CSE Community Seminar
February 14, 2025, 12-1PM
Conference Room 45-432 in Building 45
The Adjoint Method for MRI Pulse Design: Demonstration in Ritz methods
John Drago
PhD student, Electrical Engineering and Computer Science at MIT
Abstract:
We employ the adjoint method for optimizing parameterized excitation waveforms in magnetic resonance imaging (MRI) to efficiently design subject-specific excitations with a desired magnetization pattern while the subject is in the scanner. The adjoint method facilitates an efficient calculation of the cost function gradient with respect to the waveform parameters. We demonstrate the method’s ability to mitigate flip angle inhomogeneities associated with brain imaging at high field strengths (7 T) using gradient and RF waveforms expressed in a Chebyshev polynomial basis. The target excitation pattern is achieved with system- and safety-related constraints whose gradients are also efficiently calculated with the adjoint method and a rapid GPU waveform optimization. We demonstrate a reduction in optimization times sufficient to enable in-the-scanner, subject-specific pulse design. We initially design pulses over a population of subject fieldmaps to create a “universal pulse” which serves as an initial guess for subject-specific pulse constrained optimization of non-selective excitation and inversion pulses as well as slice-selective excitation pulses. We compare parameterized pulse optimization to conventional piecewise-constant optimization and demonstrate a roughly 5-10-time speed-up for both non-selective excitations and slice-selective excitations.
February 14, 2025, CSE Community Seminar
John Drago
PhD student
Electrical Engineering and Computer Science, MIT