Automated Method for Generating the Arterial Input Function on Perfusion-Weighted MR Imaging: Validation in Patients with Stroke
Michael Mlynasha,
Irina Eyngorna,
Roland Bammerc,
Michael Moseleyc and
David C. Tonga,b
a Department of Neurology and Neurological Sciences, Stanford Stroke Center, Palo Alto, CA
b Departments of Neurology and Neurological Sciences, Stanford University Medical Center, Stanford, CA
c Department of Radiology, Stanford University Medical Center, Stanford, CA
View larger version (24K):
[in a new window]
|
FIG 1. Manual and automated choice of the best AIF.
A, Manual choice, based on visual estimation of the overall shape of the curve, high signal intensity, early peak time, and small width.
B, Automatically computed AIF is identified as the one with the optimal combination of Gaussian-fit parameters (maximum A0 combined with A1 and A2 in the predefined limits) and the one satisfying the goodness-of-fit test.
| |
View larger version (22K):
[in a new window]
|
FIG 2. AIF identified at voxels 24 mm apart.
A, Manual.
B, Automated.
| |
View larger version (38K):
[in a new window]
|
FIG 3. Tmax and CBF maps, created on the basis of the manual (A, C) and automated (B, D) AIFs from Figure 1, have spatial pattern correlations of r = 0.87 and r = 0.86, respectively. Distance between corresponding AIF voxels, or d, is 30.6 mm.
| |
View larger version (37K):
[in a new window]
|
FIG 4. Tmax and CBF maps (A, C) and (B, D) corresponding to Figure 2 have spatial pattern correlations of r = 0.74 and r = 0.60, respectively.
| |
