Optimizing Selective RF Pulses for Enhanced Signal Stability in Turbo Spin Echo Using a Differentiable Extended Phase Graph Model

ABSTRACT Purpose To improve slice profile consistency across echo trains in turbo spin echo (TSE) imaging, thereby reducing image blurring and increasing the accuracy of multi echo spin echo mapping. Methods Excitation and refocusing RF pulses were optimized for TSE using a differentiable extended phase graph model that incorporates the spinor profiles of the RF pulses to calculate the magnetization slice profile across the echo train. The pulses were optimized using an L‐BFGS algorithm in PyTorch to minimize an error term on the target signal magnitude with singular value regularization to promote similarity. The performance of the optimized pulses was assessed by comparing to time bandwidth‐matched SLR RF pulses. Slice profile consistency was calculated in simulation and in a homogeneous phantom. Images were acquired in vivo to assess blurring artifacts, and measurements were acquired in a NIST phantom and in vivo to assess improvements in accuracy. Results The optimized pulses demonstrated superior performance over time bandwidth‐matched SLR pulses, with a 90% reduction in the standard deviation of the normalized integrated signal at each echo. Optimized pulses increased sharpness in vivo at the edges of CSF and veins perpendicular to the phase‐encoded direction, reduced mapping error in the NIST phantom by 91%, and produced more accurate in vivo maps. Conclusion The optimization method enables flexible design of RF pulses in echo train pulse sequences with consistent slice profiles, achieving a target signal progression while also maintaining a constant phase and FWHM between echoes.