Time-resolved multiphasic MR angiography (TRMRA) has been suggested as a useful tool for assessment of anatomical and hemodynamic information of vascular structure.^1-5 With contrast agent injection, TRMRA gives the first pass information of the bolus. Although TRMRA with contrast agent injection gives huge amount of 4D data, most previous reports have been relied on visual inspection of time series of projection images. Hence, proper processing of acquired 4D data is required to enhance clinical utility of TRMRA. If we could fully automate the extraction of hemodynamic information from TRMRA, it would be useful both for diagnostic purpose and baseline information for further processing of TRMRA data. At this point, we developed a reliable automatic extraction algorithm for arterial input function (AIF) and venous output function (VOF) in the neck region from 4D TRMRA data covering from aortic arch to head. Furthermore, the extracted functions were tested for generating a time-to-peak (TTP) map. A total 20 subjects were examined at 3T MRI (Verio, Siemens) with local IRB. After intravenous injection of a low dose (0.03 mmol/kg bolus) of gadobutrol (Gadovist), TRMRA was performed using the time-resolved angiography with stochastic trajectories (TWIST) technique (TR/TE/FA = 2.82 ms/0.99 ms/16°, matrix size = 256 × 256 × 128 × 30, 1.5 mm isotropic voxel, acceleration factor = 6, total scan time = 60 s, temporal resolution = 2 s, temporal footprint of TWIST = 8 s).
Figure 1 illustrates overall process of our extraction algorithm for arterial input and venous out functions. To avoid errors from recirculation and variability for venous drainage route of the neck, the functions were estimated in the middle slices of neck region. Firstly, maximum intensity projection (MIP) images were generated from central 64 coronal slices (Fig. 1A). Then, brain mask (Fig. 1B) and vessel mask (Fig. 1C) were generated by different threshold values (0.5 × mean of all frames for brain mask and 0.18 × difference between maximum and minimum of time frames for vessel mask). Slices for neck region (green box, N) were determined by finding a slice which has smallest voxels in the brain mask excluding lower (red box, R1) and upper (blue box, R2) slices. Initial arterial input and venous output functions (Fig. 1D) were estimated from R1 and R2 regions of the vessel mask by averaging signals. Finally, arterial input and venous output functions of the neck regions (Fig. 1E) were estimated by classifying signals in the vessel mask based on TTPs of initial estimates from R1 and R2. Two groups were classified by the middle point of TTPs of the initial estimates (green line in Fig. 1D and E), and then signals were averaged by each group.
Using the extracted AIF and VOF, a TTP map was automatically generated and displayed for coronal MIP image (64 slices). For each voxel, a TTP was calculated within the search range determined by AIF and VOF (from TTP of AIF – 4 s to TTP of VOF + 4 s). The results were compared with TTP maps calculated within the entire search range.Figure 2 shows the estimated AIF and VOF and the output images (TTP map and MIP images at peak time points of AIF and VOF) using the extracted functions. By restricting the search range for TTP calculation, the erroneous voxels due to a secondary peak or low SNR (saturated voxels in dotted circles in Fig. 3) were remarkably reduced in the resulting TTP map.We have demonstrated an automated method for extracting AIF and VOF of neck region from 4D TRMRA data. Our method uses a priori knowledge, such as the anatomic shape and the physiologic sequence of AIF followed by VOF. This priori knowledge is not sequence specific, and therefore future application of this method to other 4D data could be possible. In this study, the extracted functions were utilized for calculating an improved TTP map. Furthermore, the method can be a basis for further hemodynamic analysis of 4D TRMRA data. In addition, AIF itself could be a marker for hemodynamic index of each individual.⁶