Combining Non-Contrast CT Signs With Onset-to-Imaging Time to Predict the Evolution of Intracerebral Hemorrhage

Go to:

INTRODUCTION

Intracerebral hemorrhage (ICH), which constitutes 10%–15% of all stroke-related events, accounts for approximately half of all deaths [1]. Hematoma expansion (HE) occurs in approximately 30% of patients with ICH within 6 h after symptom onset [2], is associated with deterioration [3], and is a promising therapeutic target [4]. Despite progress in curbing HE, clinical trial interventions have not significantly improved patient outcomes [5].

The conventional definition of HE emphasizes the growth of a hematoma in the brain parenchyma [6] while ignoring the dynamic changes in intraventricular hemorrhage (IVH) [4, 7, 8]. Notably, IVH growth (IVHG) is clinically not uncommon and has been linked to a poor prognosis [7]. Recent studies have suggested that the incorporation of IVHG into the HE definition can boost the predictive accuracy of outcomes in patients with ICH and have proposed the definition of revised HE (RHE) [9, 10].

Extensive evidence supports the positive effect of non-contrast CT (NCCT) signs on HE [11, 12]. Owing to some degree of overlap between NCCT signs, it is difficult to compare their predictive accuracy with regard to clinical outcome prediction [11]. Currently, few studies have conducted multivariable analyses to evaluate the hemorrhagic growth factors associated with all NCCT signs. The time from onset to the first imaging time (onset-to-imaging time [OIT]) is another factor closely associated with hemorrhagic growth after ICH [13]. A recent study indicated that the diagnostic performance of NCCT signs in predicting HE is affected by OIT. Hypodensities were identified in four out of five patients with HE within 2 h of onset, whereas irregular shapes performed better in late presenters [14]. However, they did not provide a comprehensive survey of all currently available NCCT signs. It is also unknown whether the predictive performance of NCCT signs for IVHG or RHE is limited by OIT.

To date, a comprehensive analysis investigating the association between hemorrhagic growth (HE, IVHG, and RHE) and NCCT signs while considering OIT is lacking. Therefore, the goal of this study was to examine whether the association between the above-mentioned signs and hemorrhagic growth was modified by OIT and to determine the predictive performance of NCCT signs for hemorrhagic growth when stratified by OIT.

Go to:

MATERIALS AND METHODS

Patient Selection

The Institutional Review Board granted a waiver/exemption owing to the use of de-identified data (IRB No. PJ2022-09-30, 2022-22, 2021-036, XYYYE20220081, YX2023-134, and GRYY-LL-KJ2022-K820). We retrospectively included consecutive patients with spontaneous ICH who were admitted to six stroke centers between January 2018 and August 2022. The inclusion criteria were 1) supratentorial ICH, 2) individuals aged > 18 years, 3) OIT ≤ 6 h, and 4) follow-up NCCT within 48 h of initial ictus. The exclusion criteria were 1) multiple ICH hematomas, 2) primary IVH, 3) surgical hematoma evacuation before the follow-up NCCT, 4) secondary ICH ascribed to trauma, vascular malformation, moyamoya disease, aneurysm, neoplastic disease, or a hemorrhagic transformation of a cerebral infarction, 5) prior oral/intravenous anticoagulant therapy, or abnormal coagulation at admission laboratory values (international normalized ratio [INR] > 1.7, platelet count < 50 × 109/µL), and 6) severe NCCT imaging artifacts during examinations.

Clinical Data

Variables included demographic, laboratory, clinical, and outcome characteristics. These included age, sex, alcohol consumption, smoking, medical history (ICH, cerebral infarction, hypertension, and diabetes mellitus), admission laboratory results (glucose, platelet count, and INR), systolic blood pressure, diastolic blood pressure, and baseline Glasgow Coma Scale scores. OIT was sorted into three tertiles as previously described: ≤ 2 h, 2–4 h, and 4–6 h [15]. Neurological function was assessed through outpatient visits or telephone consultations using the modified Rankin Scale (mRS) score (dichotomized as favorable [0-2] and poor [3-6]) at 90 days [9].

Neuroimaging

NCCT images were obtained using a 5-mm slice thickness. ICH volume measurements were performed using semi-automated segmentation with 3D Slicer software (version 4.11.2; https://www.slicer.org/). Two radiologists (with 10 and 5 years of experience) independently measured the ICH volume at baseline and follow-up NCCTs. When the difference in volume measurements between the two researchers showed ≥ 1 mL, the final volume was re-assessed by both researchers through mutual discussion to reach a consensus, and the mean of both was used as a final result. Indistinguishable boundaries between the parenchymal hematoma and IVH, if present, were manually segmented by a trained radiologist (5 years of experience). ICH locations, including the basal ganglia, thalamus, and lobar region, were assigned by a stroke neurologist (10 years of experience), as previously reported [16, 17].

The NCCT signs were classified into two categories based on density (hypodensities, swirl sign, black hole sign, blend sign, fluid level, and heterogeneous density) and shape (island sign, satellite sign, and irregular shape). The NCCT signs were evaluated according to the International NCCT ICH Study Group criteria (Fig. 1) [11]. An experienced radiologist and stroke neurologist (with 10 and 5 years of experience, respectively), who were blinded to the clinical and outcome information, independently reviewed the axial NCCT images. Any discordant opinions were resolved by consulting a third senior neuroradiologist (20 years of experience). Coronal and sagittal reconstructions, if necessary, were applied for differential diagnosis to discern imaging signs from the partial volume effect, particularly in terms of hypodensities and the black hole sign.

Fig. 1

Illustrative examples of signs (arrowheads). A: Hypodensities (any hypodense area was strictly enclosed within the hematoma, regardless of size, shape, or density). B: Swirl sign (rounded, streaklike, or irregular region of hypo- or isoattenuation compared with the brain parenchyma. Does not have to be encapsulated in the hematoma). C: Black hole sign (similar to the hypodensities but with a well-defined margin and a density difference > 28 HUs between the 2 regions). D: Blend sign (the difference between a relatively hypoattenuating region and a hyperattenuating area of the hematoma should be at least 18 HU, and with an identifiable border). E: Fluid level (there was a straight line separating the two densities within the hematoma, with hypoattenuating a region above and a hyperattenuating region below). F: Heterogeneous density (3 or more foci of hypoattenuation were found in the largest hematoma region, with the axial slice equivalent to a Barras density scale ≥ 3). G: Island sign (at least 3 scattered small hematomas all separate from the main intracerebral hemorrhage or at least 4 small hematomas some or all of which may connect with the hematoma). H: Satellite sign (a hematoma (diameter < 10 mm) was 1 to 20 mm away from the main hematoma). I: Irregular shape (there were 2 or more focal hematoma margin irregularities within the largest hematoma, with the axial slice equivalent to a Barras shape scale ≥ 3). HU = Hounsfield unit

* Click for larger image

* Download as PowerPoint slide

HE was defined as absolute hematoma growth > 6 mL or relative hematoma growth > 33% from the initial to follow-up NCCTs [18], whereas IVHG was defined as either absolute IVH growth > 1 mL by comparing the initial and follow-up NCCTs or any IVH on the follow-up NCCT but without the presence of IVH on the initial NCCT [7, 9]. RHE was classified into four types as follows: HE > 6 mL, relative hematoma growth > 33%, IVHG > 1 mL, or any IVH on follow-up NCCT [10].

Statistical Analysis

Due to the retrospective nature of this study, sample size calculations were not performed. Quantitative data were evaluated for normality (Kolmogorov-Smirnov test) and equal variance (Levene’s test) before further analysis. Normally and non-normally distributed data were expressed as mean (standard deviation) or median (interquartile range [IQR]), respectively. Categorical data are presented as counts and percentages. The inter-reader agreement between different measurements of ICH volume was assessed using the Bland-Altman analysis. The agreement on each NCCT sign was evaluated using Cohen’s kappa coefficient (κ). Baseline characteristics are summarized separately for NCCT signs and presented separately for those with and without signs.

The Chi-squared (χ2) tests (or Fisher’s exact tests) and the student’s t-tests (or Mann–Whitney U-tests) were applied appropriately to the univariable analysis of hemorrhagic growth. Multivariable logistic regression was performed to determine factors associated with three types of hemorrhagic growth (HE, IVHG, and RHE) by including factors showing P values ≤ 0.1 at the univariable analysis. The adjusted odds ratio (aOR) and 95% confidence intervals (CIs) were calculated.

The sensitivity, specificity, and positive predictive value (PPV) of the NCCT signs for the three types of hemorrhagic growth (HE, IVHG, and RHE) were also calculated and stratified according to OIT. All statistical tests reported two-sided P values, with P < 0.05 as the threshold for statistical significance. The R statistical software (version 4.0.3; https://www.r- project.org/) was used to perform all statistical analyses.

Go to:

RESULTS

Study Population

Overall, 6141 patients were screened, of whom 1488 (mean age, 61 years; 65% male) were eligible (flowchart in Fig. 2). Of these ICHs, data were complete for all variables except mRS (missing data in 235 [16%]). The median OIT was 2.29 h (IQR, 1.42–3.55 h), and the follow-up NCCT time was 19.00 h (IQR, 9.89–33.04 h).

Fig. 2

Study inclusion and exclusion flowchart. ICH = intracerebral hemorrhage, OIT = onset-to-imaging time, IVH = intraventricular hemorrhage

* Click for larger image

* Download as PowerPoint slide

Quantitative data regarding the agreement of ICH volume measurements between the two researchers are available in the Supplement (Supplementary Tables 1-4 and Supplementary Figs. 1-4). The rate of NCCT signs ranged from 5.85% to 44.96%. There was substantial or excellent agreement between the observers for the identification of hypodensities (κ = 0.84), swirl sign (κ = 0.81), black hole sign (κ = 0.81), blend sign (κ = 0.85), fluid level (κ = 0.88), heterogeneous density (κ = 0.78), island sign (κ = 0.79), satellite sign (κ = 0.84), and irregular shape (κ = 0.78). The characteristics of the patients with and without imaging signs are summarized in Supplementary Tables 5-9.

Univariable Analysis of the Association with Hemorrhagic Growth (HE, IVHG, and RHE)

HE, IVHG, and RHE were observed in 418 (28%), 303 (20%), and 583 (39%) of 1488 patients, respectively. Of 583 cases with RHE, the proportion for hypodensities, swirl sign, black hole sign, blend sign, heterogeneous density, island sign, satellite sign, and irregular shape was 60%, 62%, 28%, 41%, 30%, 34%, 48%, and 43%, respectively. A similar trend in other signs was found for IVHG, except for the blend sign and fluid level. Concomitantly, patients with any type of hemorrhagic growth (HE, IVHG, or RHE) had worse mRS scores at 90 days. A more detailed comparison between patients with and without hemorrhagic growth is presented in Table 1.

Table 1

Univariable analysis for HE, IVHG, and RHE

* Click for larger image

* Click for full table

* Download as Excel file

Multivariable Analysis of the Association with Hemorrhagic Growth (HE, IVHG, and RHE)

Regardless of which definition of hemorrhagic growth was applied, hypodensities were independent predictors of HE, IVHG, and RHE (aOR = 7.99 [95% CI = 4.87–13.40], aOR = 3.64 [95% CI = 2.15–6.24], and aOR = 7.90 [95% CI = 4.93–12.90], respectively), as was OIT. Both blend sign and island sign were associated with HE (aOR = 10.60 [95% CI = 7.36–15.30] and aOR = 2.75 [95% CI = 1.64–4.67]). The same was true for the blend sign and island sign on RHE (aOR = 10.10 [95% CI = 7.10–14.60] and aOR = 2.62 [95% CI = 1.60–4.30]). Additionally, heterogeneous density was also significantly associated with RHE (aOR = 0.61 [95% CI = 0.41–0.92]). The results are presented in Tables 2, 3, 4.

Table 2

Multivariable logistic regression for hematoma expansion

* Click for larger image

* Click for full table

* Download as Excel file

Table 3

Multivariable logistic regression for IVH growth

* Click for larger image

* Click for full table

* Download as Excel file

Table 4

Multivariable logistic regression for revised hematoma expansion

* Click for larger image

* Click for full table

* Download as Excel file

Performance of NCCT Signs for Hemorrhagic Growth Stratified by OIT

Table 5 summarizes the performance results. Of note, hypodensities had a poor predictive capacity for IVHG (PPV ≤ 0.41). The PPVs of most signs of RHE were higher than those of HE. Hypodensities, blend sign, and island sign showed high PPVs ≥ 0.80 when OIT was ≤ 2 h. The highest PPV for HE was observed with the blend sign when OIT was ≤ 2 h (0.84).

Table 5

Performance of NCCT signs for hemorrhagic growth stratified by OIT

* Click for larger image

* Click for full table

* Download as Excel file

Go to:

DISCUSSION

In this study, we comprehensively investigated the potential role of NCCT in the growth of three ICH types. Hypodensities, blend sign, and island sign were independent predictors of the HE and RHE. We also tested the performance of the above-mentioned three signs in the prediction of hemorrhagic growth, and the PPV for each sign on RHE was greater than 0.80 within 2 h.

Previous studies demonstrated that IVHG was strongly associated with patient outcomes after ICH [7, 8, 10]. Patients with ICH without high-risk HE, but with IVHG, if present, are not generally considered to have poor outcomes according to the conventional definition of HE [6]. This may have resulted in an underestimation of the risk of neurological deterioration. Our findings confirmed that patients in the RHE group had a more adverse prognosis, suggesting that the definition of RHE reflects the outcome more accurately than HE or IVHG alone after ICH. As an alternative, NCCT signs have been the major focus of HE prediction, but their role in RHE has been relatively spared [17, 18, 19, 20, 21, 22, 23]. Notably, obvious overlaps exist among different signs, leading to heterogeneity in findings [11]. Therefore, a comprehensive evaluation of RHE signs is necessary.

This study measured the predictive ability of hemorrhagic growth signs after stratification based on the OIT. The sooner the signs are identified, the more accurate the prediction. This suggests that these signs have a time-dependent effect on hemorrhagic growth. n patients with ICH within 2 h of onset, the PPVs of the signs of hemorrhagic growth were significantly higher than those in patients with an onset time > 2 h, particularly for the prediction of hypodensities, blend sign, and island sign on RHE. Similar to our study, Morotti et al. [14] reported a PPV of 0.43 for hypodensities in predicting HE within 2 h in patients with ICH. The latest results from a Mobile Stroke Unit Study found that HE was more frequent in the first 2 h, with 28% occurring within the first h after onset and 17% occurring within 1–2 h after onset [24]. The present study supplements the aforementioned studies. In ICH, early comprehensive care is important because some treatments are more effective when applied [2, 3, 4]. Hemostatic therapy is more appropriate for patients at high risk of ICH growth [25]. Thus, capturing earlier signs may considerably identify patients at high risk for RHE, which could result in an improved prognosis.

It is well established that overlap inevitably exists among NCCT signs [11]. This could have contributed to the heterogeneity among the studies. Nevertheless, our results confirmed that hypodensities, blend sign, and island sign were independent of each other and positively correlated with RHE. These compounds have extremely high potential for clinical applications. Various other technologies have also been used to predict hemorrhagic growth [26, 27]. The performance of the radiomics model proposed by Xia et al. [26] was significantly better than the NCCT signs obtained in our study. The advantage of radiomics lies in the quantitative analysis of image features that do not rely on subjective judgment [28]. However, radiomics analysis is limited by the repeatability and reproducibility of radiomics features, which is not conducive to large-scale promotion and application in clinical practice [29, 30]. Furthermore, artificial intelligence-assisted diagnosis and prediction of the disease has made significant progress in recent years [27, 31]. Ma et al. [27] developed an end-to-end deep learning method to automatically segment hematomas for HE prediction, with ResNet-34 achieving an excellent area under the curve (0.9267). This technique appears promising for HE prediction, but it remains largely unknown whether it is also effective for predicting RHE and requires further investigation in the future. Although the performance of conventional NCCT signs is inferior to that of the above technologies, these indicators have convenient and rapid characteristics in clinical practice as favorable prognostic predictors [11, 19, 32], especially in emergency settings or primary health institutions [33, 34, 35].

The strengths and limitations of this study are as follows. Its strengths include the recruitment of a cohort of representative and diverse patients with ICH from six centers and the relatively large number of cases of hemorrhagic growth at 2 h of ictus onset. However, retrospective observational studies are subject to limitations inherent to the methodology. First, the follow-up NCCT time was based on clinical decisions, which may have underestimated the prevalence and lowered the detection of hemorrhagic growth. Second, the incidence and risk profiles of ICH among different hospitals might have resulted in a selection bias. This factor also significantly increases the chances of targeted surgical interventions at senior stroke centers being provided earlier than necessary, particularly in patients with a larger hematoma volume or intraparenchymal hematoma with IVH. Third, the evaluation of NCCT signs relies on visual inspection, and potential interference may arise from image noise. Judging the signs of a small hematoma is also particularly challenging when facing the current spatial resolution of imaging equipment in different centers. Finally, patients without 90-day follow-up mRS scores were excluded. Therefore, the associations between NCCT signs and mRS scores were not analyzed further.

In conclusion, this study demonstrated an association between NCCT signs and hemorrhagic growth (HE, IVHG, and RHE) after ICH. Hypodensities, blend sign, and island sign were the best NCCT predictors of RHE when OIT was ≤ 2 h. NCCT signs may assist in earlier recognition of the risk of hemorrhagic growth and guide early intervention to prevent neurological deterioration from hemorrhagic growth.

Go to:

Supplement

The Supplement is available with this article at https://doi.org/10.3348/kjr.2023.0591.

Click here to view.(710K, pdf)

References

1. 1. GBD 2019 Diseases and Injuries Collaborators. Global burden of 369 diseases and injuries in 204 countries and territories, 1990-2019: a systematic analysis for the global burden of disease study 2019. Lancet 2020;396:1204–1222. * CrossRef 2. 1. Sheth KN. Spontaneous intracerebral hemorrhage. N Engl J Med 2022;387:1589–1596. * PubMed * CrossRef 3. 1. Greenberg SM, Ziai WC, Cordonnier C, Dowlatshahi D, Francis B, Goldstein JN, et al. 2022 guideline for the management of patients with spontaneous intracerebral hemorrhage: a guideline from the American Heart Association/American Stroke Association. Stroke 2022;53:e282–e361. * PubMed 4. 1. Morotti A, Boulouis G, Dowlatshahi D, Li Q, Shamy M, Al-Shahi Salman R, et al. Intracerebral haemorrhage expansion: definitions, predictors, and prevention. Lancet Neurol 2023;22:159–171. * PubMed * CrossRef 5. 1. Sprigg N, Flaherty K, Appleton JP, Al-Shahi Salman R, Bereczki D, Beridze M, et al. Tranexamic acid for hyperacute primary IntraCerebral Haemorrhage (TICH-2): an international randomised, placebo-controlled, phase 3 superiority trial. Lancet 2018;391:2107–2115. * PubMed * CrossRef 6. 1. Yogendrakumar V, Moores M, Sikora L, Shamy M, Ramsay T, Fergusson D, et al. Evaluating hematoma expansion scores in acute spontaneous intracerebral hemorrhage: a systematic scoping review. Stroke 2020;51:1305–1308. * PubMed * CrossRef 7. 1. Yogendrakumar V, Ramsay T, Fergusson D, Demchuk AM, Aviv RI, Rodriguez-Luna D, et al. New and expanding ventricular hemorrhage predicts poor outcome in acute intracerebral hemorrhage. Neurology 2019;93:e879–e888. * PubMed * CrossRef 8. 1. Li Q, Li R, Zhao LB, Yang XM, Yang WS, Deng L, et al. Intraventricular hemorrhage growth: definition, prevalence and association with hematoma expansion and prognosis. Neurocrit Care 2020;33:732–739. * PubMed * CrossRef 9. 1. Yogendrakumar V, Ramsay T, Fergusson DA, Demchuk AM, Aviv RI, Rodriguez-Luna D, et al. Redefining hematoma expansion with the inclusion of intraventricular hemorrhage growth. Stroke 2020;51:1120–1127. * PubMed * CrossRef 10. 1. Yang WS, Zhang SQ, Shen YQ, Wei X, Zhao LB, Xie XF, et al. Noncontrast computed tomography markers as predictors of revised hematoma expansion in acute intracerebral hemorrhage. J Am Heart Assoc 2021;10:e018248 * PubMed * CrossRef 11. 1. Morotti A, Boulouis G, Dowlatshahi D, Li Q, Barras CD, Delcourt C, et al. Standards for detecting, interpreting, and reporting noncontrast computed tomographic markers of intracerebral hemorrhage expansion. Ann Neurol 2019;86:480–492. * PubMed * CrossRef 12. 1. Law ZK, Ali A, Krishnan K, Bischoff A, Appleton JP, Scutt P, et al. Noncontrast computed tomography signs as predictors of hematoma expansion, clinical outcome, and response to tranexamic acid in acute intracerebral hemorrhage. Stroke 2020;51:121–128. * PubMed * CrossRef 13. 1. Al-Shahi Salman R, Frantzias J, Lee RJ, Lyden PD, Battey TWK, Ayres AM, et al. Absolute risk and predictors of the growth of acute spontaneous intracerebral haemorrhage: a systematic review and meta-analysis of individual patient data. Lancet Neurol 2018;17:885–894. * PubMed * CrossRef 14. 1. Morotti A, Li Q, Mazzoleni V, Nawabi J, Schlunk F, Mazzacane F, et al. Non-contrast CT markers of intracerebral hemorrhage expansion: the influence of onset-to-CT time. Int J Stroke 2023;18:704–711. * PubMed * CrossRef 15. 1. Dowlatshahi D, Brouwers HB, Demchuk AM, Hill MD, Aviv RI, Ufholz LA, et al. Predicting intracerebral hemorrhage growth with the spot sign: the effect of onset-to-scan time. Stroke 2016;47:695–700. * PubMed * CrossRef 16. 1. Falcone GJ, Biffi A, Brouwers HB, Anderson CD, Battey TW, Ayres AM, et al. Predictors of hematoma volume in deep and lobar supratentorial intracerebral hemorrhage. JAMA Neurol 2013;70:988–994. * PubMed * CrossRef 17. 1. Leasure AC, Sheth KN, Comeau M, Aldridge C, Worrall BB, Vashkevich A, et al. Identification and validation of hematoma volume cutoffs in spontaneous, supratentorial deep intracerebral hemorrhage. Stroke 2019;50:2044–2049. * PubMed * CrossRef 18. 1. Dowlatshahi D, Demchuk AM, Flaherty ML, Ali M, Lyden PL, Smith EE. VISTA Collaboration. Defining hematoma expansion in intracerebral hemorrhage: relationship with patient outcomes. Neurology 2011;76:1238–1244. * PubMed * CrossRef 19. 1. Wartenberg KE. Attacking intracerebral hemorrhage expansion: noncontrast CT markers on the horizon. Neurology 2020;95:615–616. * PubMed * CrossRef 20. 1. Cai J, Zhu H, Yang D, Yang R, Zhao X, Zhou J, et al. Accuracy of imaging markers on noncontrast computed tomography in predicting intracerebral hemorrhage expansion. Neurol Res 2020;42:973–979. * PubMed * CrossRef 21. 1. Quintas-Neves M, Marques L, Silva L, Amorim JM, Ferreira C, Pinho J. Noncontrast computed tomography markers of outcome in intracerebral hemorrhage patients. Neurol Res 2019;41:1083–1089. * PubMed * CrossRef 22. 1. Kim YS, Chae HY, Jeong HG, Kim BJ, Lee SU, Kang J, et al. Size-related differences in computed tomography markers of hematoma expansion in acute intracerebral hemorrhage. Neurocrit Care 2022;36:602–611. * PubMed * CrossRef 23. 1. Morotti A, Arba F, Boulouis G, Charidimou A. Noncontrast CT markers of intracerebral hemorrhage expansion and poor outcome: a meta-analysis. Neurology 2020;95:632–643. * PubMed * CrossRef 24. 1. Bowry R, Parker SA, Bratina P, Singh N, Yamal JM, Rajan SS, et al. Hemorrhage enlargement is more frequent in the first 2 hours: a prehospital mobile stroke unit study. Stroke 2022;53:2352–2360. * PubMed * CrossRef 25. 1. Nie X, Liu J, Liu D, Zhou Q, Duan W, Pu Y, et al. Haemostatic therapy in spontaneous intracerebral haemorrhage patients with high-risk of haematoma expansion by CT marker: a systematic review and meta-analysis of randomised trials. Stroke Vasc Neurol 2021;6:170–179. * PubMed * CrossRef 26. 1. Xia X, Ren Q, Cui J, Dong H, Huang Z, Jiang Q, et al. Radiomics for predicting revised hematoma expansion with the inclusion of intraventricular hemorrhage growth in patients with supratentorial spontaneous intraparenchymal hematomas. Ann Transl Med 2022;10:8 * PubMed * CrossRef 27. 1. Ma C, Wang L, Gao C, Liu D, Yang K, Meng Z, et al. Automatic and efficient prediction of hematoma expansion in patients with hypertensive intracerebral hemorrhage using deep learning based on CT images. J Pers Med 2022;12:779 * PubMed * CrossRef 28. 1. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017;14:749–762. * PubMed * CrossRef 29. 1. Zwanenburg A, Vallières M, Abdalah MA, Aerts HJWL, Andrearczyk V, Apte A, et al. The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping. Radiology 2020;295:328–338. * PubMed * CrossRef 30. 1. Berenguer R, Pastor-Juan MDR, Canales-Vázquez J, Castro-García M, Villas MV, Mansilla Legorburo F, et al. Radiomics of CT features may be nonreproducible and redundant: influence of CT acquisition parameters. Radiology 2018;288:407–415. * PubMed * CrossRef 31. 1. Cellina M, Cè M, Irmici G, Ascenti V, Caloro E, Bianchi L, et al. Artificial intelligence in emergency radiology: where are we going? Diagnostics (Basel) 2022;12:3223 * PubMed * CrossRef 32. 1. Nehme A, Ducroux C, Panzini MA, Bard C, Bereznyakova O, Boisseau W, et al. Non-contrast CT markers of intracerebral hematoma expansion: a reliability study. Eur Radiol 2022;32:6126–6135. * PubMed * CrossRef 33. 1. Jain A, Malhotra A, Payabvash S. Imaging of spontaneous intracerebral hemorrhage. Neuroimaging Clin N Am 2021;31:193–203. * PubMed * CrossRef 34. 1. Szentes T, Kovács L, Óváry C. New hospital structure in the twenty-first century: the position of level III (tertiary) neurological and stroke care in a changing healthcare system. Springerplus 2016;5:2039 * PubMed * CrossRef 35. 1. Wang X, Birch S, Zhu W, Ma H, Embrett M, Meng Q. Coordination of care in the Chinese health care systems: a gap analysis of service delivery from a provider perspective. BMC Health Serv Res 2016;16:571 * PubMed * CrossRef