Validation of an image derived input function estimation method on PET/MR

Objectives: The study objective was to validate a recently introduced non-invasive image derived input function (IDIF) estimation method with the gold standard arterial blood sampling.

Methods: Six subjects (31-50 years old) were injected with 408±62 MBq of 15O-water simultaneously with the start of a 10 min PET scan on the SIGNA PET-MR (GE Healthcare, WI, Waukesha). During PET scanning, a sagittal vascular (inhance 3D velocity) MR series was used with the following parameters: TR=8.7 ms, TE=4.1 ms, FOV=24×21.6 cm, slice thickness=3 mm, 32 slices, velocity encoding=40, phase acceleration=2.0, and scan time=1:21 min. The PET list file was unlisted for every second and total true and scatter coincident events were plotted to identify tracer arrival into the brain arteries. Then, a short time frame over the arrival of the tracer to the cervical region was reconstructed to obtain a PET angiogram. The cervical arteries were then segmented using the MR vascular images and PETA images. Spill-over and spill- in artifacts were estimated using PETA images and the actual arterial volume was measured from the MR vascular images. The PET list file was unlisted and images were reconstructed for every 1 s for the first 30 s, every 3 s for the next 30 s, every 5 s for the 2nd minute, every 10 s for the 3rd and 4th minute and every 30 s for 5th to 10th minutes. The AIF was estimated by dividing total counts from the cervical arteries of each frame by the MR-based arterial volume. For each patient, blood samples were continuously drawn from the radial artery at the wrist using a peristaltic pump, and the tracer concentration in the arterial blood was measured using a Twilite two detector (Swisstrace) to estimate the AIF. In order to calculate the AIF at the brain arteries from these blood samples, the delay and dispersion of the arterial input function was corrected using standard PET-based methods. The CBF and distribution volume were calculated using both the IDIF method and the blood samples by minimizing the mean square of the error between the PET observations and model fit using the Nelder-Mead simplex algorithm in MATLAB (Mathworks, Wilmington, MA).

Results: Figure 1 shows the (a) PETA and (b) MR vascular images for one of the patients. The PETA images clearly show the arteries and the extent of the spill-over. Figure 2 compares the AIF curve estimated by the proposed IDIF method and the AIF curve measured by the blood samples. The comparison shows excellent correspondence between the IDIF method and the gold standard blood sampling method with 9% and 11% difference for the 1st pass and the entire AIF, respectively. The IDIF captures the AIF peak correctly and has increased signal-to-noise ratio compared to the blood sampling method. The delay and the dispersion of the AIF curve is nearly identical between the two methods. The CBF over the whole brain was measured 29.5±8.7 and 27.0±14 ml/s/100g with the AIF measured by IDIF method and blood samples, respectively with a mean difference of 14% between the two methods. The volume distribution over the whole brain was measured 0.5±0.1 for both methods with a mean difference of 15% between them.

Conclusion: As the results show, the proposed method is capable of determining a high fidelity IDIF from simultaneous PET/MRI data. Having a “blood-free” method that obviates the need for direct arterial sampling is of benefit to both investigators and their subjects, because of the high costs, inconvenience, and potential risks associated with arterial cannulation. It has applications beyond 15O-water PET, enabling pharmacokinetic modeling to be performed that is required for quantitative PET tracer studies. Research Support: GE Healthcare, Stanford University Lucas Center, Uppsala University. $$graphic_95D521E5-EADF-4C1D-B879-9168A49B1DF5$$ $$graphic_EC63587F-AF7A-478C-94C3-C944DD5FB66D$$