RF Pulse Predistortion for Low-Field MRI Based on Spin Physics Using a Neural Network Amplifier-to-Bloch Equation Model

In low-field MR, low-cost amplifiers with reduced fidelity are used. Current solutions to this problem require additional hardware and do not optimize the pulse for the final excitation profile. The proposed methodology models the amplifier with a neural network, connects the network to a Bloch simulator, and optimizes the pulses for the desired excitation profile. In simulation, a windowed sinc pulse (time-bandwidth product = 10, phase = 0 radians, duration = 1 ms) was optimized to minimize the loss between the target profile and the generated profile after the pulse passes through the amplifier to 0.05% of the starting loss.