Artifact-robust Deep Learning-based Segmentation of 3D Phase-contrast MR Angiography: A Novel Data Augmentation...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Pulsation artifacts in 3D phase-contrast magnetic resonance angiography (PC-MRA) images degrade segmentation accuracy.
• Deep learning (DL) methods show promise for medical image segmentation but are sensitive to image artifacts.
• Accurate segmentation of vascular structures in PC-MRA is crucial for clinical diagnosis and treatment planning.

Purpose of the Study:

• To develop and evaluate a novel data augmentation method for DL-based segmentation of PC-MRA images affected by pulsation artifacts.
• To improve the robustness and accuracy of DL segmentation models in the presence of simulated pulsation artifacts.
• To compare the performance of the augmented DL approach against a traditional level-set segmentation algorithm.

Main Methods:

• A novel data augmentation technique was developed by simulating pulsation artifacts via periodic errors in k-space magnitude.
• Deep learning segmentation models were trained and evaluated with and without the proposed pulsation artifact augmentation.
• The performance was assessed on PC-MRA datasets from 16 volunteers, comparing against a level-set algorithm using Dice-Sørensen coefficient, Intersection over Union, and Average Symmetric Surface Distance.

Main Results:

• Deep learning segmentation methods significantly outperformed the level-set algorithm in accuracy.
• Pulsation artifact augmentation further enhanced DL segmentation accuracy, particularly for images with lower velocity encoding.
• Quantitative metrics confirmed the effectiveness of the proposed data augmentation technique in improving segmentation results.

Conclusions:

• The novel data augmentation approach effectively improves deep learning-based segmentation of PC-MRA images corrupted by pulsation artifacts.
• This technique offers a promising solution for enhancing vascular segmentation in clinical applications of PC-MRA, especially in challenging cases.
• The findings suggest broader applicability for this augmentation strategy in medical imaging where artifact reduction is critical.