Multiparametric Magnetic Resonance Imaging for Prediction of Parenchymal Hemorrhage in Acute Ischemic Stroke After Reperfusion Therapy

Abstract

Background and purpose: Patients with acute ischemic stroke are at increased risk of developing parenchymal hemorrhage (PH), particularly in the setting of reperfusion therapies. We have developed a predictive model to examine the risk of PH using combined magnetic resonance perfusion and diffusion parameters, including cerebral blood volume (CBV), apparent diffusion coefficient, and microvascular permeability (K2).

Methods: Voxel-based values of CBV, K2, and apparent diffusion coefficient from the ischemic core were obtained using pretreatment magnetic resonance imaging data from patients enrolled in the MR RESCUE clinical trial (Mechanical Retrieval and Recanalization of Stroke Clots Using Embolectomy). The associations between PH and extreme values of imaging parameters were assessed in univariate and multivariate analyses. Receiver-operating characteristic curve analysis was performed to determine the optimal parameter(s) and threshold for predicting PH.

Results: In 83 patients included in this analysis, 20 developed PH. Univariate analysis showed significantly lower 10th percentile CBV and 10th percentile apparent diffusion coefficient values and significantly higher 90th percentile K2 values within the infarction core of patients with PH. Using classification tree analysis, the 10th percentile CBV at threshold of 0.47 and 90th percentile K2 at threshold of 0.28 resulted in overall predictive accuracy of 88.7%, sensitivity of 90.0%, and specificity of 87.3%, which was superior to any individual or combination of other classifiers.

Conclusions: Our results suggest that combined 10th percentile CBV and 90th percentile K2 is an independent predictor of PH in patients with acute ischemic stroke with diagnostic accuracy superior to individual classifiers alone. This approach may allow risk stratification for patients undergoing reperfusion therapies.

Clinical trial registration: URL: https://www.clinicaltrials.gov. Unique identifier: NCT00389467.

Keywords: biomarkers; embolectomy; hemorrhage; permeability; stroke.

Figure 1.

A 45-y-old man presented with left sided weakness and baseline National Institutes of Health Stroke Scale score of 19. Magnetic resonance imaging (MRI) was obtained 5.5 hours from symptoms’ onset presentation. Axial apparent diffusion coefficient (ADC; A), cerebral blood volume (CBV; B), and microvascular permeability (K2; C) from the pretreatment MRI are shown. Volume of interest (VOI) from the diffusion-weighted image hyperintense region was automatically generated, and after image coregistration, voxel values from VOI were used to calculate 10th percentile ADC (384×10⁻⁶ mm²/s), 10th percentile CBV (0.40), and 90th percentile K2 (0.32). Subsequent MRI 8 d later shows development of parenchymal hemorrhage in the region of infarction (D).

Figure 2.

Scatter plots for 10th percentile cerebral blood volume (CBV), 10th percentile apparent diffusion coefficient (ADC), and 90th percentile microvascular permeability (K2). There are moderate negative correlations between 90th percentile K2 with 10th percentile CBV and 10th percentile ADC and a moderate positive correlation between 10th percentile CBV and 10th percentile ADC.

Figure 3.

Classification tree model using combined 10th percentile cerebral blood volume (CBV) and 90th percentile microvascular permeability (K2). If 10th percentile CBV ≥0.47 and 90th percentile K2 <0.28, the model predicts no parenchymal hemorrhage (PH) in 55 out of 63 patients. If 10th percentile CBV <0.47 and 90th percentile K2 ≥0.28, the model predicts PH in 18 out of 20 patients.