Deep learning-based velocity antialiasing of 4D-flow MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Imaging

Background:

• Velocity aliasing is a common artifact in 4D-flow MRI.
• This artifact can lead to inaccurate blood flow quantification.
• Accurate flow assessment is crucial for diagnosing and managing cardiovascular diseases.

Purpose of the Study:

• To develop a convolutional neural network (CNN) for robust and fast correction of velocity aliasing in 4D-flow MRI.
• To evaluate the performance of the CNN against conventional anti-aliasing algorithms.

Main Methods:

• A CNN was trained on 4D-flow MRI data from 667 adult subjects, including data with simulated and existing velocity aliasing.
• The CNN's performance was compared to a conventional velocity anti-aliasing algorithm using Dice scores.
• Control data with varied velocity-encoding sensitivity (vencs) was used to validate CNN correction accuracy.

Main Results:

• The CNN demonstrated excellent performance on simulated data (Dice scores: 0.89-0.99) and detected significantly more aliased voxels in existing aliasing datasets compared to the conventional algorithm (p < 0.001).
• Correction time was comparable to the conventional algorithm (CNN: 176 ± 30s vs. conventional: 162 ± 14s).
• For control data, the CNN achieved high Dice scores (0.98 and 0.96 for 60 and 100 cm/s vencs, respectively) with moderate-to-excellent flow agreement.

Conclusions:

• Deep learning, specifically a CNN, enables fast and robust velocity anti-aliasing in 4D-flow MRI.
• The developed CNN offers a promising tool for improving the accuracy of cardiovascular flow quantification.
• This AI-driven approach enhances the clinical utility of 4D-flow MRI by mitigating aliasing artifacts.