Deep Learning-Based Analysis of Aortic Morphology From Three-Dimensional MRI.

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate quantification of aortic morphology is crucial for diagnosing and monitoring aortic diseases.
• Current manual measurement methods are time-consuming and prone to variability.
• Automated solutions are needed to improve the quality and reproducibility of aortic assessments.

Purpose of the Study:

• To develop a deep learning (DL)-based automated approach for detecting aortic landmarks and lumen from 3D MRI.
• To enable efficient and reproducible quantification of aortic morphology.

Main Methods:

• A retrospective study involving 391 individuals (healthy and patients with aortic diseases) using 3D MRI data.
• A two-stage deep learning network trained with reinforcement learning for landmark detection and segmentation.
• Evaluation using Dice similarity coefficient (DSC), Hausdorff distance (HD), average symmetrical surface distance (ASSD), and Euclidean distance (ED).

Main Results:

• The DL approach achieved high segmentation accuracy (DSC: 0.90 ± 0.05) and precise landmark detection (ED: 5.0 ± 6.1 mm).
• A strong agreement was observed between DL-derived and reference aortic indices (r > 0.95, mean bias < 7%).
• Performance metrics were comparable to inter-observer variability, except for the sinotubular junction landmark.

Conclusions:

• A novel DL-based method for automated aortic segmentation and landmark detection from 3D MRI was successfully developed.
• This approach facilitates objective and reproducible aortic morphology evaluation for clinical applications.
• The automated system demonstrates potential to enhance the assessment of patients with aortic conditions.