Gradient Waveform Pre-emphasis using Deep Neural Network

Motivation: Gradient waveform fidelity is critical in MRI for accurate signal encoding, but system imperfections often lead to waveform deviations that compromise image quality. Goal(s): To develop a deep learning-based method for accurate gradient waveform pre-emphasis that compensates for these system imperfections. Approach: A bi-directional LSTM network iteratively learns the system's gradient response, optimizing input waveforms for high-fidelity output. This was validated using 3D Cones sequences on phantoms and human brains. Results: The output gradient waveform achieved near-perfect alignment with the ideal waveform. Phantom and human brain images demonstrated enhanced spatial uniformity and reduced artifacts, showcasing improved MRI image quality. Impact: This method enhances MRI image quality by reducing artifacts and improving spatial accuracy through gradient waveform pre-emphasis using deep learning.