Introduction Assessing perfusion in acute stroke is relevant to identify patients more likely to benefit from treatment with thrombolysis (1). Currently used techniques either require intravenous contrast, are unreliable, or are expensive and thus not widely used in routine clinical practice (2). Aim: To investigate the performance of a non-invasive method (3), which measures the relative delay in local blood oxygen level dependent (BOLD) signal using resting- state functional MRI (rsfMRI) data, for the diagnosis and follow-up of perfusion deficits in patients with acute stroke in a clinical setting by directly comparing it to an established method for assessing brain perfusion (dynamic susceptibility contrast MRI, DSC-MRI). Methods • 76 patients • <24 hours of ischaemic stroke onset • Perfusion (DSC-MRI) and BOLD delay (rsfMRI) maps Results Image quality Diagnostic agreement Volumetric agreement Spatial agreement Discussion BOLD delay maps are sensitive to perfusion deficits in acute stroke but have a high rate of false-positives, possibly due to CSF artefacts. BOLD delay maps agree, in terms of location, with perfusion maps very early (<4.5h) following stroke onset. BOLD delay lesions underestimate perfusion lesion volumes, resulting in low sensitivity for perfusion-diffusion mismatch. Head motion adversely affects the spatial overlap between BOLD delay and perfusion lesions, resulting in false negative BOLD delay maps. BOLD delay in stroke circumvents the infarct core and specifically affects the surrounding tissue. Contrary to previous studies (4), this suggests that BOLD delay has neuronal rather than vascular origins. BOLD delay maps are reliable for following up stroke patients, as changes in lesion volumes between the first and second days following stroke onset were similar between BOLD delay and perfusion maps. Ahmed Khalil, Ann-Christin Ostwaldt, Till Nierhaus, Kersten Villringer, and Jochen Fiebach International Graduate Program Medical Neurosciences, Charité Universitätsmedizin Berlin Further information [email protected] 1. BOLD signal across time in one voxel (local) and whole brain (global) are extracted. 2. Signal corresponding to one repetition time (TR) at the start and end of the voxel time series is removed. 3. Local time series is passed through the global time series in a series of sequential steps (step=1TR). 4. For each step, the correlation between the local and global time series is calculated. 5. 1-4 repeated for each voxel, resulting in a TSA map showing the step yielding the maximum correlation between local and global time series. Perfusion deficits Perfusion-diffusion mismatch Sensitivity 82% (66 to 92%) 54% (34 to 72%) Specificity 41% (24 to 61%) 90% (76 to 97%) Positive predictive value 65% (50 to 78%) 79% (54 to 94%) Negative predictive value 63% (38 to 84%) 73% (58 to 85%) Yield 89% Dice similarity coefficient 0.0 0.2 0.4 0.6 0.8 Dice similarity coefficient <4.5hrs 4.5-9hrs >9hrs 0.0 0.2 0.4 0.6 0.8 Longitudinal reliability Effect of head motion Day 2 perfusion lesion volume (ml) Day 2 BOLD delay lesion volume (ml) 0 20 40 60 80 0 50 100 150 200 Change in volume (ml) Perfusion BOLD delay -50 0 50 100 150 Independent variable B P-value Maximum motion -0.023 (-0.067 to -0.005) 0.023 Mean motion -0.157 (-0.308 to -0.006) 0.042 Number of movements >0.5mm -0.002 (-0.004 to -0.001) 0.006 Time shift analysis (TSA): Percent of maps (%) Perfusion BOLD delay 0 20 40 60 80 100 Certain Uncertain Uninterpretable Percent of maps (%) Perfusion BOLD delay 0 20 40 60 80 100 High Moderate Low BOLD delay Perfusion DWI BOLD delay Perfusion Example of a false-negative BOLD delay map associated with severe head motion. The perfusion map shows a large lesion corresponding to the infarct on the right side. Results of multiple logistic regression analysis: head motion is associated with decreased spatial overlap between perfusion and BOLD delay lesions. Regression coefficients (B) represent the effect sizes on the Dice similarity coefficient. Positive correlation between BOLD delay and perfusion lesions on the second day of stroke onset (rho=0.78, p<0.0001). Tukey boxplot showing lesion volume changes between days 1 and 2 of stroke onset using the two techniques (p=0.11) Examples of high spatial overlap between BOLD delay and perfusion lesions. Distribution of the degree of spatial overlap between BOLD delay and perfusion lesions in the study sample. Middle bar is median, whiskers are interquartile range. Bland-Altman plot showing the bias (20 ml, red line) and 95% limits of agreement (-117 to 157 ml , black lines) between perfusion and BOLD delay lesion volumes. Image quality (diagnostic confidence, left; noisiness, right) of the BOLD delay and perfusion maps as assessed by the raters. The associations between map type and image quality were statistically significant (confidence, p<0.0001 ; noisiness, p<0.0001) Coverage of the infarct core Coverage ratio Perfusion BOLD delay 0.0 0.2 0.4 0.6 0.8 1.0 DWI BOLD delay Perfusion Example of low DWI lesion coverage by a BOLD delay lesion and high coverage by a perfusion lesion in the same patient. Tukey boxplots showing the degree of spatial overlap between BOLD delay and perfusion lesions in patients presenting in different time periods following stroke onset (p=0.26). Values in brackets are 95% confidence intervals Values in brackets are 95% confidence intervals Acknowledgments Thanks to Peter Brunecker for the technical advice and support, and Alexander Nave and Peter Koch for helping with the assessment of the maps. Proportion of voxels in the DWI lesion covered by perfusion and BOLD delay lesions (p=0.005). NO DELAY DELAY Literature cited (1) Thomalla et al Stroke 2006 (2) Goyal et al Radiology 2013 (3) Lv et al Annals of Neurology 2013 (4) Amemiya et al Radiology 2013