Prospective motion correction (PMC) is an effective method to correct for head motion in MRI¹. PMC performs real-time updates of the sequence geometry in accordance with tracking data of the head. Several methods exist to track head motion^2-5. In [3] a dedicated gradient sequence was used to track NMR markers. Another type of NMR markers was tracked by superimposing gradient tones⁵. These two methods come at the expense of either scan-time prolongation or alteration of the sequence. For the special case of EPI sequences with moderate spatial resolution, NMR field probes⁶ were used to implement PMC without sequence alteration⁷, relying on intrinsic high-frequency content within the gradient waveforms. However, the large class of higher-resolved and clinically relevant standard spin-warp sequences cannot be treated by that approach due to probe dephasing, caused by large gradient moments.

In this work we propose a novel strategy for addressing this issue. To achieve this, sufficiently diverse gradient time-courses measured by the probe are needed for spatial encoding. The problem of probe dephasing due to large gradient moments needs to be solved and probe relaxation prior to re-excitation has to be considered. Moreover, low frequency content of the gradient waveforms is unreliable for probe tracking⁴ and should be avoided. The proposed method employs short-lived NMR field probes, which can be re-excited quickly. This provides a means to acquire several short within-TR snippets. These snippets are chosen such that their combination delivers three linearly independent gradient time-courses. As a result, the probe position can be inferred. Fig.1 exemplifies such snippets. A sequence contains high frequencies due to gradient switching, which is mathematically a ramp-function. Hence, a robust position determination is possible. The proposed method is demonstrated in-vivo with high-resolution T2*-weighted gradient echo scans.

Four snippets were chosen as depicted in Fig.1. The snippets are concatenated and the resulting phase-signal time-course is denoted by $$$\phi(t)$$$. The Fourier transform of its temporal derivative at frequency f reads $$$FT[\dot{\phi}(t)]_f=\gamma\cdot(g_{0,f}+xg_{x,f}+yg_{y,f}+zg_{z,f})$$$, where the $$$g_{[x,y,z],f}$$$ denote the Fourier coefficients of the respective gradient signals at frequency f. $$$g_{0,f}$$$ reflects coupling to the B0-field. To determine the $$$g_{i,f}$$$, a calibration was performed by rigidly placing four probes inside the scanner at known positions⁴. Calibration coefficients were obtained for every phase-step of the spin-warp sequence. To compute {x,y,z}, the calibration is required for at least three frequencies. Seven frequencies were chosen in the range of 300-700Hz and 1700-3000Hz. Precision was identified with the RMSE of reconstructed probe positions in a static experiment. By per-TR computing of probe positions, rigid-body transformation estimates were obtained and sent to the scanner.

All scans were performed on a 7T Philips Achieva system (Philips Healthcare, Cleveland, OH) with a 32-channel receive array (Nova Medical, Wilmington, MA). Four 19F NMR field probes (diameter=1.3mm, T1=1.5ms) were used, run with a dedicated acquisition system⁸. For in-vivo PMC they were attached to the head using a setup as shown in Fig.2. With a T2*-weighted gradient echo sequence (parameters: voxel-size =0.5x0.5x2mm³, TR/TE=340/26ms, flip-angle=30°, 8 slices, duration=2:17min) the following scans were performed: a fixed spherical phantom (Agar/NiCl2) with attached probes was scanned with and without PMC to verify identical image quality. In-vivo, for scans without deliberate motion of the subject and with instructed head motion PMC was turned on and off respectively. To further assess the method’s ability to address subtle involuntary motion, a longer T2*-weighted higher-resolution scan was additionally performed with and without PMC (sequence parameters: voxel-size=0.3x0.3x2mm³, TR/TE=720/25ms, flip-angle=45°, 15 slices, duration=9:14min).

The precision with respect to translational and rotational degrees of freedom was assessed to be 10-25µm and 0.008-0.014° (RMS), respectively. This is sufficient to apply PMC to the targeted sequences. Fig.3 shows images of the phantom. The detailed structure of the air-bubbles is reproduced with high fidelity in the corrected scan. Fig.4 displays the motion parameters for the case of instructed motion. In-vivo images are presented in Fig.5. Strong motion artifacts are visible in Fig.5b/c in the uncorrected case, that do not occur in the corrected one. The higher-resolution case in Fig.5d also demonstrates the improved image quality with PMC.Without the need to increase sequence duration or adding gradients, a field-detection based motion correction has successfully been implemented on high-resolution spin-warp sequences. A caveat is that sequences may have short time periods when there is neither phase encoding nor phase spoiling, e.g. at $$$k_y=0$$$. To avoid temporary loss of information along the corresponding coordinate, the spoiler gradient can be off-set. This was done in this work and is applicable unless gradient moments have to be fully balanced (e.g. balanced-SSFP).