Prediction of malignant middle cerebral artery infarction by a rater independent model based on normalized stroke volume and subarachnoid space volume

Introduction: Surgical decompression improves outcome in patients with malignant middle cerebral artery (MCA) infarction when performed early. However, objective parameters that reliably predict a malignant course of MCA infarction have not been determined so far. We hypothesized that stroke volume and pre-stroke subarachnoid space (CSF), which indicates the intracranial volume reserve, are major determinants to develop a malignant MCA infarction. We here present a method which allows prediction of the course of large MCA infarction based on CT imaging at admission including perfusion CT.

Methods: Patients with a proximal MCA occlusion and a reduced cerebral blood flow (CBV) involving at least 1/3 of the MCA territory were included. Patients were classified as having a malignant course of MCA infarction (clinical signs of herniation) or a non-malignant course. CCT at admission was used to determine the CSF-volume in the healthy hemisphere and the intracranial volume (IV). Brain parenchyma volume (BV) was defined as the difference of IV and CSF. CT perfusion with use of a 4D adaptive spiral allowed complete coverage of the MCA territory allowing segmentation of the total volume of significantly reduced cerebral blood flow (CBV), which indicates the infarct core. Normalized CSF volume (nCSF) was defined as CSF/IV and the normalized stroke volume (nSV) was defined as CBV/BV. We then calculated the ratio of nSV and nCSF. To determine which variables best predict a malignant course of MCA infarction a ROC analysis was conducted.

Results: We included 29 patients (malignant n=15, non-malignant n=14). The ROC curve analysis is given in the table. The ratio of nSV and nCSF shows the lowest rate of false negative predictions (7.2%) in combination with a low rate of false positive predictions (6.6%). The voxel-wise probability map in a standardized brain allows direct visual estimation of brain tissue shifted over the midline (nMSF) with a priori knowledge of nCSF and nSV.

Table: Prediction of Malignant MCA Infarcation

Predicting Factors	Sensitivity,%	Specificity,%	Positive Predictive Value,%	Negative Predictive Value,%
NIHSS, >14.5	66.7	64.3	66.9	64.1
CBV volume, >222.7	73.3	100.0	100.0	77.6
CSF volume, <159.8	66.7	71.4	71.6	66.4
Intracranial volume, <1407.9	53.3	64.3	61.8	56.0
Midline shift volume, >13.2	86.7	92.9	92.8	86.6
Normalized stroke vol/normalized CSF, >1483.6	93.3	92.9	93.4	92.8

Conclusions: In this study a malignant course of MCA infarction was best predicted by the ratio of nCBV and nCSF. The suggested method therefore allows early identification of patients for surgical decompression. Since our method only requires CCT and perfusion CT it is particularly feasible in clinical practice.