US 12,320,879 B2
Method and apparatus for reconstructing an electrical conductivity map from magnetic resonance imaging
Dong-Hyun Kim, Seoul (KR); and Kyu-Jin Jung, Seoul (KR)
Assigned to UIF (University Industry Foundation), Yonsei University, Seoul (KR)
Filed by UIF (University Industry Foundation), Yonsei University, Seoul (KR)
Filed on Nov. 4, 2022, as Appl. No. 18/052,808.
Claims priority of application No. 10-2021-0165765 (KR), filed on Nov. 26, 2021.
Prior Publication US 2023/0168328 A1, Jun. 1, 2023
Int. Cl. G01R 33/56 (2006.01); A61B 5/055 (2006.01); G01R 33/50 (2006.01); G01R 33/58 (2006.01); G06T 7/00 (2017.01)
CPC G01R 33/5608 (2013.01) [A61B 5/055 (2013.01); G01R 33/50 (2013.01); G01R 33/58 (2013.01); G06T 7/0012 (2013.01); G06T 2207/10088 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30016 (2013.01)] 5 Claims
OG exemplary drawing
 
1. A method for training an electrical conductivity reconstruction network for reconstructing a brain electrical conductivity map, comprising:
generating radio frequency (RF) magnetic field data, current density data, and electric field data using a simulation software and a human head phantom;
generating a virtual magnetic resonance imaging (MRI) image and a corresponding ground-truth electrical conductivity map from the RF magnetic field data, the current density data, and the electric field data;
separating the virtual MRI image into real number data and imaginary number data based on generating the virtual MRI image;
adding noise to the real number data and the imaginary number data, respectively, and integrating noise-added real number data and noise-added imaginary number data to generate noise-added virtual MRI image;
separating the noise-added virtual MRI image into magnitude data and phase data;
acquiring patched image kernels from the noise-added virtual MRI image, based on separating the noise-added virtual MRI image; and
training the electrical conductivity reconstruction network, which is an artificial neural network comprising an input layer, a hidden layer, and an output layer, using the patched image kernels and electrical conductivities of kernel points of the ground-truth electrical conductivity map, which correspond to the patched image kernels, as training data and label data, respectively.