Quantification of Cerebral Volumes on MRI 6 Months After Aneurysmal Subarachnoid Hemorrhage
Background and Purpose—MRI can be used to assess structural damage to the brain after aneurysmal subarachnoid hemorrhage. We tuned, validated, and applied k-Nearest Neighbor-based segmentation to quantify cerebral volumes on MRI 6 months after aneurysmal subarachnoid hemorrhage.
Methods—After tuning, the accuracy of k-Nearest Neighbor-based segmentation was assessed with manual segmentations. Next, supratentorial cerebral parenchymal, peripheral cerebrospinal fluid, and lateral ventricular volumes of 55 patients were compared with those of 25 age- and sex-matched control subjects and related to clinical outcome (modified Rankin Scale).
Results—k-Nearest Neighbor-based segmentation showed good agreement with manual segmentations. Compared with control subjects, patients had a larger lateral ventricular volume (difference: log-transformed values 0.54; 95% CI,0.33–0.75), smaller peripheral cerebrospinal fluid volume (−26 mL; 95% CI, −40 to −11), and similar cerebral parenchymal volume (2 mL; 95% CI, −10 to 15). In patients, parenchymal (median split; OR, 38.8; 95% CI, 4.6–329.0) and ventricular volumes (7.4; 95% CI, 1.6–33.5) correlated with functional outcome.
Conclusions—k-Nearest Neighbor-based segmentation provides accurate cerebral volume measurements after aneurysmal subarachnoid hemorrhage. In this proof-of-principle study of this volumetric technique, we demonstrated volume changes relative to controls, which correlated with functional outcome.
The long-term functional consequences of aneurysmal subarachnoid hemorrhage (aSAH) are probably due to both focal lesions (ie, infarcts) and global change (ie, reductions in brain volume and ventricular enlargement) in the brain. Quantification of the volumes of these abnormalities on MRI may be of value for etiologic and intervention studies. However, due to the heterogeneity of the MRI abnormalities after aSAH, quantification with (semi-)automated methods is challenging and has only been described in a few published studies.1,2
We tuned and validated k-Nearest Neighbor-based segmentation (kNN)3 to quantify brain volumes on MRI after aSAH, compared volumes between patients and control subjects, and related volumes to functional outcome.
Materials and Methods
Patients (n=55) were derived from a prospective study that compared MR angiography with intra-arterial digital subtraction angiography to assess aneurysm occlusion with coils 6 months (±2 months) after aSAH4 (for details, see the online-only Data Supplement Methods). The study only included patients with a relatively favorable clinical outcome, reflected in a modified Rankin Scale of ≥3 at the time of the MRI.5 Patients with a history of stroke (aSAH or other) were excluded.
An age- and sex-matched control group consisted of 18 persons who had endovascular treatment for unruptured intracranial aneurysms and 7 persons who were screened for aneurysms (for details, see the online-only Data Supplement Methods). The study was approved by the medical ethics committee of the UMC Utrecht. All participants provided written informed consent.
The World Federation of Neurosurgical Societies SAH grading scale was recorded at admission.6 Treatment with a ventricular drain for hydrocephalus and occurrence of delayed cerebral ischemia (defined as new focal deficits or decreasing level of consciousness with new infarcts on CT) were also recorded. Clinical outcome was assessed with the modified Rankin Scale5 at the time of the MRI scan.
Scans were acquired on a 3-T Philips MR system with a standardized protocol (24 contiguous slices, voxel size: 0.45×0.45×4.00 mm3) consisting of an axial T1 (TR/TE=500/10 ms) and T2 (TR/TE=3000/80 ms). kNN was tuned and the accuracy was assessed with manual segmentations on scans of patients after aSAH with heterogeneous cerebral damage (see the online-only Data Supplement Methods). Agreement between kNN and the manual segmentations was expressed as a fuzzy similarity index.7
Supratentorial cerebral parenchymal, peripheral cerebrospinal fluid (pCSF), and lateral ventricular volumes were measured (Figure) and compared between patients and control subjects (and additionally between patients with or without infarcts) by linear regression analyses adjusted for age, sex, and intracranial volume and in additional analyses also for cerebral infarct volume. Within the patient group, the relationship between cerebral volumes (relative to intracranial volume, dichotomized by a median split) and clinical outcome after aSAH (dichotomized at a modified Rankin Scale <2 or modified Rankin Scale ≥2) was assessed by logistic regression analyses adjusted for age and sex and in additional analyses also for cerebral infarct volume.
kNN segmentations of intracranial (fuzzy similarity index: 0.98), cerebral parenchymal (fuzzy similarity index: 0.93), and lateral ventricular volumes (fuzzy similarity index: 0.92) showed good agreement with the manually segmented validation data (see online-only Data Supplement Table). kNN segmentations of pCSF volume (fuzzy similarity index: 0.71) showed moderate agreement with the validation data.
The characteristics of the patients and controls are shown in Table 1. At 6 months after aSAH, patients had a larger lateral ventricular volume and smaller pCSF volume than control subjects; cerebral parenchymal volume was not affected (Table 2).
An episode of delayed cerebral ischemia was recorded in 10 patients (19%). Infarcts were observed in 26 patients (10 of 10 with clinically manifest delayed cerebral ischemia, 16 of 45 without delayed cerebral ischemia; see also Table 1). Median infarct volumes were 8.7 mL (10th–90th percentile, 0.4–61.3) among these 26 patients. Infarcts were also observed in 2 control subjects (0.2 and 2.0 mL). Patients with cerebral infarcts had a larger lateral ventricular volume (log-transformed values: 0.32; 95% CI, 0.07–0.57) compared with patients without infarcts. Cerebral parenchymal (−5.6 mL; 95% CI, −19.0 to 7.9) and pCSF volume (−12.0 mL; 95% CI, −28.4 to 4.3) were not significantly different. The differences in pCSF and lateral ventricular volumes between patients and control subjects remained statistically significant after adjustment for infarct volume.
In the patient group, both a higher lateral ventricular volume and lower cerebral parenchymal volume (median split) were associated with worse outcome on the modified Rankin Scale (OR, 7.4; 95% CI, 1.6–33.5) and 38.8 (95% CI, 4.6–329.0), respectively; after additional adjustment for infarct volume: 3.8 (95% CI, 0.7–20.5) and 36.8 (95% CI, 3.9–346.1). A higher pCSF volume was not related to outcome (OR, 2.3; 95% CI, 0.6–8.0).
At 6 months after aSAH, patients had abnormal brain MRI volumes relative to control subjects, and these volumetric abnormalities appeared to be functionally relevant.
kNN segmentation provided accurate probabilistic measurements of cerebral volume after aSAH. We specifically tuned kNN for segmentation of cerebral volumes in patients after aSAH. The automated segmentation was supplemented by manual segmentations of focal lesions (ie, infarcts, drain trajectories), coil artifacts, and incidental findings.
The focus of our study was to demonstrate feasibility and accuracy of cerebral volume measurements after aSAH as a proof of concept for application of this technique in this patient population. The volumetric results concern a group of patients with a relatively favorable outcome and a control group that mainly consisted of people treated for unruptured aneurysms. This selection might have underestimated the impact of an aSAH on the brain. Future studies can use this technique to unravel the exact contribution of the aSAH, secondary complications, treatment, and vascular risk factor profile on change in cerebral volume.
Despite the fact that all patients in our sample regained functional independence and the modest sample size of our study, the subtle volumetric abnormalities that could be identified with kNN did correlate with functional outcome independent of infarct volume. Although this observation will have to be confirmed in larger studies, it is of importance for future studies that may wish to apply this volumetric technique in etiologic or intervention studies on functional and cognitive deficits after aSAH.
Sources of Funding
This study was supported by a high potential grant from Utrecht University.
The help of C. Jongen with the manual segmentations is gratefully acknowledged.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.112.669184/-/DC1.
- Received June 26, 2012.
- Accepted July 3, 2012.
- © 2012 American Heart Association, Inc.
- Bendel P,
- Koivisto T,
- Aikia M,
- Niskanen E,
- Kononen M,
- Hanninen T,
- et al
- van Swieten JC,
- Koudstaal PJ,
- Visser MC,
- Schouten HJ,
- van Gijn J