Prediction of Hemorrhagic Transformation in Acute Ischemic Stroke Using Texture Analysis of Postcontrast T1-Weighted MR Images

Purpose: To test the hypothesis that texture analysis of postcontrast T1-weighted MR images will predict hemorrhagic transformation (HT) in patients with acute ischemic stroke (AIS) with better accuracy than visual evidence of contrast enhancement (VE). Materials and Methods: Thirty-four AIS patients were examined within 3.5 +/- 1.5 h after stroke. T1-weighted MR images were acquired 19 +/- 7 min postconstrast injection. HT was determined by follow-up imaging at 24-72 h. Postcontrast images were evaluated for VE. Four second-order textual features were extracted (f1, f2, f3, and f9) for each patient. Receiver operating characteristic (ROC) curves were constructed for VE and for textural features, with HT as the outcome measure. Results: The f2 for HT patients (n = 12) was significantly lower than in non HT patients (1058 +/- 356) versus 1568 +/- 52.7 P = 0.005); the converse was true for f3 (0.67 +/- 0.12 versus 0.54 +/- 0.13, P = 0.07), ROC analysis indicated that the f2 and f3 textural features were the only two significant predictors of HT (P = 0.0018 and P = 0.0042). The addition of VE to either f2 or f3 did not result in a significant improvement in accuracy. Conclusion: Texture analysis of postcontrast T1-weighted images may be superior to visual evidence of enhancement for the prediction of HT.