4D segmentation of the thoracic aorta from 4D flow MRI using deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Research

Background:

• 4D flow MRI enables hemodynamic analysis in the aorta, crucial for understanding pathologies like thoracic aortic aneurysms (TAA).
• Personalized TAA management requires novel biomarkers derived from fluid-structure interaction, necessitating accurate 4D aortic segmentation.
• Current methods for generating these biomarkers are hindered by the lack of automated 4D aortic segmentation techniques.

Purpose of the Study:

• To develop an automated deep learning model for segmenting the aorta in 4D flow MRI.
• To enable precise analysis of hemodynamic changes in thoracic aortic aneurysms (TAA).
• To facilitate the creation of new biomarkers for personalized TAA management.

Main Methods:

• A U-Net based deep learning model was employed for 4D aortic segmentation, processing each 4D flow MRI frame independently.
• Segmentation performance was quantitatively assessed using Dice Score (DS) and Hausdorff Distance (HD).
• Maximum and minimum surface areas of the ascending aorta were measured and compared with cine-MRI data.

Main Results:

• The automated segmentation model achieved a Dice Score of 0.90 ± 0.02 and a mean Hausdorff Distance of 9.58 ± 4.36 mm.
• High correlation coefficients (r=0.85 for max surface, r=0.86 for min surface) were observed between 4D flow MRI and cine-MRI measurements.
• The model demonstrated accurate segmentation capabilities for 4D aortic data.

Conclusions:

• The developed automatic 4D aortic segmentation method is sufficiently accurate for clinical application.
• This technique can significantly enhance the utility of 4D flow MRI in analyzing thoracic aortic aneurysm (TAA) pathologies.
• The approach supports wider adoption of 4D flow MRI for improved patient care and research.