|Home | About | Journals | Submit | Contact Us | Français|
Intracerebral hemorrhage volume (ICHV) as a percentage of total brain volume (TBV) is a strong predictor of outcome in childhood intracerebral hemorrhage (ICH) with ICHV/TBV>2% associated with functional impairment. We aimed to determine whether easily performed approximations of ICH and brain volume can accurately and reliably stratify ICH by size.
CT scans of 18 children with spontaneous ICH were independently reviewed by 4 neurologists. ICHV as a proportion of TBV was estimated as ABC/XYZ, expressed as a percentage. A = largest axial hemorrhage diameter; B = largest axial diameter perpendicular to A on the same slice; and C = hemorrhage vertical diameter. Similarly, X = largest midline axial diameter of supratentorial brain; Y = largest axial diameter perpendicular to X; and Z = brain vertical diameter. Inter-rater reliability was measured with intraclass correlation coefficients (ICC). ICHV and TBV were measured using computer-assisted manual segmentation tracings to establish criterion validity. Each ICH was classified as small (≤2%TBV) or large (>2%TBV).
Estimates of ICHV, TBV, and ICHV/TBV using the ABC/XYZ method had outstanding inter-rater reliability (ICC 0.95–0.99). These estimates were highly correlated with volumetric measures (R2 0.77–0.96). Sensitivity of the ABC/XYZ method for identifying an ICHV >2% TBV was 100% (95%CI: 89–100%). Specificity was 95% (95%CI: 83–99%).
The ABC/XYZ method accurately and reliably estimates ICHV and TBV in children. These values can be used to quickly and easily approximate ICHV as a percentage of TBV, which has important prognostic implications.
The volume of intracerebral hemorrhage (ICH) is well established as a predictor of outcome in adults1–3. In children, ICH volume (ICHV) also predicts short-term outcome though it must be adjusted for the total brain volume (TBV) to account for the varying head sizes during development4, 5. For research purposes, ICHV and TBV measurements have utilized manual segmentation tracing, which can take up to several hours per patient. A more clinically useful technique would allow treating physicians of different training levels to rapidly and easily assess ICHV as a percent of TBV in the acute hospital setting with good reliability and accuracy. The elliptical approximation method ABC/2 has been successfully used to estimate ICHV in adults6, 7. We sought to assess the reliability and validity of a similar formula to estimate both ICHV and TBV in pediatric ICH.
With institutional review board approval, a prospective consecutive cohort of patients with spontaneous ICH was identified from a single large tertiary care children’s hospital stroke registry between January 2006 and March 2009. Inclusion criteria for this study were: age 4 months to 18 years with spontaneous supratentorial ICH confirmed by CT. Exclusion criteria were: infratentorial hemorrhage, isolated subarachnoid hemorrhage, isolated intraventricular hemorrhage, hemorrhage due to brain tumor, and hemorrhage due to hemorrhagic conversion of either arterial ischemic stroke or cerebral venous sinus thrombosis.
Head CT scans of 18 pediatric patients with acute spontaneous ICH were independently reviewed by 4 neurologists of varying experience: 1 pediatric stroke neurologist (RI), 1 adult stroke neurologist (SM), 1 pediatric stroke fellow (LB), and 1 adult neurology resident (MM). The size of the intraparenchymal ICH was estimated by ABC/2, wherein A was defined as the largest axial diameter of the hemorrhage evident on any slice, B was the largest axial diameter perpendicular to A on the same slice, and C was the vertical diameter (slice thickness multiplied by number of slices where hemorrhage was visible), as previously described by Broderick et al. 6. Similarly, TBV was estimated by XYZ/2, where X was defined as the largest anterior-posterior axial diameter of the supratentorial brain, Y as the largest axial diameter perpendicular to X, and Z as the vertical diameter (slice thickness multiplied by number of slices from foramen magnum to vertex). ICH volume was then expressed as a proportion of TBV by the formula (ABC/2)/(XYZ/2) which reduces to ABC/XYZ, and was then expressed as a percentage. A representative example of one patient’s ABC/2 and XYZ/2 measurements is depicted in Figures 1, top panel and bottom panel, respectively.
Manual segmentation tracing to assess ICHV and TBV was performed on all CT images by a single rater (LB) using a computer program, ITK-SNAP, that is available online (www.itksnap.org) 8. Similar techniques have been used reliably in both childhood ICH and in measurements of pediatric brain structures4, 8. On segmentation imaging, TBV included the cerebral hemispheres, cerebellum, and brainstem. Intraventricular hemorrhage volume was not included in the ICH volume measurements, and ventricular volume was not included in the TBV measurements. Values obtained by manual segmentation tracing were considered the reference criterion.
All measurements performed by both ABC/XYZ and manual segmentation were categorized as small if ICH/TBV was < 2% or as large if ICHV/TBV was > 2%. The threshold of 2% was chosen based on the observation that 30cc in adult ICH is an important and validated ICHV threshold for predicting outcome1–3, 6. An ICHV of 30 cc represents approximately 2% of the TBV of an average adult brain of about 1400 cc. In addition, hemorrhages >2% TBV are associated with functional impairment in childhood ICH5.
Statistical analysis was performed using STATA version 10.0 (Stata Corp, College Station, TX). Inter-rater reliability was measured with intraclass correlation coefficients (ICC) with the use of ANOVA. An ICC of 1 demonstrates perfect reliability, and an ICC> 0.8 represents excellent reliability9. Criterion validity was established by comparison of estimated values by ABC/XYZ relative to those obtained with computer-assisted manual segmentation volume measurements. Wilcoxon signed rank tests were used to determine whether the estimated and reference measurements differed significantly. Linear regression was performed to determine the correlation of the estimated measures with the volumetric measures, clustered by patient since the four estimations for each subject were not independent; the parameter R2 was reported as the result. For sensitivity and specificity, hemorrhages were dichotomized as either small (≤2%TBV) or large (>2%TBV). Sensitivity and specificity for detecting large ICHs were calculated along with their 95% binomial confidence intervals.
Head CT scans from 18 pediatric patients, median age 9.5 years (range 4.5 months to 16.6 years), were evaluated. The head CT scans were performed at a median of 9.6 hours (interquartile range [IQR]: 2.3–24 hours). Causes of ICH were arteriovenous malformation in 10, cavernoma in 3, aneurysm in 1, hematologic abnormality in 2, and unknown in 2. Hemorrhage locations were frontal in 8, temporal in 3, parietal in 3, frontotemporal in 1, occipital in 1, basal ganglia in 1, and thalamic in 1. Using computer-assisted manual segmentation volumetric measurements, median ICHV was 23 mL (IQR: 4–37mL), median TBV was 1286 mL (IQR: 1188–1343mL), and median ICHV/TBV was 1.6% (IQR: 0.3–2.8%). Using the ABC/XYZ estimation, median ICHV was 23 mL (IQR: 5–38mL), median TBV was 1192 mL (IQR: 1030–1296mL), and median ICHV/TBV was 1.8% (IQR: 0.4–3.0%).
Reliability of the ABC/2, XYZ/2, and ABC/XYZ measurements was high. The ICC for the estimation of ICH volume by the ABC/2 method was 0.99 (95% confidence interval [CI]: 0.99–1.0). The ICC for TBV by XYZ/2 was 0.95 (95% CI: 0.92–0.99). The ICC for the estimation of ICHV expressed as a proportion of TBV by the equation ABC/XYZ was 0.99 (95% CI: 0.99–1.0).
There was no difference in the estimates of ICHV by ABC/2 compared to volumetric measurements (median difference 0 mL [IQR: −1 to 1 mL]; p=0.62 by signed rank test), but the XYZ/2 method underestimated the TBV by a median of 91 mL (IQR: 32 to 145 mL; p<0.001 signed rank test). The net impact was a small but statistically significant difference in the ICHV/TBV (median difference 0.1% [IQR: 0.0–0.3%]; p=0.0005). In regression analysis, the R2 for the ABC/2 approximation and volumetric ICHV measurement was 0.97, p<0.001 (Figure 2a). The R2 for XYZ/2 and volumetric TBV analysis was 0.77, p<0.001 (Figure 2b). The R2 for ABC/XYZ and volumetric analysis of ICHV as a proportion of TBV was 0.96, p<0.001 (Figure 2c).
Using the reference criterion, ICHV/TBV was ≤2% in 10 patients, >2% – 4% in 6 patients, and >4% in 2 patients. When ABC/XYZ was dichotomized as ≤2% vs. >2%, 70 of 72 measurements (18 ICHs × 4 raters) were classified correctly compared to the reference criterion. The sensitivity of the ABC/XYZ method for identifying an ICHV that was >2% of TBV was 100% (95% binomial CI: 89–100% and specificity was 95% (95% binomial CI: 83–99%).
This study demonstrates that the elliptical approximation method ABC/XYZ may be used to estimate ICH volume in children as a proportion of total brain volume with excellent reliability and validity. The XYZ/2 approach tends to underestimate TBV compared to volumetric measurements, presumably because the pediatric brain is not truly elliptical. This limitation could result in overestimation and misclassification of some small hemorrhages as large, particularly those just below the 2% ratio. The net effect is that this method maximizes sensitivity with minimal loss of specificity at this threshold. Nevertheless, despite this potential pitfall, 70/72 categorizations were correct.
Several other limitations of the ABC/XYZ method should be noted. In several adult studies, the ABC/2 method for hemorrhage volume measurement has been shown to overestimate hemorrhage volume for both irregularly-shaped hemorrhages and for large hemorrhages10–12. With 18 subjects, sub-analyses of the effect of various hemorrhage shapes or of various hemorrhage sizes could not be performed. Additionally, this method was not tested in patients with ICH due to brain tumors. The ABC/XYZ equation might not work as well in this setting since it can be difficult to differentiate hemorrhage from underlying tumor in some cases. Furthermore, this method was not tested in patients with hemorrhagic transformation of arterial ischemic stroke, hemorrhagic transformation of cerebral sinus venous thrombosis, primary intraventricular, subarachnoid, or epidural hemorrhage. The ABC/XYZ method may not be valid for these hemorrhages since their shapes are often more irregular and depart from true ellipses.
Despite the potential limitations of this method, the calculation of ABC/XYZ is easily performed in just minutes using measurement tools available in standard image viewing software and a calculator, and can be implemented by raters of different experience with minimal training. Until now, there has been no accurate method to evaluate hemorrhage size as a percent of total brain volume quickly in the acute setting. Therefore information of critical prognostic value has been unavailable to clinicians and parents. The ABC/XYZ formula will be useful to clinicians for both outcome prediction in the acute setting and for choosing appropriate participants for future acute trials. The ABC/XYZ method should be validated in a separate cohort to determine widespread applicability.
We acknowledge The Children’s Hospital of Philadelphia Research Institute. It has generously supported Dr. Ichord’s development of the Pediatric Stroke Program at The Children’s Hospital of Philadelphia as well as the development and maintenance of The Children’s Hospital of Philadelphia Stroke Registry.
We also acknowledge the contributions of our research assistants, Alana M. Feiler, Stephen J Tang BS, and Grady H. Hedstrom, BS for their help with image formatting. We would like to thank Justine Shults, PhD for her advice on statistical methods.
Dr. Beslow is supported by grants from the NIH (T32-NS007413) and from the L. Morton Morley Funds of The Philadelphia Foundation.
Dr. Ichord is supported by grants from the NINDS (RO1-207320801, K23-NS062110).
Dr. Mullen is supported by a grant from the NHLBI (K12-HL083772).
Dr. Licht is supported by grants from the NINDS (K23-NS52380), Dana Foundation and Wolfson Fund for Neurological Research.
Dr. Smith is supported by a grant from the NINDS (K12-NS049453).
Dr. Jordan is supported by a grant from the NINDS (K23-NS062110).
Dr. Messé is supported by an American Heart Association Beginning Grant-in-Aid.
None of the authors has any conflict of interest or disclosures.