| A new strategy is developed for designing magnetic resonance imaging (MRI) gradient waveforms. It describes a waveform as a set of discrete amplitudes and selects such amplitudes and other design parameters (e.g. repetition time, echo time, sampling time, number of repetitions, and flip angle) to optimize several design criteria subject to imaging conditions and hardware restrictions. The design criteria are selected to achieve images optimal in terms of contrast, resolution, slice selection, field-of-view, chemical shift, motion sensitivity and waveform slewings, while minimizing the total acquisition time.; All design criteria are modeled as objective functions, and relevant imaging requirements and hardware restrictions are modeled as constraints. The resulting multiobjective model is highly nonlinear due mostly to the complicating timing parameters, and large due to the gradient amplitudes. Artificial constraints imposed by traditional multi-lobe design procedures are eliminated.; An effective multilevel procedure based on variable decomposition is developed where the top level master program contains the sequence and timing parameters, which are identified as complicating variables. The remaining variables are handled in three separate quadratically constrained linear programming (QCLP) subproblems. A dual two level method is developed for solving these and the more general QCQP.; A semi-interactive goal-based method is developed to help the waveform designer find the best compromise solution. The method avoids punishment of over-attainment while using available trade-off information for computing satisfactory and locally feasible improving search steps. An event driven design support system with tools for evaluating the achieved solutions and for making trade-offs between different objectives has also been implemented.; Gradient waveforms have been designed for cervical spine imaging and imaging of fatty infiltrated liver; generated images have been compared to those of a standard sequence. Theoretical contrast per unit time was predicted to improve by 152%, with 70% a verified clinically, and sensitivity to several moments was reduced while keeping resolution, chemical shift and field-of-view fixed. Clinical tests in both cases, particularly for the liver, show the importance of the RF pulse profile at high flip angles (above 90°), tissue parameters, and field inhomogeneity. |
