Acute ischemic stroke (AIS) is one of the major diseases causing disability and death. Ischemic stroke, especially intracranial large blood vessel occlusion, imposes a large financial burden on society and families due to its poor prognosis [1]. Identifying the etiology of AIS is therefore of great significance for the selection of secondary prevention measures and treatment choices. For secondary prevention, anticoagulant drugs such as warfarin are preferred for cardioembolic (CE) [2] stroke. Compared to non-cardioembolic (NCE) thrombi, CE thrombi are increasingly considered to be more difficult to remove, with catheter aspiration often used as a supplemental method to stent retrieval during mechanical thrombectomy [3, 4, 5, 6]. This may be due to a stiffer thrombus and greater friction coefficient [7]. It is therefore important to identify the source of thrombi early, allowing for timely treatment with mechanical thrombectomy. Atrial fibrillation (AF) is the main cause of CE stroke, but some AF is not discovered until the left atrial appendage thrombi are shed off. No documented prior history of AF increases the difficulty in identifying CE stroke at admission due to the limitation of emergency electrocardiograms (ECGs) on detecting paroxysmal AF. CE stroke can also be triggered by other cardiogenic causes, such as patent foramen ovalis. Therefore, other indicators besides AF are needed to identify CE stroke.

Some readily available clinical indicators can help to diagnose the etiology of stroke. CE stroke is more common in women with high baseline National Institutes of Health Stroke Scale (NIHSS) [8, 9, 10, 11]. Patients with hypertension often progress to NCE, mainly large artery atherosclerosis (LAA) stroke. Compared to NCE thrombi, CE thrombi usually have higher thrombus permeability, which indicates the ability of contrast agents to permeate the irregular space in the thrombus [12, 13]. Conventional permeability is evaluated as the increased thrombus attenuation between noncontrast computed tomography (NCCT) and CT angiography (CTA) images [14, 15]. This method requires precise co-registration, which is time-consuming and difficult to achieve in clinical practice, as well as being prone to inevitable measurement error [16]. Dual-energy CT (DECT) generates iodine overlay maps based on material decomposition technology. It can directly obtain the penetration amount of the contrast agent by displaying the iodine concentration (IC) in the thrombus [17, 18, 19]. The IC of the thrombus was normalized to the IC of the contralateral normal vessel to derive a normalized IC (NIC) value, thereby eliminating the potential effect of individual circulation differences on the IC of the lesion. The NIC of the thrombus is a new permeability parameter that is assessed using iodine overlay maps alone without the co-registration process. Therefore, we hypothesized that NIC may have higher clinical value for identifying CE stroke, with a simpler operation process and less measurement error. In addition, previous studies have suggested that there were differences in the thrombus composition between CE and NCE stroke [13, 20, 21, 22]. CE thrombi consist of fewer red blood cells (RBCs) and more fibrin/platelet (FP) conglomerations, while the opposite is seen for NCE thrombi. DECT allows the operator to analyze how the material-specific attenuation varies with X-ray photon energy [23]. We suspected that DECT could distinguish between thrombi with similar X-ray attenuation but different histopathologic compositions by atomic numbers and spectral attenuation curves, as represented by the effective atomic number (Zeff) and slope ($\lambda$_{Hounsfield unit [HU]}), respectively. To our knowledge, there have been no reports on the performance of DECT thrombus variables in combination with clinical data for the identification of CE stroke.

The aims of this study were therefore to identify quantitative DECT thrombus parameters that are associated with CE stroke, and to subsequently combine these variables with clinical data in a nomogram to identify CE stroke.

Determining the source of thrombi in stroke patients is of great value for the selection of secondary prevention and treatment methods [2, 3, 4, 5, 6]. AF is an important diagnostic proof of CE stroke while paroxysmal AF is usually difficult to be detected by emergency ECG at admission because the ECG shows normal results during non-episodic periods of paroxysmal AF. Studies have shown that thrombus permeability parameters assessed on conventional NCCT and CTA scans can help to predict CE stroke [12, 13], but this approach is heavily dependent on accurate image co-registration. In the event that co-registration is unsuccessful, measurement errors are unavoidable [16]. Moreover, the measurement process is complex and time-consuming, thus limiting its clinical utility. On the other hand, DECT can produce iodine overlay maps that directly assess the thrombus permeability, thereby eliminating the image co-registration process [17, 18, 19]. This approach offers greater simplicity, efficiency, and precision in clinical settings. Moreover, DECT can provide additional tissue information due to the acquisition of high- and low-energy data which may be helpful in identifying the source of thrombi [25, 26]. The current study aimed to construct a nomogram model based on DECT to determine the etiology of stroke when the AF history is unknown, or when the stroke is caused by other cardiogenic causes. We found the following: (a) NIC, Zeff, $\lambda$_HU and baseline NIHSS were all significantly greater in patients with CE stroke than in those with NCE stroke (p 0.001, p = 0.009, p = 0.007, and p 0.001, respectively). Gender, diabetes mellitus, hypertension and smoking history were each significantly associated with stroke etiology (p = 0.002, p = 0.018, p = 0.022, and p = 0.002, respectively); (b) multivariable analysis identified four variables (NIC, gender, hypertension, baseline NIHSS) that were significantly associated with CE stroke (p 0.001, p = 0.002, p = 0.019, and p = 0.003, respectively); (c) The nomogram constructed using these four variables identified CE stroke with an AUC of 0.929 in the study sample and 0.899 in the validation cohort. Calibration plots both revealed good consistency between the predicted and observed probabilities.

Thrombi consist of FP, RBCs, a small number of white blood cells (WBCs), and other components that are cross-linked to form irregular gaps. Together, these constitute the histological basis of thrombus permeability [20]. In agreement with previous studies [12, 13], we observed higher permeability in CE thrombi with greater NIC (CE vs NCE: 0.104 $\pm$ 0.043 vs 0.051 $\pm$ 0.033, p 0.001). In multivariable analysis, NIC was independently associated with the stroke etiology. Most studies have reported that CE thrombi consist mainly of FP while NCE thrombi consist mainly of RBCs [13, 20, 21, 22]. Fibrin is thought to be attractive to iodide, and the fibrin network is a porous and organized structure that is more conducive to the penetration of contrast agents [27, 28], meaning that thrombi dominated by FP often show high permeability. This may explain why CE thrombi tend to be more pervious. In contrast, less permeability was more frequently observed in thrombi dominated by RBCs, as these cells are firmly bound to each other, making them more impenetrable to contrast agents [29]. Therefore, NCE thrombi are often less pervious. However, a study by Boodt et al. [15] suggested that thrombus permeability is not related to stroke etiology. This discord may be explained by the fact that Boodt et al. [15] also included patients with small artery occlusion in their study, such as M2 distal segment and M3 segment of the middle cerebral artery, in addition to intracranial proximal artery occlusion. Although the scope of their study was wider than the present work, thrombi in small arteries were too small to place ROIs, hence the measurement error was increased. Furthermore, Boodt et al. [15] used traditional permeability parameters that were assessed on NCCT and CTA images, requiring precise co-registration. This is difficult to achieve in clinical practice, leading to uncertainty in the actual measurements. The DECT permeability parameter (NIC) employed in our study was assessed directly on iodine overlay maps without the need for co-registration, and thus more accurately reflected the actual thrombus permeability.

The present study also found that Zeff and $\lambda$_HU were greater in CE thrombi than in NCE thrombi. The Zeff and $\lambda$_HU of the thrombus assessed on CTA images represent not only the thrombus itself, but also the contrast agent that penetrated into the thrombus which is a strong indicator of stroke etiology. After adjusting for NIC and clinical variables, no statistically significant association was detected between Zeff (or $\lambda$_HU) and CE stroke. Therefore, we conclude that Zeff and $\lambda$_HU of the thrombus itself cannot be used to distinguish CE stroke from NCE stroke.

This research also found that CE stroke was more commonly seen in female patients with increased baseline NIHSS and without hypertension. Studies on gender differences in stroke have generally found that women are more likely to suffer from CE stroke, whereas men are more likely to suffer from LAA stroke due to higher rates of smoking and drinking [8, 9]. Zotter et al. [10] and Guglielmi et al. [11] suggested that CE thrombi were often larger than LAA thrombi, which could easily lead to proximal intracranial occlusions and consequently a higher baseline NIHSS, in agreement with the findings of our study. Hypertension is generally recognized as a risk factor for atherosclerosis, so a high prevalence of LAA stroke, rather than CE stroke, was noted in patients with hypertension. The present study used NIC, gender, baseline NIHSS and hypertension history to develop the nomogram, which was demonstrated to have a good validity and reliability. The AUC of this nomogram was slightly higher than that of AF, although the difference was not statistically significant (0.929 vs 0.869, p = 0.130). Therefore, our nomogram can be used to identify CE stroke when the existence of AF is difficult to confirm. Other cardiogenic causes besides AF can also lead to CE stroke, such as patent foramen ovalis. It is easy to be misdiagnosed stroke etiology in this condition when based only on AF. In addition, AF also occurs in some NCE stroke patients. These may explain why the AUC of AF in identifying CE stroke was slightly lower than that of our nomogram.

Timely and rapid identification of CE stroke would greatly assist neuro-interventional physicians in making optimal mechanical thrombectomy plans [30, 31]. Our study used the emergency DECT parameter (NIC) and clinical indicators (gender, baseline NIHSS, and hypertension history) to construct a nomogram that can be a useful tool for identifying CE stroke at admission. At later stages of treatment, some occult AF or other cardiogenic causes are still undetectable, and our nomogram could then also be used to determine the source of thrombi.

There were several limitations to this study. First, only patients with anterior circulation stroke and proximal intracranial occlusions were included. Whether our result is applicable to patients with distal artery occlusion of anterior circulation or posterior circulation requires further study. Second, the sample size of our study was modest due to the limitations of DECT scanning. DECT is an advanced CT equipment capable of performing both conventional and dual-energy scans. Although it is more expensive than traditional CT, DECT offers superior image quality and richer diagnostic information without increasing the radiation dose to patients. With the growing clinical adoption of DECT in recent years, large multicenter DECT studies on stroke will become feasible in the near future. Third, the retrospective study design meant that some clinical data, such as the use of anticoagulants, was missing. This can be overcome by conducting further prospective studies. Fourth, Zeff and $\lambda$_HU of the thrombus were assessed on DECTA images, which undergo interference from the contrast agent in the thrombus. For this reason our study used NIC, which represents the extent of penetration of the contrast agent, to adjust the results. Further studies using dual-energy NCCT are needed to directly assess the Zeff and $\lambda$_HU of the thrombus itself. Fifth, we excluded patients with vascular calcifications near the thrombus. This may have introduced selection bias, as it excluded some patients with LAA stroke, thereby potentially skewing the accuracy of the nomogram. However, only patients with intracranial proximal artery occlusion (mainly the M1 and proximal M2 segments of the middle cerebral artery) were included in our study. Vascular calcifications are relatively uncommon in such patients, which may have reduced the potential effect of selection bias on our research findings. Finally, our study lacked histopathological analysis of thrombi, which could have provided important validation of the observed associations between DECT parameters and thrombus composition; further studies are required.

We demonstrated that a DECT thrombus permeability parameter, NIC, is a useful indicator of CE stroke. A nomogram constructed using NIC, gender, baseline NIHSS and hypertension history showed good diagnostic performance for CE stroke. We believe this nomogram may be especially valuable when the AF history is unknown, or the stroke is caused by other cardiogenic causes.

AIS, Acute ischemic stroke; CE, Cardioembolic; NCE, Non-cardioembolic; AF, Atrial fibrillation; ECG, Electrocardiogram; NIHSS, National Institutes of Health Stroke Scale; LAA, Large artery atherosclerosis; NCCT, Noncontrast computed tomography; CTA, CT angiography; DECT, Dual-energy CT; IC, Iodine concentration; NIC, Normalized iodine concentration; RBC, Red blood cell; FP, Fibrin/platelet; Zeff, Effective atomic number; HU, Hounsfield unit; $\lambda$_HU, Slope; SOE, Other determined etiology; DECTA, Dual-energy CTA; GSI, Gemstone Spectral Imaging; WC, Water concentration; ROI, Region of interest; TOAST, Trial of Org 10172 in Acute Stroke Treatment; OR, Odds ratio; AUC, Area under the curve; ROC, Receiver operator characteristic; IQR, Interquartile range; WBC, White blood cell.

The datasets generated or analyzed during the study are mainly human CT images, and are not publicly available due to the ethics policies of the hospital, but are available from the corresponding author on reasonable request.

YD and YZhang designed the research study. YD and YZhao performed the research. YZhang, YM and JY put forward suggestions for modification. YD, ZS, YM and JY analyzed the data. YD and YZhao wrote the initial draft of the manuscript. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

The study was carried out in accordance with the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of the institution (the First Affiliated Hospital of Soochow University) (Protocol No. 059, 2022), and the requirement for informed consent was waived due to the retrospective nature of the study.

This research received no external funding.

The authors declare no conflict of interest.