Fig 1.
Example of real-time TSP maps in an ischemic stroke patient with a right-sided perfusion deficit.
The left panel shows the cursor placed within presumed healthy tissue showing uniform values approaching 1.0 for most of the brain except for focal, heterogeneous perfusion deficit lesions. The right panel shows the seeding voxel within the perfusion deficit. Note that these maps will change in real-time as the investigator moves the cursor around the brain to select a new seeding voxel.
Table 1.
Demographic and clinical characteristics of the study groups.
Fig 2.
Example of the automated TSP maps in an ischemic stroke patient with a right-sided perfusion deficit.
The images were generated using an iterative method that first calculated an average time-series of healthy tissue (top left) to generate a Pearson’s correlation map of all voxels in the brain based on correlation with the average time-series of healthy tissue (top right). Signal intensity time-series for voxels in healthy and under perfused tissue are in the bottom panel. Voxels in healthy tissue demonstrate a signal intensity time-series that is like the average signal intensity time-series for all healthy tissue, while a voxel in the perfusion deficit will have a signal intensity time-series that is delayed, dispersed and/or decreased.
Fig 3.
TSP maps in the 20 TIA imaging-negative patients.
The TSP maps were uniform with values approaching 1.0 for the whole brain. The color scale for all TSP maps runs from <0 to 1, blue to red.
Fig 4.
An overview of the automatically generated TSP and TTP maps for all 20 ischemic stroke patients.
The slice-of-interest is shown for the mid segment of perfusion deficit. The top row for each patient delineates the TSP map with a Similarity Index (Pearson’s r), color scale from ≤0 to 1 (blue to red). Below the numbered TSP, the corresponding TTP map was individually windowed for each patient. This was done since TTP values are non-standardized in general and can be affected by factors such as speed of bolus injection.
Fig 5.
Bland-Altman plots of the inter-rater reliability between the two readers for TSP, TTP and MTT map-based lesion volumes.
Fig 6.
Panel A shows the inter-rater bias measures for TSP, TTP or MTT map-based lesion volumes. Panel B shows that the Pearson's correlation between TSP and TTP map-based lesion volumes was high (r(18) = 0.73, p<0.0003). Panel C shows that the effective CNR was greater for TSP (352.3) compared to TTP (p<0.03) and MTT maps (p<0.03).
Fig 7.
Figure comparing the spatial overlap between lesion volumes drawn on TSP and TTP maps for each patient.
Overlap from lesion volumes from each rater was analyzed separately and averaged for each patient.
Fig 8.
Example of TSP maps with varying Pearson’s R correlation thresholds in a patient with a right-sided perfusion deficit.
The red encircled map shows the threshold used for the automated lesion detection. No visual differences were found between TSP maps constructed with these different thresholds (B) Mean signal values for lesion and healthy tissue (based on the unbiased perfusion lesion) showed relatively stable TSP values within TSP maps for healthy and lesion tissue demonstrating robustness. (C) The mean difference in signal between healthy and perfusion tissue increased slightly (by ~0.03 in the 0.9 map compared to the 0.6 map) in TSP values.