Acceleration techniques like Echo-Planar Spectroscopic Imaging (EPSI)^1-3 reduce the long acquisition time of Chemical Shift Imaging (CSI). The “flyback” echo-planar trajectory⁴ is more robust against gradient errors (e.g. eddy currents) compared to the symmetric implementation. There is no analytic solution for the design of an SNR optimized readout/flyback trapezoid pair for a given spectral and spatial resolution. Beside hardware constraints like maximum slew rate and gradient strength of the gradient system physiological constraints to prevent peripheral nerve stimulation (PNS) need to be taken into account for the design of the gradient waveform. Traditionally, a trajectory is designed for a given set of parameters like spectral and spatial resolution and gradient system. This waveform is checked against PNS thresholds and if violating constraints, the trajectory is redesigned with derated gradient specifications (Smax and/or Gmax), iterating the design until an allowed trajectory is reached. Goal of this work is to implement an SNR-optimized algorithm that allows calculation of the gradient waveform directly on-the-fly based on the user prescription (e.g. bandwidth, spatial resolution) and is therefore not limited to pre-calculated waveforms like in⁵. Trapezoids and ramp times for the “flyback” echo planar trajectory (Fig. 1) were calculated using a dB/dt-optimized algorithm following the linear formulation of the allowable change in gradient field ΔG_lim as a function of ramp time Δt in⁶ (ΔG_lim = SR_min (Δt + c)) including parameters of the gradient system (rheobase, chronaxie and effective coil length). This also took into account that the flyback EPSI readout train was just on one logical gradient axis and no gradient trapezoids during that time on the other two. Other constraints for the design of the trapezoids were the gradient system hardware that defines the maximum gradient strength and the minimum rise time for a full amplitude ramp and that only a discrete number of possible receiver bandwidths given by the maximum receiver bandwidth of the system divided by a decimation (integer) number were possible. In addition all gradient times had to be a multiple of 4μs (gradient rate) due to system constraints. The implemented algorithm to calculate the trapezoids maximizes SNR for a given set of input parameters like the field-of-view (fov), the number of sampling points in the spatial direction (res) and the spectral bandwidth which defines the maximum time of the trapezoid pair (T). Therefore the sampling time for a single data point (Δt) was maximized. Trapezoid waveforms were calculated on the host for various prescriptions so no pre-calculated gradient waveforms for a limited number of spatial resolutions were required. This was implemented within a Point-Resolved Spectroscopy (PRESS) sequence.Results for different gradient configurations and settings are shown in Table 1 and compared with results reported previously⁵. The results for the zoom gradient configuration of TRM are the same for SNR Efficiency E_SNR and spectral bandwidth 1/T for both spatial resolutions (5mm and 10mm) shown in⁵ Fig. 1. Changing prescription and re-calculation of waveforms led to no delay meaning the time required for the re-calculation is in the range of a few ms. XRMB configuration (gradient system of MR750, GE Healthcare) shows better performance than the zoom mode of the TRM configuration (gradient system of HDx) used in reference⁵ but is limited by PNS (see column ‘gflyb SR’ in Table 1). PNS evaluation for the full waveform using the convolution model demonstrate no threshold violation (Fig. 2). An example of an in-vivo 2D data set is shown in Fig. 3. A fast algorithm was implemented for “flyback” echo planar trajectory optimization within a PRESS sequence that is not only taking hardware constraints into account but also PNS and therefore supports all different gradient systems. The applied linear formulation for PNS is slightly more conservative than the convolution model for a "flyback" EPSI train and therefore on the safe side. Thus no pre-calculated waveforms for a limited number of spatial resolutions are required.