Quantification of intravoxel incoherent motion with optimized b-values using deep neural network.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Biophysics

Background:

• Intravoxel incoherent motion (IVIM) imaging provides insights into tissue microcirculation.
• Accurate quantification of IVIM parameters is crucial for clinical applications.
• Current methods for IVIM parameter estimation can be sensitive to the choice of b-values and noise.

Purpose of the Study:

• To develop a framework for simultaneously optimizing b-values and training a deep neural network (DNN) for accurate IVIM parameter quantification.
• To enhance the precision of intravoxel incoherent motion (IVIM) parameter estimation.

Main Methods:

• A deep neural network (DNN) framework was developed for IVIM parameter quantification.
• Optimal b-values were simultaneously selected during DNN training using Monte Carlo simulations.
• The framework was evaluated for accuracy and noise sensitivity using simulated and in vivo 3T MRI data.

Main Results:

• Simultaneous optimization of b-values and DNN training minimized IVIM parameter quantification errors.
• Perfusion coefficient (Dₚ) and perfusion fraction (f) accuracies were more sensitive to b-value selection than the diffusion coefficient (D).
• Optimized b-values varied with noise levels, highlighting the need to consider noise in the optimization process.

Conclusions:

• The proposed scheme effectively quantifies IVIM parameters by simultaneously optimizing b-values and training DNNs.
• The trained DNN can accurately estimate IVIM parameters from diffusion-weighted images acquired with optimized b-values.
• This approach offers a robust method for precise IVIM analysis in MRI.