Deep-learning-based image segmentation for image-based computational hemodynamic analysis of abdominal aortic...

Area of Science:

• Medical Imaging
• Computational Fluid Dynamics
• Artificial Intelligence in Medicine

Background:

• Computational hemodynamics analysis of abdominal aortic aneurysms (AAA) requires patient-specific models.
• Manual segmentation of computed tomography angiography (CTA) scans for AAA model creation is time-consuming, hindering clinical application.
• Automated segmentation methods are needed to accelerate the process.

Purpose of the Study:

• To investigate the feasibility of deep-learning-based image segmentation for automating AAA model creation.
• To compare the performance of ARU-Net and CACU-Net for AAA segmentation in computational hemodynamics.
• To assess the time reduction and accuracy of automated segmentation compared to manual methods.

Main Methods:

• Two deep learning models, ARU-Net and CACU-Net, were employed for image segmentation of 30 CTA scans.
• Morphological features and hemodynamic metrics were compared between deep learning and manual segmentation models.
• Performance was evaluated using DICE scores, correlation coefficients, and Bland-Altman analysis.

Main Results:

• Both ARU-Net and CACU-Net achieved a DICE score of 0.916 and correlation above 0.95, indicating high agreement with manual segmentation.
• Bland-Altman analysis confirmed good agreement between automated and manual segmentation results.
• Automated model generation time was reduced from approximately 2 hours to 10 minutes, with CACU-Net showing superior accuracy and speed.

Conclusions:

• Automated image segmentation using deep learning significantly reduces the time and cost associated with patient-specific AAA model creation.
• CACU-Net and ARU-Net are effective deep learning methods for AAA segmentation.
• CACU-Net demonstrates superior performance in accuracy and efficiency for computational hemodynamics applications.