Automatic aorta and left ventricle segmentation for TAVI procedure planning using convolutional neural networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Surgery

Background:

• Transcatheter aortic valve implantation (TAVI) is a critical procedure for high-risk patients with aortic valve defects.
• Effective preoperative planning, including accurate valve sizing, is essential to minimize TAVI complications like paravalvular leakage and stroke.
• Current planning involves manual measurements and 3D visualization, which can be time-consuming and prone to error.

Purpose of the Study:

• To develop a fully automatic method for segmenting the aorta and left ventricle.
• To improve the accuracy and efficiency of preoperative planning for TAVI procedures.
• To address the lack of existing methods for parallel aorta and left ventricle segmentation.

Main Methods:

• A fully automatic neural network approach utilizing a 2D U-Net architecture was proposed.
• The convolutional neural network was trained on 44 CTA datasets (22 raw, 22 deformed) and tested on 18 additional datasets.
• Cross-validation was performed on 8 datasets during each epoch of the network's learning process.

Main Results:

• The proposed 2D U-Net model achieved a high precision in segmenting aorta and left ventricle structures.
• A mean Dice coefficient score of 0.95 with a standard deviation of 0.02 was obtained.
• The results demonstrate the effectiveness of the automated segmentation in generating accurate label maps.

Conclusions:

• The developed automated segmentation method significantly enhances preoperative planning for TAVI.
• The neural network approach offers a precise and efficient solution for aorta and left ventricle segmentation.
• This automation has the potential to reduce TAVI-related complications and improve patient outcomes.