1. Image Reconstruction using Dynamic EPI Phase Correction Magnetic resonance imaging (MRI) studies using echo planar imaging (EPI) employ data acquisition techniques in which data is collected on both positive and negative polarity lobes of readout gradient waveforms. Phase errors are inherent in this data collection scheme and must be corrected for during image reconstruction. One technique used for EPI phase correction is to acquire a set of non-phase encoded data during an EPI reference scan, computing phase correction coefficients to be used during image reconstruction of the subsequent EPI scan(1,2). If phase characteristics change during EPI scans with longer durations, such as functional magnetic resonance imaging (fMRI), diffusion tensor imaging (DTI) and multi-phase EPI scans, increased Nyquist ghosting and image translation in the phase direction may result (3,4,5). We propose a method of dynamic EPI phase correction in which phase characteristics are monitored during an EPI scan, and the original phase correction parameters determined prior to the scan are adjusted to insure consistent image fidelity is maintained throughout the duration of the scan. Conventional EPI reconstruction techniques employ a phase correction scheme where the phase correction coefficients are estimated prior to the scan. Empirical evidence suggests that changes in phase characteristics over the duration of certain lengthy EPI multi-phase scans may lead to image quality degradation over time. One approach for solving this problem is to collect extra non-phase encoded frames of data during an EPI scan. At the beginning of the scan no adjustments are made to phase correction coefficients used during image reconstruction. Reference data acquired during the first several phases is used to establish baseline values from which adjustments are made later on. The extra non-phase encoded reference data frames collected during a scan are processed in a similar manner that is used during an EPI reference scan. After Fourier transformation and phase computation, adjacent frames of data are subtracted from each other. After phase unwrapping, a magnitude weighted least-squares fit is applied to the phase difference data to determine a first order approximation of the phase characteristics for each phase difference frame. It is then possible to compute linear and constant coefficients for each frame of reference data. We propose using at least 3 extra frames of non-phase encoded reference data during an EPI scan to monitor phase errors. Linear coefficients determined from frames collected with positive gradient polarity are averaged, then subtracted from the average of those collected with negative gradient polarity. The resulting linear coefficient for each phase, a t, is subtracted from the baseline value yielding an adjustment factor,  a t, which is added to all linear phase correction coefficients used during image reconstruction for the current phase. Another type of correction compensates for apparent translation of the image in the phase direction. The constant coefficients acquired from two frames of data collected with the same gradient polarity, separated in acquisition space by a single frame of reversed polarity data, are subtracted to determine a value representative of the gap, g t, between these frames of data. This parameter is subtracted from the baseline value yielding an adjustment factor,  g t, which is multiplied by the frame number, and incrementally added to constant phase correction coefficients used during image reconstruction for the current phase. Both the  a t and  g t parameters are filtered in time using a recursive smoothing filter. For multiple-channel scans, the  a[i] t and  g[i] t parameters computed from each channel, i, are averaged and a single adjustment factor is applied to the phase correction coefficients for each channel. Fig. 1: Ghosting levels over time obtained from a single-channel 3T fMRI study without dynamic EPI phase correction for a GE MRS phantom with EPI-GRE, TE=26 msec, TR=2000, FOV=24cm, slice thick=3mm, 36 slices, and 140 phases. Dynamic EPI phase correction provides a robust and efficient means of adjusting linear phase correction coefficients to insure that Nyquist ghosting levels remain consistent during the course of a scan. In addition, image translation in the phase direction due to B0 drift is eliminated by adjusting a constant phase term that increments from echo to echo. Collecting extra reference data during an EPI scan may reduce the maximum number slices that can be acquired within a TR. References 1. CB Ahn and ZH Cho, IEEE Trans on Med Imag, MI-6, no 1, pp 32-36, 1987. 2. RS Hinks, et al. US Patent Application, 60/615,248, Sept 2004. 3. HA Ward, et al., MRM, vol. 43, pp. 459-469, 2000. 4. HA Ward, et al., Proc. of ISMRM, 9th Meeting, p. 295, 2001. 5. S Thesen, et al., Proc. of ISMRM, 11th Meeting, p. 1025, 2003. Data was collected using a 3.0 T General Electric (GE) Signa MR scanner (GE Healthcare, Waukesha, WI, USA) equipped with a high bandwidth (1MHz) data acquisition subsystem and a TwinSpeed gradient coil capable of 40 mT/m at a maximum slew rate of 150T/m/s. Conventional gradient recalled echo (GRE) and spin echo (SE) EPI scans were performed with a standard GE single channel head coil as well as an 8-channel head coil (MRI Devices, Waukesha, WI, USA). Raw data from each experiment was saved and re- processed off-line to show the effects of the new algorithm. Results Introduction Theory Methods Conclusions R. Scott Hinks 1, Fred J. Frigo 1, Bryan J. Mock 1, Xiaoli Zhao 1, Ryan Sangill 1, 2, Bruce D. Collick 1 1 GE Healthcare, Waukesha WI, United States, 2 CFIN, Aarhus University Hospital, Aarhus, Denmark EPI-GRE and EPI-SE pulse sequences were modified to collect extra non-phase- encoded reference data during the EPI scan. During image reconstruction, the non- phase-encoded reference frames of data are Fourier transformed, and phase subtraction is performed to compare the phase of each frame of data to a corresponding frame from the start of the scan. This phase difference data is then row-subtracted from neighboring frames of data, and a magnitude-weighted least-squares fit is used to estimate first-order phase correction coefficients from the phase difference data. These estimates are temporally filtered with a low-pass filter, and then used to adjust the original phase correction coefficients obtained prior to a scan. Fig. 2 shows the results of applying this technique to data obtained with the following parameters: single channel head coil, EPI- GRE, TE=26 msec, TR=2000, FOV=24cm, slice thick=3mm, 36 slices, 140 phases. Fig. 3 shows results of this technique for a human volunteer. Fig. 2: Ghosting levels over time obtained from a single-channel 3T fMRI study with dynamic EPI phase correction for a GE MRS phantom with EPI-GRE, TE=26 msec, TR=2000, FOV=24cm, slice thick=3mm, 36 slices, and 140 phases with the same raw data used for images shown in Fig. 1. Fig. 3: fMRI results from a human volunteer using dynamic EPI phase correction with an 8-channel coil..