Stroke is one of the main causes of death and disability in the world. In the USA, stroke has been one of the leading causes of death since 1924 (1). The global population comprised a significant number of elderly individuals, causing an increased incidence of age-related diseases such as stroke. Therefore, an accurate definition of the mechanism of stroke is crucial, as this will guide the selection of the most effective care and therapy (2, 3). There are many factors contributing to the overall outcome and prognosis in stroke patients. In addition to other predictors of stroke prognosis, such as neurological severity, age, other comorbidities, and epidemiological factors, the etiological subtype in ischemic stroke is an important factor for predicting the prognosis. For example, patients with lacunar infarcts have a better prognosis up to 1 year after onset than other stroke subtypes due to a different pathophysiological mechanism, although the long-term prognosis after lacunar stroke may not greatly differ from that after non-lacunar stroke. In contrast, patients with stroke of cardioembolic (CE) or large artery atherosclerosis (LAA) etiology tend to have worse prognoses for recovery than those with other ischemic stroke subtypes (4, 5). In this context, it is important to appropriately diagnose the patient’s etiological subtype of ischemic stroke not only for assisting treatment decisions by physicians but also for designing strategies for secondary stroke prevention (6).

There are several etiologic classification systems for ischemic stroke, such as the Harvard Stroke Registry (7), Stroke Data Bank (8), Trial of ORG 10172 in Acute Stroke Treatment (TOAST) system (9), Baltimore-Washington system (10), Causative Classification of Stroke System (CCS) (11), and atherothrombosis, small vessel disease, cardiac causes, and other uncommon causes (12). These are further subcategorized into the following two types according to the major approaches for the etiologic classification of stroke: phenotype and causative types. The phenotypic subcategory is determined based on an organization of abnormal test findings that may fit into major etiologic groups. Using this classification, patients could be categorized into more than one etiologic subtype. In contrast, the causative subcategory is assigned according to the most suspicious presumed mechanism of stroke. TOAST and CCS are the most commonly available examples of causative systems. However, with recent advances in diagnostic technology, multiple coexisting potential causes of ischemic stroke are frequently identified, which makes the selection of a causative subgroup a subjective process in clinical practice.

The TOAST classification system is primarily based on clinical features and also incorporates diagnostic information from various modalities including computed tomography, magnetic resonance imaging (MRI), transthoracic echocardiography, extracranial carotid ultrasonography, and cerebral angiography. This classification includes five subtypes of cerebral infarction: (1) LAA, (2) small vessel occlusion (SVO), (3) cardioembolism (CE), (4) strokes of other determined etiology (SOD), and (5) strokes of undetermined etiology (SUD) (Table 1). The SUDs are further sub-classified into “two or more causes,” “negative evaluation,” and “incomplete evaluation” (9). Although the TOAST system has been widely used in stroke-related research and in therapeutic studies, it has a few weaknesses. First, the reliability of TOAST classification is only fair to moderate with kappa values ranging from 0.42 to 0.54 (9, 13–16). Second, a result of “undetermined etiology” is often reached, partially because patients with more than two causes of stroke are classified as “undetermined etiology,” even though the individual potential causes of stroke are identifiable (6, 11, 17). Third, only strokes with more than 50% stenosis of the corresponding artery are classified as LAD (18). This means that cases with less than 50% stenosis are ignored, despite a potential to generate distal embolization. Fourth, embolic infarctions with vulnerable plaque with less than 50% stenosis are classified as SUD and aortic atherosclerosis is not described in the TOAST system even though CCS is classified as CE. Furthermore, advanced imaging techniques such as high-resolution vessel wall imaging and hemodynamic 4D-flow imaging, which can detect minimal changes in intracranial abnormalities as well as the underlying pathologic changes, can cause more confusion for classifying TOAST subtypes. Complementary classification systems are needed to resolve these problems.

In this study, we focused on the hyperintense acute reperfusion marker (HARM) in post-contrast fluid-attenuated inversion recovery (FLAIR) images in patients with acute ischemic stroke to resolve the problems of TOAST classification. The HARM is defined as a delayed enhancement of the subarachnoid or subpial space on post-contrast FLAIR images in patients with acute stroke and is believed to be associated with permeability changes in the blood–brain barrier (BBB) (19, 20). It has been reported that BBB breakdown, as indicated by HARM, was more common in patients who underwent reperfusion and was exacerbated by thrombolytic therapy (20). Recently, Ostwaldt et al. (21) demonstrated an association of HARM with early contrast material enhancement of the infarcted area in acute stroke and Lee et al. (22) reported that early HARM sign in acute stroke patients and most of HARM sign in that study was founded in the LAA and CE infarctions. They speculated that HARM may occur frequently in association with small embolic infarctions or reperfusion. We also recently observed relative high incidence of HARM in the specific cause of ischemic stroke. This study is based on a priori hypothesis: Embolic subtypes of acute stroke (e.g., LAA and CE) show more HARM compared with other subtypes of acute ischemic stroke based on conventional TOAST classifications. Therefore, the aims of this study were (1) to demonstrate the relationship between HARM and causative stroke etiology based on conventional TOAST classification and (2) to evaluate the relationship between HARM and stroke locations with respect to vascular territories and anatomic compartments. Based on the regions involved, HARM assessments were dichotomized into “regional pattern” and “randomized pattern.” The regional pattern described a HARM localization correlating with the same vascular territory as the acute infarction whereas the randomized pattern was assigned when a generalized contrast enhancement was detected in the ipsilateral or contralateral cerebral cortex regardless of acute infarction. Ultimately, we are going to evaluate the role of HARM for classifying stroke etiology based on TOAST classification. Understanding the clinical significance of HARM in acute stroke and incorporating these findings into the stroke subtype classification scheme may be a rational approach for acute stroke treatment and prevention.

In the current study, relative HARM incidence was significantly higher in the LAA subgroup than in the other subgroups (p = 0.001), while the SVO subgroup showed a relatively lower HARM incidence compared with the other subgroups (p < 0.001). Although CE itself showed no statistically significant difference in HARM incidence compared with the other etiologic subgroups, the relative HARM incidence in the overall embolic infarctions, including the LAA and CE subgroups, was higher compared with that in the other subgroups (p < 0.001). In terms of stroke location, the relative HARM incidence in the MCA territorial infarction and cortical infarction subgroups was higher than that in other vascular and anatomic territories. These findings suggest that HARM on post-contrast FLAIR images is associated with specific stroke causes, especially embolic infarction.

HARM is well known to represent the BBB permeability in ischemic stroke patients (19, 20). Ischemic stroke evokes a marked biphasic opening of the BBB. Early opening of the BBB takes place within the first 30 min of reperfusion and is followed by partial closing. The initial acute opening is described as a “hemodynamic” BBB opening resulting in cytotoxic edema (23, 24). Subsequently, a second delayed opening occurs 24–48 h later after reperfusion; this second opening results in vasogenic edema and is induced by “inflammation” (25). In the early opening phase, ischemic stroke restricts the delivery of oxygen and glucose to the brain tissue, which leads to a cascade of ischemic events and subsequent decrease in blood flow and ATP levels. Consequently, metabolic failure, ionic imbalance, and finally cellular death progresses within minutes (26). Following this, activated astrocytes and pericytes participate in the inflammatory cascade and release a number of substances that may be harmful to brain tissue (27). This wide variety of responses results in a number of effects including the loss of tight junctions between endothelial cells and finally results in the increase in BBB permeability (28).

We believe that HARM is more visible in the embolic infarctions, as compared with other TOAST subtypes, because of a lack of tolerance to the sudden hemodynamic changes in embolic infarction cases. High ischemic tolerance has been noted in atherosclerotic thromboembolic infarction rather than in embolic infarction and in old age patients rather than in young age stroke patients. Furthermore, the high recanalization rate in cardioembolic infarction is attributable to HARM incidence. In 4.7–12% of cases, cardioembolic infarction showed a sudden onset of neurologic deficit followed by dramatic improvement of the initially severe neurologic symptoms. This may be due to distal migration of the embolus followed by recanalization of the occluded vessel (29). These sudden hemodynamic changes might contribute to the relatively higher incidence of the HARM in embolic infarction.

There are other findings to back up these theories in the current study. Within the LAA etiology group (n = 75), 30 patients were HARM positive and the remaining 45 patients were HARM negative. We speculate that these results may be due to different pathophysiological mechanisms. The pathophysiological mechanisms of LAA have been proposed as (1) hypoperfusion of the involved vascular territory, (2) unstable plaque with artery-to-artery embolism, and (3) hypoperfusion-mediated microembolus formation. It is difficult to infer which mechanism may be responsible for an individual patient’s condition in clinical practice (30). We hypothesized that the presence of HARM in acute stroke patients diagnosed with LAA etiology may be attributable to the artery-to-artery embolism, whereas the absence of HARM may be attributable to hypoperfusion.

Other studies have also found a higher incidence of HARM in MCA territorial infarction and cortical infarction compared to infarction at other vascular territories and anatomic compartments. Using high-resolution phase-contrast MRI, most of the IA blood flow was found to be distributed to the MCA (31). We hypothesized that the higher incidence of HARM in MCA territorial infarction may be attributable to the embolic stroke etiology in terms of IA flow dynamics. This result is in agreement with the findings of Garcia et al. (32), which show that most emboli lodge in the MCA distribution because 80% of the blood carried by large neck arteries flows through the MCA. Regarding stroke location according to the involved anatomic compartment, our study demonstrated a higher incidence of HARM with cortical infarction. This result is also in line with a previous study in which acute stroke lesions in the cortex and subcortical area were associated with CE and LAA infarction (33).

However, there were no significant differences in the relative HARM incidence between the CE subgroup itself and other etiologic subgroups, contrary to our priori hypothesis. This result could be explained by the relatively high proportion of “undetermined etiology” cases in the current study. Under the conventional TOAST classification, the presence of atrial fibrillation combined with other possible stroke etiologies belongs to the category of “undetermined etiology.” However, we speculated that the different histological compositions of the cardiac origin emboli and artery-to-artery emboli from large artery disease could have produced the differences in relative HARM incidence between artery-to-artery emboli and cardiogenic emboli. Emboli of cardiac origin are derived from blood stasis in the posterior left atrium and appendage and are mainly composed of fibrin and red blood cells (RBCs). In contrast, large artery-to-artery emboli are derived from atheromatous plaques in the aorta and carotid artery at the site of ulcerated plaque and are mainly composed of platelets and fibrin (34, 35). We hypothesized that large artery-to-artery emboli are more likely to induce BBB disruption compared to cardiac emboli, as RBC-rich emboli are promptly degraded by an intrinsic thrombolytic cascade. Notably, the composition of emboli may play a role in the success rate of thrombolysis with tPA since RBC-rich thrombi are known to be more vulnerable to thrombolysis compared to platelet-rich thrombi (36, 37).

In the current study, the majority of the SVO etiologic groups did not show HARM (p < 0.001). In a previous study, BBB dysfunction, indicated by IV injected contrast leakage, was demonstrated in white matter disease and dementia (38, 39), as well as in lacunar infarction (40). However, in the current study, BBB disruption, as indicated by HARM, was relatively low in the SVO etiologic group. In terms of our hypothesis that embolic sources of acute infarction cause more HARM than other stroke etiologies, our results are in line with that of a previous study, which demonstrated that cardiac or artery-to-artery embolism is unlikely to be a frequent cause of most lacunar strokes (41).

In contrast, among the SVO etiologic group (n = 88), six patients showed HARM signs. This finding may be attributable to other causes or TIA rather than acute lacunar infarction itself. Recent studies have shown that TIA and imaging-negative strokes might be HARM positive and associated with focal BBB disruption (40, 42). Lee et al. (43) suggested that a finding of HARM without diffusion abnormalities represents a specific type of TIA with favorable short-term neurologic outcomes, compared with imaging-negative stroke. The prevalence of SVO may be artifactually elevated to a misclassification of stroke subtypes (18). SVO usually presents as single subcortical infarction. Furthermore, recent studies have shown that single subcortical infarctions frequently are caused by branch occlusion associated with parental artery atherosclerotic plaque, arteriosclerotic proximal small vessel disease, or other embolic sources, rather than lipohyalinotic small artery disease (44).

We analyzed the topographic pattern of HARM with regard to stroke locations. Of the 67 patients in the HARM-positive group, 55 patients showed a linear cortical or sulcal enhancement along the same vascular territory near the acute infarction (regional pattern), whereas 12 patients showed ipsilateral or contralateral enhancement regardless of the acute infarction vascular territory (randomized pattern). Of note, LAA showed a significantly higher association with the regional HARM pattern relative to the other subgroups (p = 0.002). Anatomically, the perivascular space is filled with interstitial fluid or CSF, encompassing the surface and penetrating the blood vessels of the brain. It has been demonstrated that systolic arterial pulsation is derived from CSF motion in the perivascular space (45). We speculate that this pulsatile CSF motion, termed “extracerebral CSF dynamic,” may influence the extent of contrast spillage indicated by the HARM pattern.

Blood–brain barrier opening indicated by a HARM sign is distinct from enhancement of the leptomeninges (46, 47) and from the parenchyma enhancement that is observed during the subacute stage of cerebral ischemia (48). A recent study reported that acute ischemic stroke patients with HARM showed a higher relative signal intensity compared with HARM-negative patients on FLAIR images in the adjacent brain parenchyma corresponding to the DWI-positive tissue, which is likely due to parenchymal enhancement (21). This result indicates that HARM does not only represent a contrast leakage from the pial system into the CSF space. However, it is not yet clear whether HARM is also associated with extravasation of contrast agent into the parenchyma.

In previous HARM studies, the commonly used gadolinium-containing contrast agents were Magnevist (20, 49) or Gadovist (50). In the present study, we administered Gadovist at a dose of 0.2 mmol/kg. Ostwaldt et al. (51) revealed that increasing the dosage of Gadobutrol is also associated with increased HARM incidence in acute stroke patients. Post-contrast FLAIR images obtained at 5 min after contrast administration showed what was described as “early” HARM in a previous study (22). It should be emphasized that these “early HARM” signs are different from those of other previous studies in which HARM was depicted on follow-up unenhanced FLAIR images at 2–48 h after contrast administration in the initial MRI. In the current study, HARM was detected in 25.38% of acute ischemic strokes compared to the 5.5% reported by Lee et al. (22). This discrepancy might be attributable to the treatments applied before MRI image acquisition in the present study. In addition, the interval between symptom onset and MRI acquisition varied from 0 h to 7 days in the current study compared to 0–24 h in the study by Lee et al. (22). However, the incidence of HARM evaluated at 2–48 h after contrast administration was 33–43% (20, 49, 50).

Although the HARM and risk of hemorrhagic transformation associated with sudden hemodynamic changes after carotid artery stenting or temporary balloon occlusion of the carotid artery have been widely studied in previous studies, there is little information on the relationship between the HARM and stroke etiology. The current study demonstrated that the HARM is associated with embolic infarction. This hypothesis is supported by our findings of a relatively higher incidence of the HARM in MCA and cortical infarction compared with infarction at other vascular territories and anatomic compartments. The results from this study may be translated into clinical practice and management, suggesting that evaluation of embolic sources needs to be done thoroughly, especially in HARM-positive strokes. It has been demonstrated that the HARM is not a specific radiologic finding in acute stroke and can be understood as a common finding in other BBB-disrupting conditions, such as postcarotid stent, postcardiac surgery, and traumatic brain injury (22). However, understanding the clinical significance of HARM signs in acute stroke and incorporating these findings into a stroke subtype classification scheme may be a rational approach for acute stroke treatment and prevention. Moreover, strict application of the conventional TOAST criteria could lead to categorization of the majority of strokes in the “undetermined etiology” category (52). Elucidating the most likely mechanism among coexisting potential stroke causes would therefore be useful. We expect that with the recent advances in stroke imaging, including the HARM on post-contrast FLAIR, it is possible to establish the correct classification of stroke.

With recent advancements in imaging, high-resolution MRIs can detect small plaques causing infarction, even in patients with normal vascular findings, which would not be classified as LAA etiology under previous classification systems (18). Despite the practical issues in using MRI for acute stroke evaluation, MRI can be useful for the accurate etiologic diagnosis of not only stroke but also stroke prognosis.

This study has a number of limitations. First, the sample population may be biased, as it is a retrospective single-center study from a single region. This cohort has relatively mild stroke with median NIHSS around 4–5. Moreover, the distribution of stroke subtypes in our study showed a higher incidence of SVO than the population average in Korea (18). Second, symptom onset varied between the early hyper-acute (0–6 h), late hyper-acute (6–24 h), and acute (24 h to 1 week). The third drawback of this study is the limited evaluation of the most probable cause of stroke etiology in the undetermined group based on conventional TOAST classification. Numerous patients in the current study had a stroke of undetermined etiology, comprising up to 25%. Fourth, IV tPA and IA thrombolysis were performed in very few patients at varied timings; therefore, we could not identify the relationship between the HARM and applied treatments. Fifth, the infarct volume is not supplemented due to we included both single and multiple infarctions in this study. Further studies employing larger datasets with prospective study design and pathologic confirmation are needed to confirm the results in current study.

In conclusion, the presence of the HARM was significantly associated with embolic infarctions including LAA and CE. Conversely, SVO was exclusively associated with the absence of the HARM. Second, MCA and cortical infarction showed a more pronounced HARM compared to infarctions at other vascular territories and anatomic compartments. According to the results in the current study, we speculate that the presence of HARM on post-contrast FLAIR images was associated with specific stroke causes especially in embolic causes. However, future studies are required to confirm these preliminary findings and to determine if HARM can provide independent information on stroke causes.

This retrospective study was approved by our institutional review board. Informed consent was waived because only imaging findings and demographic data were retrieved from the picture archiving and communication system (PACS) server and medical records.

HC and H-GK collected the data and wrote the article. KL conceived and designed the study and wrote the article. EK, WC, BK, SH, and D-IC critically revised the article for important intellectual content.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.