maps. All aligned FA images were transformed onto the MNI152 template using affine registration. Then, mean FA skeleton was thresholded at a value of FA >0.2 to exclude non-white matter regions. Each individual’s aligned FA map was then projected onto this skeleton, resulting in a 4D skeletonized FA image containing all subjects’ data. Data on MD, AD, and RD were also aligned to the standard space using the transformation matrices derived from the FA registration and projected onto the mean FA skeleton [19].

To characterize white matter microstructural integrity of each individual, we measured FA, MD, AD, and RD in the skeleton. The skeleton was segmented in 20 predefined regions of interest (ROI) by using the Johns Hopkins University (JHU) DTI-based white matter atlas. The binary mask derived from the voxel-wise statistical analysis. For each subject, the mean values of the diffusion metrics (FA, MD, AD, and RD) were extracted from whole-brain white matter and specific white matter tracts derived from the intersection of the statistical mask and the JHU atlas.

Statistical analysis

This study assessed whether demographic and clinical data were statistically different between HCs and patients, good and poor outcome groups. The independent t-test was used if continuous data followed a normal distribution and exhibited homogeneous variance and data were presented as mean (SD). If the data were non-normally distributed or had heterogeneous variance, the non-parametric Mann-Whitney U test was used, and data are presented as median (interquartile range). For categorical variables, chi-square tests were performed and percentages were obtained.

Voxel-wise cross-subject statistics were performed on the skeletonized data using the “randomize” tool with 5000 iterations to compare differences between HCs and cardiac arrest patients, the patients with good outcomes and poor outcomes [20]. Age and sex as covariances were included in the design matrix as necessary. Threshold-free cluster enhancement (TFCE) was applied to correct for multiple comparisons, with a significance threshold of p < 0.05.

To assess the predictive value of variables demonstrating significant inter-group differences, we constructed six distinct models: one based on clinical variables, one on whole-brain DTI measures, and four on ROI-specific metrics. Within each model, stepwise regression was employed to evaluate the predictive value of individual variables. For the ROI-based predictive models, the mean values of each DTI metric (FA, MD, AD, RD) from all 20 JHU white matter atlas regions were used as the initial candidate predictors for the respective stepwise regression analyses. For variables achieving statistical significance (p < 0.05) in their respective models, receiver operating characteristic (ROC) curves were generated. The area under the curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value for predicting good outcome were then calculated. Internal validation was performed using bootstrapping with 1000 resamples to calculate an optimism-corrected AUC and mitigate the risk of overfitting. All analyses were performed using RStudio (version 2024.12.1 + 563).

Results

Participant characteristics

The study enrolled 33 patients who met the inclusion criteria in the Chaoyang hospital. Among them, 3 patients were excluded because they did not complete the MRI exam. Another 2 patients were excluded because of bad image quality, mainly due to severe artifacts caused by excessive head motion during visual inspection of images. Finally, this study included 28 comatose cardiac arrest patients (54.36 ± 3.01; 71% male) and 28 HCs (48.57 ± 2.16; 50% male). No significant differences were observed between cardiac arrest patients and HCs in age (p = 0.124) and sex (p = 0.101). Among cardiac arrest patients, comparisons between good (n = 10) and poor outcome (n = 18) groups revealed no significant differences in age (p = 0.464), sex (p = 0.454), witnessed arrest rate (p = 0.662), or time to ROSC (p = 0.291). However, ventricular fibrillation as the initial rhythm was more frequent in the good outcome group (80% vs. 28%, p = 0.008) (Table 1).

Table 1.

Patients characteristics

Characteristics	All patients (n = 28)	Good outcomes (mRS 0–2; n = 10)	Poor outcomes (mRS 3–5; n = 18)	p
Demographic characteristics
Age, years	54.36 ± 3.01	60.00 (35.25)	56.00 (20.50)	0.464
Male sex, n (%)	20 (71)	8 (80)	12 (67)	0.454
Characteristics of the cardiac arrest
Witness, n (%)	26 (93)	9 (90)	17 (94)	0.662
Time to ROSC, min	21 (19.5)	20 (6.75)	25.5 (29.5)	0.291
Initial rhythm, n (%)				0.008
PEA	15 (54)	2 (20)	13 (72)
VF	13 (46)	8 (80)	5 (28)
Etiology				0.274
Cardiogenic	24 (86)	10 (100)	14 (78)
Respiratory origin	2 (7)	0 (0)	2 (11)
Other	2 (7)	0 (0)	2 (11)
Time from CA to MRI, days	4.04 ± 2.91	5.40 ± 3.34	3.28 ± 2.42	0.063
Clinical characteristics
NSE, ng/mL	56.80 (60.8)	40.84 (25.21)	88.05 (55.7)	0.044
pH	7.34 ± 0.16	7.38 ± 0.04	7.32 ± 0.20	0.261
PaCO2, mmHg	34.54 ± 8.32	37.83 ± 7.04	32.72 ± 28.58	0.103
PaO2, mmHg	152 (80.5)	135 (85.75)	170.5 (67)	0.179
P/F	322 (178)	444.67 (155.83)	294.28 (123.37)	0.058
Whole brain parameters
FA	0.51 (0.07)	0.54 (0.03)	0.49 (0.07)	< 0.001
MD	7 × 10− 4 (2.25 × 10− 4)	7.5 × 10− 4 (1.0 × 10− 4)	6.5 × 10− 4 (2.0 × 10− 4)	0.019
AD	1.15 × 10− 3 (3.25 × 10− 4)	1.2 × 10− 3 (0.0 × 10− 4)	1.0 × 10− 3 (3.75 × 10− 4)	< 0.001
RD	5 × 10− 4 (2 × 10− 4)	6 × 10− 4 (0.0 × 10− 4)	5 × 10− 4 (2 × 10− 4)	0.005

Data are presented as mean (SD), n (%), or median (IQR). ROSC, return of spontaneous circulation; PEA, pulseless electrical activity; VF, ventricular fibrillation; CA, cardiac arrest; MRI, magnetic resonance imaging; NSE: neuron - specific enolase; PaCO₂: arterial partial pressure of carbon dioxide; PaO₂: arterial partial pressure of oxygen; P/F: PaO₂/FiO₂ ratio (where PaO₂ represents arterial partial pressure of oxygen and FiO₂ represents fraction of inspired oxygen); FA: fractional anisotropy; MD: mean diffusivity; AD: axial diffusivity; RD: radial diffusivity

The NSE were significantly different between the good and poor outcome groups. There were no significant differences in pH, PaCO₂, PaCO₂, and PaO₂/FiO₂ ratio (P/F) (Table 1).

Tract-based spatial statistics

After controlling for age and sex as covariates, TBSS of FA, MD, AD, and RD identified significantly widespread decreases in patients compared to HCs (Fig. 1A), and in patients with poor outcomes compared to the good outcomes (Fig. 1B) (TFCE-corrected p < 0.05).

Fig. 1.

The spatial distributions of differences in fractional anisotropy, mean diffusivity, axial diffusivity, and radial diffusivity. The blue areas show significantly decreased values in (A) patients compared with HCs, (B) the patients with poor outcomes compared with the good outcomes (p < 0.05). HC: health control; CA: cardiac arrest; FA: fractional anisotropy; MD: mean diffusivity; AD: axial diffusivity; RD: radial diffusivity

Figure 2 showed the specific white matter changes in the ROIs of FA, MD, AD, and RD respectively in patients with good outcomes and poor outcomes. Only the right cingulum hippocampus of FA showed no difference between the good and poor outcome groups. And the values of other all regions were significantly decreased in the patients with poor outcomes.

Fig. 2.

Histogram based on ROI metrics

Prognostic value of variables

In the clinical variables model, significant variable was NSE + initial rhythm (AUC: 0.894). The whole brain DTI measures, which included FA and RD, increased the AUC to 0.984 (95% CI 0.925–1.000), with 97.7% specificity and 92% sensitivity. In the ROI-specific metrics model of the multivariate analyses, significant regions selected by stepwise regression were the right anterior thalamic radiation (ATR_R) in FA (AUC 0.853), the right cingulum hippocampus (CH_R) and left inferior fronto-occipital fasciculus (IFOF_L) in MD (AUC 0.916), the CH_R and right uncinate fasciculus (Unc_R) in AD (AUC 0.978), and CH_R in RD (AUC 0.818) as shown in Table 2; Fig. 3.

Table 2.

Area under the receiver operating characteristic curve

Characteristics	Area under the ROC curve (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	Predictive positive value (95% CI)	Negative positive value (95% CI)
Clinical variable
NSE	0.733 [0.523, 0.925]	0.407 [0.000, 0.929]	0.850 [0.529, 1.000]	0.466 [0.000, 1.000]	0.748 [0.571, 0.933]
Initial rhythm	0.753 [0.575, 0.912]	0.653 [0.000, 1.000]	0.788 [0.545, 1.000]	0.534 [0.000, 0.875]	0.834 [0.625, 1.000]
NSE + initial rhythm	0.894 [0.738, 1.000]	0.731 [0.200, 1.000]	0.863 [0.647, 1.000]	0.753 [0.333, 1.000]	0.857 [0.654, 1.000]
Whole brain DTI measures
FA	0.891 [0.743, 0.994]	0.756 [0.222, 1.000]	0.855 [0.643, 1.000]	0.754 [0.444, 1.000]	0.876 [0.720, 1.000]
RD	0.807 [0.645, 0.926]	0.694 [0.000, 1.000]	0.776 [0.500, 1.000]	0.565 [0.000, 0.875]	0.847 [0.667, 1.000]
FA + RD	0.984 [0.925, 1.000]	0.920 [0.700, 1.000]	0.977 [0.850, 1.000]	0.965 [0.750, 1.000]	0.956 [0.824, 1.000]
ROIs of FA
ATR_R	0.853 [0.683, 0.977]	0.665 [0.143, 1.000]	0.847 [0.643, 1.000]	0.716 [0.286, 1.000]	0.835 [0.682, 1.000]
ROIs of MD
CH_R	0.907 [0.783, 0.993]	0.766 [0.375, 1.000]	0.866 [0.692, 1.000]	0.761 [0.499, 1.000]	0.879 [0.737, 1.000]
IFOF_L	0.755 [0.444, 0.917]	0.486 [0.000, 1.000]	0.794 [0.467, 1.000]	0.573 [0.000, 1.000]	0.765 [0.579, 1.000]
CH_R + IFOF_L	0.916 [0.802, 1.000]	0.774 [0.400, 1.000]	0.884 [0.714, 1.000]	0.799 [0.538, 1.000]	0.880 [0.733, 1.000]
ROIs of AD
CH_R	0.960 [0.871, 1.000]	0.865 [0.666, 1.000]	0.927 [0.765, 1.000]	0.880 [0.633, 1.000]	0.925 [0.800, 1.000]
Unc_R	0.851 [0.690, 0.981]	0.684 [0.000, 1.000]	0.813 [0.562, 1.000]	0.656 [0.000, 0.933]	0.844 [0.667, 1.000]
CH_R + Unc_R	0.978 [0.904, 1.000]	0.917 [0.667, 1.000]	0.973 [0.857, 1.000]	0.958 [0.778, 1.000]	0.953 [0.818, 1.000]
ROIs of RD
CH_R	0.818 [0.610, 0.961]	0.602 [0.000, 0.909]	0.826 [0.538, 1.000]	0.685 [0.000, 1.000]	0.799 [0.615, 0.950]

ROC: receiver operating characteristic; NSE: neuron - specific enolase; DTI: diffusion tensor imaging; FA: fractional anisotropy; RD: radial diffusivity; MD: mean diffusivity; AD: axial diffusivity; ATR_R: Anterior_thalamic_radiation_R; CH_R: Cingulum_hippocampus_R; IFOF_L: Inferior_fronto_occipital_fasciculus_L; Unc_R: Uncinate_fasciculus_R

Fig. 3.

Receiver operating characteristic curve. NSE: neuron - specific enolase; DTI: diffusion tensor imaging; FA: fractional anisotropy; MD: mean diffusivity; AD: axial diffusivity; RD: radial diffusivity; ATR_R: Anterior_thalamic_radiation_R; CH_R: Cingulum_hippocampus_R; IFOF_L: Inferior_fronto_occipital_fasciculus_L; Unc_R: Uncinate_fasciculus_R; ROI: region of interst

Discussion

This study demonstrates that DTI-derived parameters, particularly FA and RD, exhibit exceptional prognostic accuracy for predicting six-month neurological outcomes in comatose cardiac arrest patients. The combination of FA and RD achieved an AUC of 0.98, surpassing the predictive performance of clinical variables such as NSE and blood gas parameters. These findings align with prior evidence implicating white matter integrity as a critical determinant of functional recovery after hypoxic-ischemic injury. Reduced FA and RD in patients with poor outcomes likely reflect axonal degeneration and demyelination, processes exacerbated by prolonged cerebral hypoxia during cardiac arrest.

We found that the patients with good outcomes had a higher rate of ventricular fibrillation as the initial rhythm. The ventricular fibrillation was one type of shockable rhythms, which was found highly associated with outcome after in-hospital cardiac arrest [21]. The same trend showed in the out-of-hospital study that survival from a shockable rhythm greatly exceeds survival after pulseless electrical activity [22]. Goto et al. found that out-of-hospital patients with initial nonshockable rhythms may later develop shockable rhythms during resuscitation that was associated with 1-month good outcomes [23]. This may be because shockable rhythms are related to the effectiveness of early defibrillation.

NSE is the most commonly used hematological indicator for monitoring the status of patients with return of spontaneous circulation after cardiac arrest, and shows an ability to distinguish good and poor outcomes among individuals [24, 25]. In an observational study, NSE at day 3 led to an AUC 0.89 in predicting neurological outcomes [26]. However, in our study the NSE at day 3 had an AUC 0.733 with specificity of 85% for a sensitivity of only 40.7%. Several factors may explain this finding. First, a well-recognized lack of standardization in NSE assays across institutions and platforms could lead to substantial inter-laboratory variability and impact absolute values and prognostic thresholds [27]. Second, as a blood test, NSE was vulnerable to interference from a range of biological and preanalytical factors, and exhibited marked inter-individual and temporal variability. Finally, and most importantly, our study specifically aimed to discriminate good from poor outcomes among comatose patients of cardiac arrest. In our study, the patients had already passed the earliest, highest-mortality filter. In studies that focused primarily on predicting poor outcome, NSE often performed very well because non‑survivors frequently have markedly elevated values. Our results might reflect the real-world performance of NSE in the specific context of prognostication among comatose survivors, where the predictive value of NSE for identifying good outcomes appeared more limited than the prediction of poor outcomes [28]. Indeed, in this study, despite the initial rhythm, the other characteristics had no significant differences in the good and poor outcome groups. This baseline homogeneity strengthens the reliability of our observed lower specificity, which may stem from differences in measurement protocols or inherent biological variability, suggesting our findings may better reflect real-world predictive performance.

The results of this study are aligned with previous studies, that have shown widespread decreased FA values in patients compared to HCs, with further differentiation between good and poor outcome groups, which may indicated brain injury and microstructural abnormalities in comatose patients of cardiac arrest [12, 14]. The reduced FA in white matter reflects compromised microstructural integrity primarily attributed to hypoxia-induced oligodendrocyte injury, triggering sequential pathological cascades including cellular edema, necrosis, demyelination, and subsequent axonal degeneration [29–31]. Particularly, we observed that MD, AD, and RD were decreased in patients with poor prognosis, which is distinctly different from the increased diffusion metrics associated with isolated demyelination in chronic disorders [32–34]. The underlying mechanism is attributed to restricted diffusion secondary to cytotoxic edema leading to severe cellular swelling, which peaks in severity around 5 days post-cardiac arrest – a time window notably aligned with our observational period [35–37]. The dilution effect of cytotoxic edema also could lead to a decrease in FA. The lack of significant correlation between these DTI metrics and the time-to-MRI in our study suggested that this state of profound diffusion restriction was the dominant and relatively stable microstructural change characteristic during this specific phase. Therefore, we inferred that during this critical time window, DTI metrics may provide optimal diagnostic accuracy.

Extensive studies showed utility for the FA derived from DTI in prognostic value in poor outcomes [38, 39]. However, Koskensalo et al. reported different results that the performance of FA is inferior to NSE in predicting poor prognosis in cardiac arrest patients [40], possibly due to differences in time to undergo MRI (their median = 53 h vs. our median = 4 days). Our results indicate that the DTI-derived measures, particularly the combined FA + RD model, significantly outperform NSE and individual DTI parameters in prognostic accuracy achieving an outstanding AUC of 0.984, perfect specificity (97.7%), and high sensitivity (92%). This suggests that microstructural integrity assessed by DTI is a more reliable biomarker for identifying patients likely to achieve a good neurological outcome after ROSC. This synergistic effect likely reflects complementary pathophysiological information that FA captures microstructural integrity while RD specifically sensitizes to myelin damage—a critical pathological feature in hypoxic-ischemic injury [41]. Furthermore, regional analyses highlighted the significance of specific white matter tracts, with AD in the CH_R and Unc_R showing excellent predictive performance. These regions are critical for cognitive and memory functions often impaired following hypoxic-ischemic injury related to cardiac arrest. The results potentially indicated selective vulnerability of axonal projections in this memory-critical region [42]. Overall, DTI’s ability to detect subtle white matter changes non-invasively supports its integration into multimodal prognostic models, potentially improving clinical decision-making and patient counseling in post-cardiac arrest care.

Our study has several limitations. First, the limited sample size may result in inflated effect estimates, particularly impacting the predictive validity of our findings. Second, the exclusion of patients uncompleting MRI exam may introduce selection bias. Third, the cross-sectional design of MRI acquisition (median 4 days post-cardiac arrest) cannot capture temporal changes in white matter pathology. Future multicenter studies with longitudinal imaging and larger cohorts are needed to validate these findings and establish standardized DTI thresholds for clinical use.

Conclusions

This study demonstrated that the combination of DTI-derived FA and RD provided outstanding prediction of long term neurological outcomes after cardiac arrest. The superior standalone performance of FA + RD underscored their potential as primary prognostic tools where advanced imaging is available. Integrating these metrics into multimodal prognostic models could significantly improve clinical decision-making and prevent premature WLST. Future multicenter studies are needed to validate and translate these biomarkers into clinical practice.

Abbreviations

AD

Axial diffusivity

ATR_R

The right anterior thalamic radiation

AUC

Area under curve

CH_R

The right cingulum hippocampus

DTI

Diffusion tensor imaging

EICU

Emergency intensive care unit

FA

Fractional anisotropy

HC

Healthy control

IFOF_L

The left inferior fronto-occipital fasciculus

JHU

Johns Hopkins University

MD

Mean diffusivity

MRI

Magnetic resonance imaging

mRS

Modified Rankin Scale

NSE

Neuron-specific enolase

RD

Radial diffusivity

ROC

Receiver operating characteristic

ROI

Region of interest

TBSS

Tract-based spatial statistics

TFCE

Threshold-free cluster enhancement

Unc_R

The right uncinate fasciculus

WLST

Withdrawal of life sustaining treatment

Author contributions

Ziren Tang, and Qi Yang, conceptualization and design of the study; Xuejia Jia and Rui Shao, and Yingying Li, material preparation, and data collection; and Xuejia Jia, Rui Shao, and Xiuqin Jia, the manuscript drafting and data analyses. All the authors reviewed the manuscript and provided feedback.

Funding

This study was funded by the National Natural Science Foundation of China (grant number: 82025018, U24A20753), the Beijing Municipal Natural Science Foundation (L252089), and Beijing Hospitals Authority’s Ascent Plan (DFL20220303, DFL20240302).

Data availability

Not applicable.

Declarations

Competing interests

The authors declare no competing interests.