STRENGTHS AND LIMITATIONS OF THIS STUDY

We collected a panel of imaging markers on multimodal CT and systematically analysed their predictive abilities for haemorrhagic transformation.

The median time from stroke onset to initial CT was 5 hours when the midline shift may be undetectable, resulting in the low positive rate of midline shift in this study.

This was a real-world study, so the enrolled patients were arranged for different neuroimaging examinations based on their medical conditions.

The study was a single hospital-based study, limiting its generalisability.

Introduction

Stroke is associated with significant mortality and high disability-adjusted life-years lost around the world.1 Ischaemic stroke is the most common subtype of stroke.2 As a devastating complication after acute ischaemic stroke (AIS), haemorrhagic transformation (HT) occurs as a natural history of AIS and is aggravated by some treatments with bleeding risk.3 The reperfusion therapy is an effective treatment for AIS recommended by the guideline,4 while the fear of HT especially symptomatic HT between patients and doctors hampers its wide application in clinical practice. Apart from the treatment-related HT, spontaneous HT is also not uncommon and is associated with poor clinical outcomes after AIS.5 Early prediction of HT after AIS is helpful for the stratification of patients at high risk of bleeding and the decision-making about further individual treatments with bleeding risk. To date, increasing studies have focused on the prediction of HT after reperfusion therapy,6–8 whereas the research investigating the overall HT and spontaneous HT is limited.

CT is an economical and fast imaging technology, making it essential for the diagnosis and management of AIS. Recently, a growing number of studies reported the association between various CT parameters and HT and showed promising predictive ability.9–11 However, the acquisition of several CT factors such as permeability surface and the exact lesion volume needs advanced technologies or special software, which may be unavailable in routine clinical assessment in most medical centres. The predictive value of multiple imaging factors in combination for HT has not been fully understood yet. In addition, the previous radiological research usually enrolled patients treated with reperfusion therapy. Limited studies have systematically addressed the imaging predictors of the overall HT and HT in patients who do not receive reperfusion therapy. Therefore, we aimed to (1) verify the predictive ability of individual imaging markers on multimodal CT for HT and (2) identify the key CT markers that can accurately predict HT, particularly spontaneous HT while maintaining easy and rapid assessment in the early stage of stroke.

Materials and methodsStudy population

Patients with AIS admitted to the Department of Neurology, West China Hospital, Sichuan University, within 24 hours after onset from 1 January 2016 to 30 June 2018, were prospectively and consecutively enrolled as the initial cohort. Out of this cohort, we included subjects who (1) were older than 18 years old and (2) underwent initial brain CT within 24 hours after admission and follow-up non-contrast CT (NCCT) or MRI scan within 30 days after admission. We excluded patients with (1) evidence of haemorrhage on the initial imaging or (2) non-diagnostic imaging quality.

Data acquisition

A standardised form was used to collect baseline data on age, sex, comorbidities (hypertension, diabetes mellitus, atrial fibrillation, hyperlipidaemia), current smoking and drinking, National Institutes of Health Stroke Scale (NIHSS) score, blood pressure and laboratory tests on admission, the Trial of ORG 10172 in Acute Stroke Treatment (TOAST) and treatments during hospitalisation (intravenous thrombolysis, endovascular treatments, antiplatelets and anticoagulation). Laboratory data on the first visit included leucocyte count and levels of blood glucose, triglyceride, cholesterol, low-density lipoprotein, high-density lipoprotein, serum creatinine, total bilirubin, alanine transaminase, aspartate aminotransferase and alkaline phosphatase.

Imaging protocol

The initial CT scans were performed within 24 hours after admission, and the follow-up neuroimaging examinations were completed within 30 days after admission. All patients underwent initial NCCT. Based on the clinical arrangement, some patients underwent NCCT, followed by CT angiography (CTA) and/or CT perfusion (CTP). All CT examinations were performed on a 64-slice scanner (Siemens). The imaging protocol for NCCT was as follows: axial scanning mode of the head, 120 kVp, 300 mAs, 5 mm slice thickness at 5 mm intervals. CTP was performed following the administration of 100 mL of nonionic iodinated contrast injected at a rate of 5 mL/s in a sequential algorithm in a section 4 cm wide, divided into eight layers, each 5 mm thick. The rotation time was 0.28 s, the voltage of the lamp was 70 kV and the anode current was 150 mA; each scan lasted 60 s. CTA was acquired using the helical scanning mode of the head. Before the scan, 100 mL of nonionic iodinated contrast was power injected into an antecubital vein at a rate of 5 mL/s. The remaining CTA parameters were 100 kVp, 86 mAs and 0.625 mm slice thickness at 0.625 mm intervals. The CTP source images were sent to a commercial workstation (Syngo MMWP (VE40A); Siemens Healthcare) for postprocessing, where the plane of symmetry was manually set; cerebral blood flow (CBF), cerebral blood volume and mean transit time (MTT) were generated. The CTA source images were sent to the same workstation and reconstructed to 20 mm thick axial and coronal maximum intensity projections.

The follow-up MRI scans were performed on a 3 T Siemens scanner. The imaging protocol for MRI was as follows: axial T1-weighted (repetition time = 1600 ms; echo time = 8.6 ms), T2-weighted (repetition time = 4500 ms; echo time = 105 ms) and fluid-attenuated inversion recovery images (repetition time = 6000 ms; echo time = 100 ms). The slice thickness was 5 mm, and the matrix size was 256 × 256 pixels.

Imaging analysis

For initial NCCT, four variables were reviewed: (1) the early hypodensity involved >1/3 of the middle cerebral artery (MCA) territory, (2) the Alberta Stroke Programme Early CT Score (ASPECTS), (3) the presence or absence of midline shift and (4) the presence or absence of hyperdense middle cerebral artery sign (HMCAS). ASPECTS was a 10-point scoring system with anatomical regions distributed over the MCA territory.12 The midline shift was defined as a midline shift of more than 5 mm at the septum pellucidum level or more than 2 mm at the pineal gland level.13 The HMCAS was defined if the lumen of the MCA appeared more hyperattenuated than adjacent or equivalent contralateral arteries but non-calcified.14

For CTA, the collateral circulation was assessed based on an ordinal scoring system proposed by Maas et al.15 The system is a 5-point score that compares collaterals on the symptomatic hemisphere against the contralateral hemisphere. The Sylvian fissure vessels and leptomeningeal collaterals were used as internal controls. The grades assigned were as follows: (1) absent, (2) less than the contralateral side, (3) equal to the contralateral side, (4) greater than the contralateral side and (5) exuberant. The poor collateral circulation was defined as grades 1–2.

For CTP, the presence or absence of infarct core and penumbra was measured. To eliminate the differences between individuals and the selection of regions of interest in both hemispheres, we used a relative value (ie, the ratio of the absolute perfusion data in the ischaemic hemisphere compared with the contralateral normal side) for data analysis. The infarct core was defined by CBF as less than 30% of the contralateral side.16 The penumbra was defined by MTT more than 145% of the contralateral side.17

Follow-up NCCT or MRI was examined to detect HT. HT was further categorised as haemorrhagic infarction and parenchymal haematoma (PH) using the European Cooperative Acute Stroke Study (ECASS) II criteria.18 HT was classified as symptomatic HT when the patients showed an increase of four points or more in the NIHSS score or asymptomatic HT when the patients showed no worsening neurological manifestations.19 HT in patients who did not receive reperfusion therapy (intravenous thrombolysis or endovascular treatment) was recorded as spontaneous HT. Two trained neurologists blinded to patients’ information independently read the above imaging signs. Disagreement was solved through discussion or advice from a third researcher.

Outcomes measurement

The primary outcome was measured as any HT on the follow-up imaging. The secondary outcomes were measured as the presence of PH, symptomatic HT and spontaneous HT.

Statistical analysis

Continuous variables were reported as mean ± SD or median with IQR, while categorical variables were reported as frequencies and percentages. Intergroup differences in continuous variables were assessed for significance using the Student’s t-test or the Mann-Whitney U test, while differences in categorical variables were assessed using χ2 or Fisher’s exact test. The independent relationships between imaging factors and HT were evaluated using multivariate logistic regression. Variables with p < 0.05 in comparing the baseline characteristics between patients with and without HT were entered into the multivariate analysis. Given that the comorbidities (ie, hypertension, diabetes mellitus, atrial fibrillation and hyperlipidaemia) may impact the incidence of HT,3 20 they were all adjusted as the confounders in the multivariate model. The results were shown as ORs with 95% CIs.

The predictive ability of each CT factor for HT was assessed via sensitivity, specificity, Youden index, positive predictive value (PPV) and negative predictive value (NPV). The area under the receiver operating characteristic curve (AUC) was also calculated as an index of the predictive ability of the combination of multiple imaging factors. A two-sided p < 0.05 was considered statistically significant. Statistical analysis was performed by using SPSS V.25.0 and GraphPad Prism V.8.0.

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

Results

During the study period, the initial cohort consisted of 812 patients, and 49 patients were disqualified for further analysis, of whom 22 patients did not have initial brain CT within 24 hours after admission, 21 did not have follow-up CT or MRI, 2 presented with haemorrhage on initial CT and 4 had poor quality of neuroimaging. Finally, a total of 763 patients (mean age 67.6 ± 14.0 years, 59.1% males) were included in this study. Among them, 118 patients (15.5%) received reperfusion therapy, of whom 71 patients (9.3%) were treated with intravenous thrombolysis alone, 40 (5.2%) with endovascular treatment alone and 7 (0.9%) with bridging therapy. The median time from onset to intravenous thrombolysis was 3 hours (IQR 2–4 hours), and the median time from onset to endovascular treatment was 6 hours (IQR 6–7 hours). Six hundred and ninety-nine patients (91.6%) used antiplatelets during hospitalisation, of whom 229 (30.0%) received dual antiplatelet therapy. One hundred and twenty-eight patients (16.8%) were treated with anticoagulation. The median time from onset to antiplatelets was 20 hours (IQR 13–29 hours), and to anticoagulation was 7 days (IQR 3–14 days). The median time from stroke onset to initial imaging was 5 hours (IQR 3–9 hours). All patients had initial NCCT on admission, 573 patients (75.1%) had CTA, 360 patients (47.2%) had CTP and 358 patients (46.9%) completed NCCT, CTA and CTP scans. The detailed information on each imaging marker is listed in online supplemental table S1. The median time from stroke onset to follow-up imaging was 4 days (IQR 3–8 days). Follow-up NCCT was performed in 105 (13.8%) patients, MRI in 223 (29.2%) patients and both modalities in 435 (57.0%) patients. One-hundred and forty-four patients developed HT, with 96 (12.6%) experiencing PH, 36 (4.7%) symptomatic HT and 108 (14.2%) spontaneous HT.

The baseline characteristics of the groups with and without HT are outlined in table 1. Compared with the patients without HT, the patients with HT were older (p = 0.01), had a higher NIHSS score (p < 0.001), a higher level of leucocyte count (p = 0.02) and lower systolic blood pressure on admission (p = 0.001), and were more likely to receive reperfusion therapy (p < 0.001). As for the comorbidities, the HT group had a lower rate of history of diabetes mellitus (p = 0.04) and a higher proportion of atrial fibrillation (p < 0.001) than the non-HT group. The rates of prior hyperlipidaemia (p = 1.00) and hypertension (p = 0.98) were similar in these two groups. The patients treated with antiplatelet agents in the HT group were less than those in the non-HT group (p = 0.001). There was a significant difference between the two groups in the distribution of TOAST classification (p<0.001). These variables with p < 0.05 and the four comorbidities were considered the confounders of the development of HT and were adjusted in further multivariate analysis. On baseline NCCT, the group with HT had higher proportions of the large extent of early hypodensity, midline shift and HMCAS, and presented with lower ASPECTS than the group without HT (all p < 0.001, figure 1). The patients with HT were more likely to show poor collateral circulation on baseline CTA and infarct core and penumbra on baseline CTP (all p < 0.001, figure 1).

Figure 1

The percentages of the CT markers in the group with or without HT. *p < 0.001. #Data were available in 573 patients. ##Data were available in 360 patients. ASPECTS, the Alberta Stroke Programme Early CT Score; HT, haemorrhagic transformation; MCA, middle cerebral artery.

Table 1

The baseline characteristics between patients with and without HT

After adjusting for age, history of diabetes mellitus, atrial fibrillation, hyperlipidaemia and hypertension, NIHSS score, systolic blood pressure, reperfusion therapy, antiplatelet treatment, leucocyte count and TOAST classification, the above CT signs were all the independent predictors for the occurrence of HT (table 2). The predictive value of each CT marker for HT is shown in table 3. The presence of poor collateral circulation had the highest Youden index of 0.56 for predicting HT among these imaging signs. Notably, although the midline shift had a low sensitivity (3.5%) for predicting HT, its specificity was up to 99.8%, with the PPV reaching 83.3%. The detailed characteristics of the patients with midline shift (n=6) are shown in online supplemental table S2. All patients with midline shift experienced a large extent of early ischaemic sign, and two patients received reperfusion therapy. Five patients with midline shift who developed HT were diagnosed with moderate-to-severe stroke (NIHSS > 8) on admission.

Table 2

Multivariable adjusted ORs of CT markers for HT

Table 3

Sensitivity, specificity, PPV and NPV of CT markers in discrimination between HT and non-HT

We further assessed the predictive ability of the combination of the imaging markers for HT. Given that CTA and CTP are not the routine neuroimaging scans for patients with AIS on admission in most stroke centres, and the acquisition of imaging data on CTA and CTP relies on specific imaging postprocessing techniques, we finally focused on the combination of the imaging signs on NCCT. As both the early hypodensity of the MCA territory and ASPECTS represented the extent of early ischaemic signs, we excluded ASPECTS from the analysis to avoid collinearity. Eventually, the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS was included in the final analysis. The AUC of the combination of the three markers for predicting HT was 0.80 (95% CI 0.75 to 0.84). Only three patients simultaneously presented with the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS on NCCT. Interestingly, the specificity and PPV of the coexistence of the three factors to predict HT reached 100%.

The predictive performance of the CT markers for the subtypes of HT was similar to the results of the overall HT. The presence of poor collateral circulation had the highest Youden index to predict PH (online supplemental table S3) and spontaneous HT (online supplemental table S4) among the CT factors. Regarding the prediction of symptomatic HT, ASPECTS ≤7 showed the highest Youden index, followed by penumbra and poor collateral circulation (online supplemental table S5). The AUCs of the combination of the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS for predicting PH, symptomatic HT and spontaneous HT were 0.81 (95% CI 0.76 to 0.87), 0.78 (95% CI 0.68 to 0.87) and 0.79 (95% CI 0.74 to 0.85, figure 2), respectively. Three patients who simultaneously presented with the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS all developed PH but not symptomatic HT. Among patients who were not treated with reperfusion therapy, the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS coexisted in only one patient, and the patient developed spontaneous HT afterwards. The specificity and PPV of the coexistence of the three factors to predict PH and spontaneous HT were also 100%.

Figure 2

AUCs of the combination of the early hypodensity >1/3 of the middle cerebral artery territory, midline shift and hyperdense middle cerebral artery sign for predicting the overall HT (A), PH (B), spontaneous HT (C) and symptomatic HT (D). AUC, area under the curve; HT, haemorrhagic transformation; PH, parenchymal haematoma.

As the necessity of CTA and CTP is increasing to evaluate endovascular treatment, we preliminarily analysed the predictive value of the combined multimodal CT parameters for HT. Considering that both the infarct core and penumbra were the surrogates of the extent of ischaemia on CTP, the infarct core was eliminated from the model to avoid collinearity. Finally, five parameters on multimodal CT (ie, the early hypodensity >1/3 of the MCA territory, midline shift, HMCAS, poor collateral circulation and penumbra) were incorporated to predict HT. The AUC of the combination of these five factors was 0.85 (95% CI 0.79 to 0.91) for predicting HT, 0.86 (95% CI 0.80 to 0.92) for PH, 0.86 (95% CI 0.78 to 0.94) for symptomatic HT and 0.86 (95% CI 0.79 to 0.93) for spontaneous HT.

Discussion

In the present study, we investigated the predictive ability of a panel of imaging markers on multimodal CT for HT after AIS. Seven CT factors were all independently associated with the high risk of HT. Although the midline shift had a low sensitivity, its high specificity was worthy of clinical concern when assessing the risk of HT. The early hypodensity >1/3 of the MCA territory, midline shift and HMCAS on NCCT was identified as the key CT factors for the early prediction of HT. The combination of the three markers showed good discrimination for the overall HT, PH and spontaneous HT.

Our findings confirmed and extended prior studies on the relationship between CT factors and HT after AIS.9 21 22 Both the early hypodensity of the MCA territory and ASPECTS were the independent predictors of HT, suggesting that the early ischaemic sign on CT plays an important role in the development of HT. This result was consistent with previous studies.12 23 24 We found that HMCAS was associated with a high risk of HT, similar to the study by Zou et al.9 It is notable that there were only six patients presented with midline shift on initial CT in our study. Although the midline shift was significantly associated with HT after adjustment, this independent association needs to be validated in the future due to the low positive rate of midline shift (<1%) and the wide range of 95% CI of the OR (1 to 135). Our results also showed that the poor collateral circulation on CTA could predict the occurrence of HT. When an artery is blocked, the compensatory blood flow derived from the collateral circulation is delivered to downstream regions of the brain. Therefore, the ischaemic tissue with poor collateral circulation may suffer from a large infarction or a worsening stroke, leading to the development of HT consequently.25 Regarding the imaging signs on CTP, we investigated the predictive ability of the infarct core and penumbra. These two factors are commonly used to evaluate the ischaemic degree and can be easily defined by computing the relative CBF and MTT. One study by Schaefer et al reported that all ischaemic tissue with a mean CBF ratio <18% developed HT,26 which was similar to our results in some sense. In another study, mean relative MTT with relative MTT >130% is an independent predictor of HT.27 Although the exact values of relative MTT and relative CBF used to predict HT in previous studies differed from this study, all our results revealed that the severity of ischaemia is closely associated with the development of HT.

To date, there are limited data on the detailed predictive value of the CT markers for predicting HT. Therefore, we systematically analysed the predictive ability of the CT signs for the overall HT and its subtypes. In this study, the presence of poor collateral circulation showed the highest Youden index to predict the overall HT, PH and spontaneous HT. However, the assessment of collateral circulation depends on CTA and needs neuroradiological training, which may be unavailable in the acute phase of stroke in community hospitals and, therefore, limits its generalisation in practice. Similarly, although some imaging factors on CTP also have an excellent ability to predict HT, the complex process during the examination and data analysis makes it difficult to apply for clinicians. NCCT is the first choice of neuroimaging examination for patients with AIS due to its convenience and economy. Thus, the imaging factors on NCCT with good predictive ability would be regarded as the key CT markers for the early prediction of HT. Among the imaging factors on NCCT, ASPECTS showed the highest Youden index for predicting HT. Increasing studies have confirmed the vital role of APSECTS in predicting HT. However, few reported the exact predictive value when assessing ASPECTS individually. Our study answered this question in detail. Even though the Youden index of the early hypodensity >1/3 of the MCA territory for predicting HT was a little lower than that of ASPECTS ≤7 (0.49 vs 0.55), its PPV was higher (53% vs 44.8%). Moreover, the measurement of the 1/3 rule is more manageable than ASPECTS, so the 1/3 rule may be more suitable for the early prediction of HT. The sensitivity of midline shift was only 3.5%, while its specificity was up to 99.8%, with the PPV and NPV being more than 80%. This finding indicated that although the absence of midline shift might be unable to rule out the risk of HT, the patients who have midline shift in the early stage of stroke are much more likely to experience HT subsequently. HMCAS also had a high specificity and NPV of nearly 90% to predict HT. Consequently, we identified the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS as the key CT factors of HT, and each of them had its unique advantages in the prediction of HT.

A systematic review of 55 studies concluded that the association between individual factors and HT was modest, and it was not a reliable way to assess the risk of HT based on the presence or absence of one variable.28 Several studies have demonstrated the usefulness of adding multiple biomarkers for predicting HT.29 30 However, the predictive significance of multiple imaging parameters has not been fully understood. We addressed this question and observed a good performance of the combination of the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS for predicting HT. Interestingly, the coexistence of the three factors had the specificity and PPV of 100% in this study, suggesting that patients who presented with the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS simultaneously on NCCT on admission would all develop HT during hospitalisation. The coexistence of the three factors had an extremely low false-positive rate to predict HT. This finding would help to identify patients at high risk of HT in the acute setting and guide further therapeutic strategies individually. In addition, we calculated the AUC of the combined multimodal CT parameters by adding the factors on CTA and CTP, which also showed a good predictive ability for HT and its subtypes. With the increasing application of CTA and CTP in the assessment of endovascular treatment, there may be a trend towards combining multimodal CT parameters to predict HT in clinical practice in the future.

Several clinical factors are associated with an increased risk of HT. We found that patients with HT were more likely to have a history of atrial fibrillation than patients without HT. In the study of 101 patients, patients with atrial fibrillation had more frequent PH than patients without HT,31 which was in accord with our findings. Like atrial fibrillation, diabetes mellitus is another important risk factor of HT.20 Although the rate of prior diabetes mellitus was lower in patients with HT than those without HT in this study, the blood glucose level on admission was higher in the HT group. This discrepancy may be because some HT patients were unaware they suffered from diabetes mellitus before this stroke. Öcek et al noted that serum low-density lipoprotein, triglycerides and total cholesterol levels were significantly lower in patients with HT than those without HT.32 A meta-analysis by Whiteley et al reported that a history of hypertension was associated with an increased risk of postalteplase HT.28 Therefore, we additionally included prior hyperlipidaemia and hypertension as the confounders that may impact the incidence of HT. It is well recognised that some treatments after stroke, particularly reperfusion therapy, increase the risk of HT. In the ECASS-II study, the incidence of HT reached 27% after thrombolysis, which was significantly higher compared with placebo (17.6%).19 A meta-analysis by Hao et al noted that the rate of HT was up to 35% after mechanical thrombectomy.33 In line with previous studies, we found that patients with HT had a higher proportion of reperfusion therapy than those without HT. Antiplatelets, particularly dual antiplatelet therapy, also pose a risk of HT.20 Notably, we found that the in-hospital use of antiplatelets was less in the HT group than in the non-HT group. The median time to initiate antiplatelets in the HT and non-HT groups was 28 hours and 18 hours, respectively. It suggests that some cases of HT might occur before the use of antiplatelets, resulting in a lower rate of antiplatelets in the HT group in the present study. Anticoagulants are commonly used for secondary prevention following cardioembolic stroke. Smythe et al reported that early initiation of anticoagulants was associated with a significant increase in HT.34 The patients with and without HT were similar in the proportion of anticoagulation in our cohort. This may be partly due to the relatively late initiation of anticoagulation after stroke in these patients (the median time 7 days).

Some limitations of this study should be noted. First, this was a single hospital-based study, which may limit the generalisability of our results to other populations. The multicentre studies preferably including different ethnicities are needed to verify our findings in the future. Second, the median time from stroke onset to initial CT in the present study was 5 hours when the midline shift may be undetectable, resulting in the low positive rate of midline shift and potential bias. Third, this was a real-world study. The enrolled patients were arranged for different neuroimaging examinations based on their medical conditions. Therefore, not all patients received CTA and CTP scans. In addition, both CT and MRI were used to detect HT in this study. MRI may be more sensitive to detect HT than CT. Nevertheless, their sensitivity for detecting PH is similar, and we did not use the special MR sequences such as susceptibility-weighted imaging and T2*-weighted gradient echo images which are very sensitive to detect microbleeds. What is more, about two-thirds of patients who received follow-up MRI had follow-up CT as well. The researchers had double checked the consistency of the imaging data on MRI and CT in these patients to determine the presence or absence of HT eventually. Thus, the different follow-up imaging technologies used in this study would not significantly affect the conclusions.

Conclusions

The present study indicates that the early hypodensity >1/3 of the MCA territory, ASPECTS ≤7, midline shift, HMCAS, the poor collateral circulation, infarct core and penumbra was individually associated with HT after AIS, independent of potential confounders. The high specificity of midline shift suggests the need to consider it as an imaging indicator when assessing the risk of HT. The early hypodensity >1/3 of the MCA territory, midline shift and HMCAS on NCCT was identified as the key imaging factors for the early prediction of HT due to their good predictive abilities and straightforward assessments. Incorporation of the three key factors had a good predictive performance for HT, particularly for PH and spontaneous HT. Our results revealed that using the simple imaging parameters on multimodal CT for early predicting HT after AIS could be an economical and practicable way, and the coexistence of the early hypodensity >1/3 of the MCA territory, midline shift and HMCAS might be a valuable index to identify individuals at high bleeding risk and guide further treatments.

Data availability statement

The data supporting the findings of this study are available on reasonable request from the corresponding author.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

The Biomedical Research Ethics Committee of West China Hospital, Sichuan University approved this prospective study (NO. 2016(339)), which was conducted in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. Informed content was obtained from all patients or their guardians.