Optimization of Quantitative Sequences Using Automatic Differentiation of Bloch Simulations

We apply automatic differentiation (widely used in deep learning) and a simple Bloch simulation to efficiently optimize any sequence that can be simulated. Automatic differentiation allows us to optimize sequences without an analytical expression for the magnetization.

 

The code is available here.

 

Automatic differentiation allows us to optimize Magnetic Resonance Fingerprinting sequences an order of magnitude faster than conventional methods with 10x fewer lines of code.

 

Lee PK, Watkins LE, Anderson TI, Buonincontri G, Hargreaves BA. Flexible and efficient optimization of quantitative sequences using automatic differentiation of Bloch simulations. Magn Reson Med. 2019 May 26. doi: 10.1002/mrm.27832.

Online Journal Article

Parameter maps from the fully sampled MRF acquisition for the CRLB optimized and conventional sequences are shown. White and grey matter ROIs are outlined in red. Zoomed versions of the ROIs outlined in blue show that the spatial variation of the parameter estimate is reduced in the CRLB optimized sequence. The effect is most prominent in the T2 maps.

Ph.D. Student in Electrical Engineering, admitted Autumn 2016
Professor of Radiology (Radiological Sciences Laboratory) and, by courtesy, of Electrical Engineering and of Bioengineering

Contact Information

Lucas Center for Imaging 
1201 Welch Rd, Stanford, CA 94305-5488

Directions: Lucas Ctr. or Porter Dr. Locations