Motion-resolved volumetric imaging (4D-MRI) has been an attractive research topic in the last several years. One important application with the recent introduction of simultaneous PET/MR imaging has been the use of 4D-MRI to generate a motion model, which is subsequently applied to correct for physiologic motion in the relativity low resolution PET data (1). For this application, a rapid, continuous and comprehensive 4D-MRI data acquisition is desirable, so that other clinically relevant images can be collected within the same imaging session (2). The current standard method utilizes a self-navigated stack-of-stars radial sequence (1, 3). However this approach suffers from limited spatial coverage along the partition direction and relatively long acquisition time (~10 min). In this study, we developed, implemented, and tested a novel self-navigated 3D spiral sequence that provides higher design flexibility and faster acquisition speed for respiratory motion correction. With this new acquisition scheme, respiratory motion-resolved images covering the entire abdomen with 1.8 mm isotropic resolution could be obtained in only 1:12 min.

Sequence design: A 3D golden-angle spiral projection gradient echo sequence that enables capture of respiratory motion along the foot-head (F-H) direction was designed to acquire MR images covering the entire upper abdominal volume with isotropic resolution. Multishot 2D spirals (N=21 shots, spherical field of view 350 mm, 1.8 mm isotropic resolution, readout duration 4.5 ms per arm) were acquired continuously with increment of the azimuthal angle of the projection along the F-H axis (Fig. 1). Golden-angle increments were employed in order to ensure good k-space distribution in the subsequent motion sorting procedure, similar to previous approaches (4). With this hybrid spiral projection sampling scheme, each 2D readout yields a projection plane along the principal motion axis acquired at a fast rate (214 ms), while the complete 3D acquisition yields isotropic depiction of the full abdominal volume, which can be retrospectively sorted in into distinct respiratory motion states using the motion signal extracted from the acquired data.

In vivo MR imaging: The sequence designed above was tested on healthy volunteers who gave informed consent. Imaging was performed on a 3.0-T clinical scanner (Skyra, Siemens Healthcare, Erlangen) equipped with spine and body matrix coil and 40 mT/m, 180 mT/m/ms gradient sets with the following parameters: TR/TE=10.2/0.7 ms, flip angle=10^o. The volunteers were instructed to breathe freely during the entire scan and the acquisition time for the 3D spiral projection sequence was 1:12 min including 333 projections. Fat suppression was performed using 1-1 binomial pulses. The RF pulse selected a 300 mm-thick sagittal slab including all abdominal organs and excluding the volunteer’s arms.

Image reconstruction: 1D projection profiles were first generated by summing each 2D spiral projection plane along the foot-head axis, and respiratory motion was then derived from the 1D projection profiles from a coil-element that is close to the diaphragm. The self-navigation trace was extracted using 3 different methods: (i) 1D image registration using the log difference as similarity metric, (ii) center of mass, and (iii) by computing the 1st principle component in the time domain. The k-space data were sorted into 7 different respiratory bins and each bin was reconstructed separately using the non-Cartesian conjugate gradient-SENSE (CG-SENSE). Reconstruction time was about 10 min per bin on a Linux server running Matlab. A CUDA-accelerated nonuniform FFT toolbox was used for gridding operation (5).

Figure 2 shows high quality projection profiles that provide sufficient repetition rate to capture respiratory motion. Among the three motion extraction algorithms, the registration approach provided the most reliable tracking of respiratory motion signal, comparing to the true displacement values in the projection profiles. Figure 3 shows a reformatted 3D dataset reconstructed using 35 projections at end expiratory phase. The CG-SENSE method leads to improved image quality with notable reduction of streak artifacts, which are present in the direct reconstruction. Figure 4 compares the averaged reconstruction with the motion-resolved reconstruction at end-inspiration and end-expiration states. Motion sorting leads to reduction of overall blurring. Signal intensity profiles taken along the liver-lung interface show respiration-related displacements of order 10-15 mm, which is consistent with previously reported values.

Our new acquisition approach that utilizes a hybrid spiral projection trajectory enables reliable tracking of physiological motion and efficient motion-resolved imaging of the whole abdominal volume with 1.8 mm isotropic resolution in 1:12 min scan time. Such a fast acquisition can provide comprehensive models for PET motion correction without affecting routine clinical workflow.

Future work includes evaluating this technique for cardiac motion estimation, as well as improving image reconstruction using multi-dimensional compressed sensing techniques (4).