1713
IVIM quantification and b-value optimization using deep neural network
Wonil Lee1, Byungjai Kim1, Jongyeon Lee1, and HyunWook Park1
1KAIST, Daejeon, Korea, Republic of
The trained DNN and the optimized b-values by the proposed method quantified IVIM parameters more accurately than combination of the conventional b-value optimization schemes with DNN fitting method.
Figure 1. Overall diagram of the proposed IVIM quantification method using DNN. a) Diagram for training of DNN and for optimizing of b-values. b) Diagram for quantification of IVIM parameters from the diffusion-weighted MRI signals.
Figure 3. Total parameter errors of the IVIM parameters from the three estimation methods (MS-LSR fitting, Bayesian fitting, and DNN) when the diffusion weighted images were acquired using the b-values optimized by uniform sampling, Jalnefjord’s method, Zhang’s method, and the proposed method. The red error bar indicates the standard deviation obtained from 50 simulations.