OPTIMIZATION OF PARAMETER VALUES FOR COMPLEX PULSE SEQUENCES BY SIMULATED ANNEALING - APPLICATION TO 3D MP-RAGE IMAGING OF THE BRAIN

A number of pulse sequence techniques, including magnetization prepare d gradient echo (MP-GRE), segmented GRE, and hybrid RARE, employ a rel atively large number of variable pulse sequence parameters and acquire the image data during a transient signal evolution. These sequences h ave recently been proposed and/or used for clinical applications in th e brain, spine, liver, and coronary arteries. Thus, the need for a met hod of deriving optimal pulse sequence parameter values for this class of sequences now exists. Due to the complexity of these sequences, co nventional optimization approaches, such as applying differential calc ulus to signal difference equations, are inadequate. We have developed a general framework for adapting the simulated annealing algorithm to pulse sequence parameter value optimization, and applied this framewo rk to the specific case of optimizing the white matter-gray matter sig nal difference for a T-1-weighted variable flip angle 3D MP-RAGE seque nce. Using our algorithm, the values of 35 sequence parameters, includ ing the magnetization-preparation RF pulse flip angle and delay time, 32 flip angles in the variable flip angle gradient-echo acquisition se quence, and the magnetization recovery time, were derived. Optimized 3 D MP-RAGE achieved up to a 130% increase in white matter-gray matter s ignal difference compared with optimized 3D RF-spoiled FLASH with the same total acquisition time. The simulated annealing approach was effe ctive at deriving optimal parameter values for a specific 3D MP-RAGE i maging objective, and may be useful for other imaging objectives and s equences in this general class.