PETRA quiet MRA: Improved Robustness with 3D Elastic Cross-Registration
Yutaka Natsuaki1, Robert Grimm2, Xiaoming Bi1, David Grodzki2, Peter Schmitt2, and Gerhard Laub1

1Siemens Healthcare, Los Angeles, CA, United States, 2Siemens Healthcare, Erlangen, Germany


Previously, Ultra-Short TE sequences and their subtraction-based non-contrast MR angiography (e.g. PETRA qMRA) have shown promising results in intracranial applications, in particular with tortuous carotid arteries that has been problematic for the current clinical gold standard 3D Time Of Flight (TOF). As with any subtraction based MRA techniques, PETRA qMRA is also sensitive to motion occurred in subsequent acquisitions. The current work demonstrates that the proposed 3D elastic cross-registration can solve this limitation for the PETRA qMRA, and it takes to the next level in motion robustness and in the highest attainable isotropic resolution.


For non-contrast intracranial MR angiography (MRA), the Ultra-short TE (UTE) sequences (e.g. PETRA 1) and their subtraction-based MRA techniques (e.g. PETRA qMRA 2-3) have shown great promises of being a viable alternative to the current gold standard in 3D Time-of-Flight (TOF) 4 with well-known flow dephasing issues. However, as with any subtraction based MRA techniques, PETRA qMRA too is limited by how well two data sets from subsequent measurements are aligned. The current work demonstrates that the 3D elastic cross-registration can potentially solve this limitation, and enables PETRA qMRA to reach new level in motion robustness and in the highest attainable isotropic resolution.


With the PETRA qMRA, two imaging volumes (i.e. labeled and control data) are acquired (Fig.1). The labeled data set applies the slice-selective saturation pulse at the upstream of the carotid artery, effectively darkening the arterial inflow. The subsequent control data features the same saturation pulse, only prescribed outside of the imaging volume for retaining the MT effects while not interfering with the imaging volume. Two data sets are then subtracted to generate desired Arterial MRA with back ground tissue and venous flow suppressed.

As with any subtraction-based MRA techniques, the two subsequent measurement data need to be aligned. Otherwise, mis-registered data subtraction generates undesired background residual artifacts and interferes with the intended arteries. The individual PETRA measurement, on the other hand, is robust to motion thanks to the 3D radial acquisition. Combining these two motion properties, we hypothesize that if the two PETRA data sets are properly registered via post-processing, their subtraction will result in the optimal angiography data. In addition, with such post-processing, the longer scan time with improved resolution is now attainable. The current work utilizes 3D elastic registration originally proposed by Chefd’Hotel et al 5-6. Since this post-processing has an overall smoothing effect on the registered images, both labeled and control data are cross-registered prior to the subtraction (Fig.2).

The sequence prototype was implemented on 3T scanners (MAGNETOM Skyra and Prisma, Siemens Healthcare, Erlangen, Germany). The technique was validated with healthy volunteers (n=5) under a local IRB approved protocol. PETRA qMRA data was acquired with the following high resolution protocol: FOV 220mm3, 75000 radial spokes; isotropic 0.57mm3; flip angle = 6°; TR/TE 4.25/0.07msec; BW 1860 Hz/pixel; slice selective Saturation pulse applied once per 25 TRs; total scan time 2 x 6:01min. The acquired DICOM data were then post-processed offline on a standard laptop PC for the proposed cross-registration (syngo via Frontier MR Elastic Registration Toolbox prototype, Siemens Healthcare, Erlangen, Germany). The resulting MRA data sets (without vs with elastic cross-registration) were compared for overall image quality, delineation of vessel details, and presence/absence of artifacts.


All scans, reconstructions, and post-processing were performed successfully. To our surprise, each 3D elastic registration of PETRA isotropic data (384x384x384 matrix size) only took few seconds on a standard non-GPU laptop PC. Fig.3 shows a representative volunteer data acquired with the high resolution protocol. While smoothing effects from the post processing were apparent in subtracted data, it did not smooth out the target arteries including small distal branches. As expected, the 3D elastic registration has shown clear improvements over the unregistered data in residual background artifacts and noises (e.g., spurious arteries (red arrows), residual brain tissue (green oval) and distal arteries interference (blue oval)).


For an intracranial MRA with rigid brain, 3D elastic registration is probably overkill. However, it still provides the best possible registration with such high efficiency, and as the image resolution gets higher, the data are more prone to the registration error. In the current work, we intentionally increased the resolution (0.57mm3) and the total scan time (12 min). Because of this, the resulting data without registration appears far noisier than the previously reported 0.86mm3 isotropic resolution images with a total scan time of 8 minutes. Nevertheless, a total scan time of 12 min is too long to be clinically relevant and further acceleration (e.g. radial SENSE 7) must be considered.


The 3D elastic cross-registration helps improving the robustness of the PETRA qMRA, especially for high-resolution scan protocols. Further work includes clinical validation of the method on intracranial MRA patients and technical developments in PETRA acquisition to be more efficient.


No acknowledgement found.


1. Grodzki DM, et al. MRM 67:510–518 (2012)

2. Grodzki DM, et al. Proc. MRA Club (2014)

3. Natsuaki Y, et al. Proc. ISMRM (2015)

4. Parker DL et al. MRM 17:434-451(1991)

5. Chefd’hotel C, et al. Proc. IEEE Biomed lma. (2002)

6. Chefd’hotel C, et al. Proc. ICCV (2001)

7. Pang J, et al. MRM 71:67-74 (2013)


Fig.1: Schematic diagram of PETRA qMRA, a subtraction based non-contrast MR angiography. Two data sets with different contrasts (Control (bright blood) and Labeled (dark blood)) are acquired, and then subtracted pixel-by-pixel basis to get rid of background and venous flow.

Fig.2: Schematic diagram of the proposed 3D elastic cross-registration. First, B (Label) is registered to A (Control), resulting in the registered Label B’. Then A is then cross-registered to the new Label B’, resulting in the registered Control A’. These registered data are subtracted for the final MRA.

Fig.3: Representative high-resolution PETRA qMRA without (upper row) and with (lower row) the proposed registration. Axial source subtraction, coronal & sagittal thin MIPs are shown. Without registration, background residuals on Src.Subtract were propagating to cause artifacts on the ThinMIPs (e.g., spurious arteries, residual brain tissues and distal arteries interference).

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)