Improving Accuracy of Intravoxel Incoherent Motion Reconstruction using Kalman Filter in Combination with Neural Networks: A Simulation Study.

Abstract

Background: The intravoxel Incoherent Motion (IVIM) model extracts perfusion map and diffusion coefficient map using diffusion-weighted imaging. The main limitation of this model is inaccuracy in the presence of noise.

Objective: This study aims to improve the accuracy of IVIM output parameters.

Material and methods: In this simulated and analytical study, the Kalman filter is applied to reject artifact and measurement noise. The proposed method purifies the diffusion coefficient from blood motion and noise, and then an artificial neural network is deployed in estimating perfusion parameters.

Results: Based on the T-test results, however, the estimated parameters of the conventional method were significantly different from actual values, those of the proposed method were not substantially different from actual. The accuracy of f and D* also was improved by using Artificial Neural Network (ANN) and their bias was minimized to 4% and 12%, respectively.

Conclusion: The proposed method outperforms the conventional method and is a promising technique, leading to reproducible and valid maps of D, f, and D*.