Multiclass Support Vector Machine-Based Lesion Mapping Predicts Functional Outcome in Ischemic Stroke Patients

摘要: Purpose The aim of this study was to investigate if ischemic stroke final infarction volume and location can be used predict the associated functional outcome using a multi-class support vector machine (SVM). Material Methods Sixty-eight follow-up MR FLAIR datasets patients with known modified Rankin Scale (mRS) after 30 days were used. The infarct regions segmented calculate percentage lesioned voxels in predefined MNI, Harvard-Oxford cortical subcortical atlas as well four problem-specific VOIs, which identified from database voxel-based lesion symptom mapping. An overall 12 SVM classification models for predicting corresponding mRS score generated overlap values different brain region definitions, laterality information, optional parameters volume, admission NIHSS, patient age. Results Leave-one-out cross validations revealed that including information about terms measurements led improved prediction results compared not utilizing information. Furthermore, integration features all cases tested. additional best predictions precise multi-value accuracy 56%, sliding window (mRS±1) 82%, binary (0-2 vs. 3-5) 85%. Conclusion Therefore, graded SVM-based quantification leads promising but needs further validated an independent rule out potential methodical bias overfitting effects. could valuable tool combined voxel-wise tissue based on multi-parametric acquired at acute phase.