Pulseq-zero: PyPulseq sequence scripts in a differentiable optimization loop

Motivation: A common source of problems in MRI sequence development and optimization is the transfer of the sequence between simulation and measurement. By using a single sequence definition understood by both, this could be avoided. Goal(s): Enable the use of a single definition of an MRI sequence for simulation, optimization of sequence parameters, and measurement. Approach: PyPulseq is integrated with a fast, physics-based simulation, which allows a single sequence definition to be used for both simulation and measurement. Automatic differentiation allows efficient optimization of sequence parameters based on the simulated reconstruction. Results: Timing and variable flip angles of two PyPulseq examples were successfully optimized. Impact: Pulseq is a widely used and vendor agnostic library for writing MRI sequences. With pulseq-zero, existing and new sequence scripts can be simulated and optimized directly, with only minor modifications. The resulting sequences can be measured on real scanners immediately.