Introduction: The DWI/FLAIR mismatch has been proposed as an imaging pattern to predict stroke onset before 4.5h in patients with unknown time of symptom duration. Limitations of the DWI/FLAIR mismatch include the additional imaging time required to obtain FLAIR, difficulty of assessment in patients with severe white matter disease, poor interrater agreement, and motion sensitivity of longer MRI sequences. We hypothesised that automated analysis of DWI imaging would be at least as accurate to predict stroke onset before 4.5h.
Methods: Data from the Axis 2 trial were used and patients were included in whom a DWI lesion was detected by automated software (RAPID). The visual DWI/FLAIR mismatch was rated in accordance with the criteria of the ongoing WAKE-UP trial. For every patient the relative DWI (rDWI) was calculated in a voxel based manner as the ratio of the diffusion intensity in that voxel and the median diffusion intensity of a sphere with radius 15 mm around the homologue voxel in the contralateral hemisphere. The mean and standard deviation (SD) of the rDWI was obtained in the DWI lesion. Receiver operating curves (ROC) and corresponding area under the curves (AUC) for predicting the 4.5h time-window were determined.
Results: We included 200 patients. In 8 patients (4%) the visual DWI/FLAIR rating was hampered due to either extensive white matter disease or poor quality of FLAIR imaging. The DWI/FLAIR mismatch had an AUC of 0.66 to predict stroke onset before 4.5h vs an AUC of 0.71 for rDWI SD (p for difference=0.4). The optimal rDWI SD threshold for predicting stroke ≤ 4.5h was 0.23. Using this threshold, stroke onset was accurately predicted in 73% of patients vs 65% accuracy with the visual DWI/FLAIR mismatch (p for difference=0.1). (table 1)
Conclusion: Our results suggest that rDWI may have advantages over the visual DWI/FLAIR mismatch for predicting the time of stroke onset. rDWI provides an accurate and objective assessment with the potential for fully automated processing.