Comparison between voxel-based and subtraction methods for measuring diffusion-weighted imaging lesion growth after thrombolysis

Abstract

Background Infarct growth (IG) is used as surrogate end-point in therapeutic trials. For practical reasons, infarct growth is commonly assessed using simple subtraction of acute from follow-up diffusion-weighted imaging (DWI) lesion volumes. However, the volume subtraction method will underestimate true infarct growth in case of diffusion-weighted imaging lesion reversal. Aim To measure the size of the difference between true infarct growth on voxel-based coregistration and infarct growth approximated with simple volume subtraction. Methods We retrospectively analyzed 322 consecutive stroke patients (median (IQR) age: 70 years (57–80), National Institute of Health Stroke Score at admission 14 (8–19)), who underwent a magnetic resonance imaging before (DWI1) and ≈24 h (DWI2) after IV-thrombolysis. IGvoxel-based was defined as the volume of signal changes on DWI2 that did not overlap with that on coregistered DWI1. This was compared with simply subtracting DWI1 from DWI2 lesion volume (IGsubtracted). We also compared these two metrics for the prediction of three-month unfavorable outcome (mRS ≥ 2) using c-statistics of multivariable models, adjusted for age, and National Institute of Health Stroke Score. Results Infarct growth volume metrics were strongly correlated (ρ = 0.94), but IGsubtracted substantially underestimated IGvoxel-based (median (IQR): 9.52 (0.23–38.9) vs. 16.98 (4.4–45.4) mL). Of the 75 patients with shrinking or stable diffusion-weighted imaging lesion using volume subtraction, IGvoxel-based was ≥5 mL in 20 (27% of the subset, 6.2% of the whole population). Moreover, IGvoxel-based better predicted unfavorable outcome than IGsubtracted (c-statistics = 0.86 (95% CI, 0.82–0.90) vs. 0.82 (0.78–0.87), P = 0.003). Conclusion At early post-thrombolysis time points, the simple subtraction of lesion volumes masked substantial diffusion-weighted imaging lesion growth in 6.2% of patients. Although more time-consuming, the voxel-based method may impact results of trials that use infarct growth attenuation as an end-point.