Figure 1.
Database screening results and final study inclusion rate.
In both databases the exclusion criterion with the highest exclusion rate was a stroke-to-imaging-time higher than 12 hours. In the 1.5 T database, the number of patients, which had to be excluded due to insufficient image quality (mainly of FLAIR images), was much higher (17.1%) than in the 3 T database (2.0%).
Figure 2.
Labeling of lesions in DWI imaging according to the FLAIR pattern.
A) In case of a FLAIR lesion encompassing the whole DWI-lesion, the hyperintensity was delineated as a region of interest (ROI) (A, second row). The FLAIR-ROI was then copied on the DWI and filled with 6 mm circular ROIs (A, third row)). These ROIs were classified as FLAIR+. B) In cases, in which the FLAIR-ROI did not completely match the DWI-lesion (B, second row), ROIs inside the FLAIR-ROI were classified as FLAIR+, and those outside as FLAIR- (B, third row). C) If no FLAIR lesion was identified (C, second row), the whole DWI lesion was filled with circular ROIs, which were classified as FLAIR- (C, third row). These steps were performed equally in ADC-maps. D), E) and F) show examples in analogy to the scheme, D) showing a patient, where the delineated FLAIR-ROI encompasses the whole DWI-lesion, E) depicting a patient, where the FLAIR-lesion only partially covers the DWI-lesion. Lastly, F) shows a patient, where all DWI-ROIs were labelled as FLAIR- in the absence of a visible FLAIR lesion.
Table 1.
Clinical data, imaging data and comparison of patient groups.
Table 2.
Detailed results of the unadjusted ROC analysis for all 3 raters at 1.5 and 3 T.
Figure 3.
Adjusted ROC curves for the detection of presence of FLAIR-lesions by a relative DWI- and ADC-threshold.
ROC-curves belonging to the detailed data presented in table 3 and 4 (please see legends of table 3 and 4 for further details). DWI-models for Group A (1.5 T) and B (3 T) (A,B) and ADC-models for Group A and B (C,D).
Table 3.
Detailed results of the adjusted ROC analysis for DWI and all 3 raters at 1.5 and 3 T.
Table 4.
Detailed results of the adjusted ROC analysis for ADC and all 3 raters at 1.5 and 3 T.
Figure 4.
Adjusted linear regression analysis to evaluate the association of relative DWI-intensity and time-from-stroke-onset.
Adjusted linear regression analysis was performed to identify a possible association of relative (A) DWI-intensity and (B) ADC-intensity (y-axis,) and time-from-stroke-onset (x-axis) at 1.5 T (blue circles) and 3 T (green circle). At both field strengths, a significant association was found for DWI (A) with moderate adjusted Rsquare values (1.5 T: 0.28; 3 T: 0.44). Adjusted correlation (Spearman's rank correlation) was: 1.5 T = 0.45 (p<0.001), 3 T = 0.69 (p<0.001). In contrast, no association was found for ADC-maps (B) with adjusted Rsquare values near zero (1.5 T: 0.04; 3 T: 0.01) and weak to no adjusted correlation (1.5 T = −0.22, 3 T = 0.05). Plots are shown in logarithmic scale.