Fast MRI typically requires fast switching of gradient coils in order to traverse the k-space with the minimal encoding time and to complete the data acquisition in a fraction of a second. However, fast switching gradient fields can cause strong eddy current, which can deteriorate images if effects associated with eddy current are neglected. Several approaches can mitigate this challenge, such as gradient pre-emphasis adjustment¹ and k-space trajectory calibration². Although artifacts due to imperfect gradient magnetic fields can be partially resolved by pulse sequences and image post processing, an accurate estimate of the k-space trajectory is still challenging, because all parameters related to final images do not exactly match between the calibration scan and the actual imaging scan. Different from EPI, spiral imaging is an efficient approach to obtain MRI with high spatiotemporal resolution³ and has been extensively applied to dynamic cardiovascular imaging⁴ and functional MRI⁵. However, spiral imaging drives the gradient coil with high slew rate and strength (particularly when traversing the periphery of the k-space). Consequently eddy-current and concomitant artifacts are prominent when high spatiotemporal resolution is required. Previous studies demonstrated that field probes^6,7 can be applied to correct artifacts due to eddy current, B₀ off-resonance effect and concomitant field. However, how the deviation of a k-space trajectory from the theoretical one changes across slices remains unexplored. Here we use a dense 2D field probe array with 24 probes moving over a 3D volume to estimate magnetic field distribution dynamically and to demonstrate that trajectory calibration is slice-dependent.The construction of our probe system has been reported previously⁸ (Figure 1(a)). The 3D imaging correction was realized by first measuring the NMR signal in all probes using the target pulse sequence and then moving the 2D probe array along the slice direction in steps of 1 cm. This allows up to magnetic field measurements at 360 positions (24 positions per slice; 15 slices). Using a spiral sequence (TR = 800ms, a = 60^o, TE = 60ms, and resolution = 2mm x 2mm x 2mm; 110 T/m/s slew rate), we obtained 20,000 data samples from each probe. The instantaneous frequency caused by gradients during readout was estimated by first taking the ratio between the accrued phases in two consecutive measurements and the dwell time, and then subtracting a bias frequency measured from the FID at the beginning of measurements. At each slice, we calculated the magnetic field distribution at 20,000 instants during the readout. Instantaneous magnetic field across imaging plane was estimated by fitting a polynomial with up to the 3^rd order using probe measurements. To demonstrate the effectiveness of image reconstruction using dynamic magnetic field information, we acquired 7 slices of brain images using the same multi-slice spiral sequence (2mm slice thickness, 8mm slice gap). Finally, image reconstruction using experimentally measured and theoretically designed field information was done by using the weighted correlation method⁹ with field map correction.Figure 1(b) shows the theoretical k-space trajectory and trajectories estimated using probe measurements matched to the imaging slice, 3cm away from the imaging slice, and 7cm away from the imaging slice. Trajectories estimated using the probe measurements matched to the imaging slice and probe measurements at 3cm away from the imaging slice were similar. Both theoretical trajectory and the trajectory estimated with 7cm away were significantly different. Figure 2 shows that the reconstructed images without dynamic field correction and reconstructed image using field information 7cm away both show visible blurring (purple arrow). The reconstructed image using dynamic field information 3cm away or using matched dynamic field information is less blurring. The ratio between l₂ norm of difference image and l₂ norm of corrected image is 19.3%, 5.7% and 17.0% for the uncorrected image, the reconstructed image using 3cm away field information and the reconstructed image using 7cm away field information respectively. Figure 3 shows seven slices of corrected brain images acquired using spiral pulse sequence and GRE images acquired at the same locations.Our results suggest that realistic k-space trajectories varied across slices, and reconstructed image quality degrade as the imaging plane and the calibration plane were away. The results shows that eddy currents and magnetic field gradient are not spatial invariant. They also suggest that conventional trajectory calibration using pulse sequences can be inaccurate, because such a method assumed the calibrated trajectory is the same at all locations inside the imaging volume.