Multimodal Imaging Biomarker-Based Model Using Stratification Strategies for Predicting Upper Extremity Motor Recovery in Severe Stroke Patients

Abstract

Background. Various prognostic biomarkers for upper extremity (UE) motor recovery after stroke have been reported. However, most have relatively low predictive accuracy in severe stroke patients. Objective. This study suggests an imaging biomarker-based model for effectively predicting UE recovery in severe stroke patients. Methods. Of 104 ischemic stroke patients screened, 42 with severe motor impairment were included. All patients underwent structural, diffusion, and functional magnetic resonance imaging at 2 weeks and underwent motor function assessments at 2 weeks and 3 months after stroke onset. According to motor function recovery at 3 months, patients were divided into good and poor subgroups. The value of multimodal imaging biomarkers of lesion load, lesion volume, white matter integrity, and cortical functional connectivity for motor recovery prediction was investigated in each subgroup. Results. Imaging biomarkers varied depending on recovery pattern. The integrity of the cerebellar tract ( P = .005, R 2 = .432) was the primary biomarker in the good recovery group. In contrast, the sensory-related corpus callosum tract ( P = .026, R 2 = .332) and sensory-related functional connectivity ( P = .001, R 2 = .531) were primary biomarkers in the poor recovery group. A prediction model was proposed by applying each biomarker in the subgroup to patients with different motor evoked potential responses ( P < .001, R 2 = .853, root mean square error = 5.28). Conclusions. Our results suggest an optimized imaging biomarker model for predicting UE motor recovery after stroke. This model can contribute to individualized management of severe stroke in a clinical setting.