The inclusion criteria were as follows: (1) acute ischemic stroke (AIS) caused by intracranial large artery occlusion (LVO), include intracranial carotid artery (ICA), middle cerebral artery (MCA); (2) etiology of LVO was intracranial atherosclerotic disease (ICAD); (3) the onset-to-presentation time was within 24 h; (4) underwent endovascular treatment (EVT) and achieved successful recanalization that can stenosis degree of culprit artery; (5)aged 18 years or older; and (6) the pre-stroke modified Rankin Scale (mRS) score was between 0 and 2.

The exclusion criteria were as follows: (1) The cause of LVO was cardioembolic, dissection, moyamoya disease, vasculitis, or undetermined; (2) posterior circulation LVO; (3) tandem occlusion; (4) distal medium artery occlusion; (5) underwent stenting; (6) repeated intraprocedural re-occlusion or early re-occlusion post procedure; (7) lack of high-resolution magnetic resonance imaging (HR-MRI); (8) poor imaging included DSA/CTA/MRA quality to analysis; (9) absence of 90-day follow-up outcome data.

ICAD was defined as a significant fixed focal stenosis at the site of occlusion evidenced by final angiography or during endovascular treatment (3). Arteriosclerotic plaques were observed at culprit artery on high-resolution MRI performed after procedure.

Thrombectomy with retrieval stent or/and aspiration catheter was determined by neuro-interventionist. All treatment decisions during EVT, including emergent angioplasty, were made in real-time based on standard angiographic findings and were not influenced by the TS classification since this was a retrospective study. Performing emergent angioplasty was more depended on reperfusion status rather than residual stenosis degreee (13). It was performed only when high-grade residual stenosis co-existed with adequate antiplatelet coverage and at least one of the following conditions: (1) impaired distal flow (eTICI grade ≤ 2a) after mechanical thrombectomy; (2) re-occlusion or significant flow impared during the procedure; and (3) angiographic evidence of high re-occlusion risk (e.g., significant residual stenosis with delayed antegrade flow or poor collateral compensation).

The TS classification system was determined based on DSA images, with two aspects of the images being considered, namely, the existence of thrombus and the residual stenosis degree. The residual stenosis degree was measured in the absence of vasospasm or dissection appearance or residual thrombus after successful recanalization achieved by mechanical thrombectomy. Cases with dissection appearance or residual thrombus or vasospasm on DSA imaging were excluded from this study. The degree of stenosis was assessed using the Warfarin–Aspirin Symptomatic Intracranial Disease Study (WASID) method. According to this method, the degree of stenosis was classified as mild to moderate (<70%) or severe (70–99%) (14). The presence of a definite thrombus was defined in the following two ways: (1) Macroscopic thrombus visible on the extracted stent- retriever or aspiration catheter which was documented in medical record; and/or (2) distal embolism within the target vessel territory after a thrombectomy pass, as recorded in the procedural note and confirmed by subsequent DSA. The degree of stenosis was semi-automatically quantified using the built-in measurement tools on the DSA workstation.

The TS classification consists of three types (Figure 1):

The TS classification was determined by two independent neurologists, Dr. Pan and Dr. Lin, who had not participated in the writing of the criteria (group A), and two independent neuro-interventionists, Dr. A and Dr. B, who had participated in the writing of the criteria (group B). In cases of inter-observer disagreements, a consensus was reached to make the final determination. Dr. C is an attending neurologist with 7 years of experience in neurointerventional surgery. Dr. D is a resident neurologist with 6 years of experience in neurology and 3 years of experience in neurointerventional surgery. Dr. A and B are consulting neuro-interventionists, each having performed more than 500 operations per year.

All angiographic classifications, including the residual stenosis degree and the reperfusion score, were evaluated by the two independent raters (Z. N. P. and D. L. L.). Interobserver disagreements were resolved by consensus. The reperfusion statuses were measured using the extended Thrombolysis in Cerebral Infarction (eTICI) scale (15). An eTICI of 2b-3 was classified as successful reperfusion (15). Intraprocedural re-occlusion rate was high in ICAD-related LVO (16), so thrombectomy reperfusion should be observed at less 20 min during procedure (13). The first pass effect (FPE) is defined as the achievement of complete reperfusion (eTICI 3) after a single pass of thrombectomy (17).

Occlusion length was measured base on the venous phase of multi-phase CTA imaging (18). Perfusion maps and ischemic core volumes (CTP volumes) were auto-calculated by MIStar software (Apollo Medical Imaging Inc., Melbourne, VIC, Australia) (19). Delayed time (DT > 3 s) was defined as hypoperfusion brain tissue, while the ischemic core volumes were defined by relative cerebral blood flow (rCBF) < 30%. The mismatch volume was calculated as the volume of DT > 3 s minus the volume of rCBF < 30%, and the mismatch ratio was determined by dividing the volume of DT > 3 s by the volume of rCBF < 30%.

Clinical outcome was assessed by 3-month’s mRS, and favorable outcome was defined as mRS of 0–2, excellent outcome as mRS 0–1, and symptomatic intracranial hemorrhage (sICH), defined as any type of hemorrhage associated with an increase in the NIHSS score by ≥4 points (20).

Patients were categorized into three groups based on the TS classification criteria by four independent readers. In cases of disagreement, the relevant cases were re-examined, and any disputes were resolved through consensus. The interobserver reliability was assessed using the kappa (k) coefficient, with k values ranging from 0.70 to 0.89 indicating good consistency, and values between 0.9 and 1 representing excellent consistency. Quantitative variables were presented as the mean and standard deviation (SD) or the median and interquartile range, while categorical variables were expressed as the number of cases (N) and frequencies (%). For continuous variables with a normal distribution, Student’s t-test was conducted, whereas the Mann–Whitney U-test was utilized for continuous variables without a normal distribution or ordinal variables. Analysis of variance (ANOVA) was performed for comparisons among the three groups, followed by Bonferroni correction for pairwise comparisons. The chi-square (χ²) test was applied to categorical variables. Variables with a p < 0.05 were incorporated into the multivariable regression model to ascertain the correction between TS Subtype with EVT strategy and clinical Outcome, with odds ratios (ORs) being calculated. All statistical analyses were executed using SPSS 20.0 (IBM, Armonk, New York, USA) and R statistical software (Version 4.1.3), and a p < 0.05 was deemed statistically significant.

Anonymized data will be shared upon request to any qualified investigator.

Between January 2020 and December 2022, a total of 1,042 patients with intracranial LVO who underwent EVT were initially included in this study (Figure 2). After excluding patients for the following reasons, 105 patients (mean age 61.9 ± 10.5 years; male: female ratio 87:18) were finally enrolled: (1) cardiac embolism (n = 427), other determined causes (n = 56), and undetermined causes (n = 197); (2) tandem occlusion (n = 167); (3) distal medium artery occlusion (n = 32); (4) emergent stenting (n = 16); (5) failed to achieve successful reperfusion or re-occlusion during procedure or early after procedure (n = 19); (6) lack of high-resolution magnetic resonance imaging (HR-MRI) data (n = 51); (7) poor imaging quality (n = 4); and (8) absence of 90-day follow-up outcome data (n = 8).

The included patients were independently evaluated by two neurologists who had not participated and two neuro-interventionists who were involved in the development of the TS subtype diagnosis system. Prior to patient assessment, the four raters reached a consensus to strictly adhere to the classification system’s criteria without any exceptions. The thrombus retrieved by the device was physically documented and traceable in each patient’s record, while the degree of stenosis was semi-automatically quantified using the built-in measurement tools on the DSA workstation. So, excellent inter-observer agreement for TS classification was achieved in both individual groups (A and B) and across the combined cohort, with κ = 0.99. Among the 105 patients, the four raters concurred on the final diagnosis of the subtypes, except for one patient. This patient had a residual stenosis of less than 70% and no thrombus trapped by the stent. It was speculated that the thrombus could have been dissolved when the stent was unsheathed and intravenous tirofiban was administered. Consequently, this patient was classified as TS - type I.

Based on the TS classification criteria, 31.4% (33/105) of patients were classified as TS-type I (a representative case is depicted in Figure 3A), 43.8% (46/105) as TS-type II (Figure 3B), and 24.7% (26/105) as TS-type III (Figure 3C).

The baseline characteristics are presented in Table 1. The TS-type I group exhibited a larger baseline infarct core volume (11 mL vs. 3 mL vs. 4 mL, p = 0.008) compared to the other two groups. The TS-type III group had a shorter occlusion length (7.4 mm vs. 7.7 mm vs. 3.5 mm, p < 0.001) and a lower prevalence of HAS (24.2% vs. 32.6% vs. 0%, p = 0.015) than the other subtypes.

The radiological and clinical outcomes are summarized in Table 2. Compared to the other subtypes, TS-type III patients more frequently underwent emergent angioplasty (0% vs. 19.6% vs. 53.8%, p < 0.001) and had a lower rate of FPE (90.9% vs. 63.0% vs. 34.6%, p < 0.001). TS-type II patients were less likely to achieve eTICI 2c-3 reperfusion (97.0% vs. 82.2% vs. 100%, p = 0.017), required a higher number of thrombectomy passes (1.0 vs. 1.5 vs. 1.0, p = 0.003), needed more pass of thrombectomy (≥3 passes; 0% vs. 13% vs.0%, p = 0.021).

When compared with TS-type I patients, the likelihood of performing emergent angioplasty was significantly higher in TS-type III patients (aOR,44.0; 95% CI,5.1–369.8; p < 0.001) and TS-type II patients (aOR, 9.0; 95% CI,1.1–73.3; p = 0.047), after adjustment for age and NIHSS (Table 3).

Our study has several limitations. First, its retrospective single center nature may introduce uncontrolled bias, though prospective data collection helped minimize the influence, in future, multi-center studies are needed to validate the practicability and clinical significance of TS classification. Second, the TS subtype classification, based on EVT results (e.g., thrombus trapped by the stent and residual stenosis), is only applicable to patients who underwent successful EVT recanalization. Future studies will prospectively validate the TS system’s utility in guiding EVT strategies and use larger samples to identify clinical/imaging markers for predicting TS subtypes, thereby informing preoperative EVT decision-making. Third, bridging intravenous thrombolysis may dissolve thrombi at proximal or distal occlusion sites, potentially affecting TS subtype classification. In forthcoming work we would derive a predictive model for TS classification that incorporates clinical presentation and baseline brain imaging, thereby mitigating this limitation. Fourth, the relationship between TS classification and long-term clinical outcomes was not examined in the present study and warrants dedicated investigation in future work.

Once the TS classification has been established, three priority directions emerge for future research: (1) Prospective, interventional trials stratified by TS subtype, for example, head-to-head comparisons of optimal endovascular strategies for Type I versus Type III lesions; (2) Prospective cohort studies examining the association between TS subtype and short and long-term clinical outcomes. (3) Development and validation of predictive models that integrate clinical presentation and baseline brain imaging to pre-procedurally estimate TS subtype.