TAVI-PREP: A Deep Learning-Based Tool for Automated Measurements Extraction in TAVI Planning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Cardiovascular Surgery

Background:

• Transcatheter aortic valve implantation (TAVI) is a minimally invasive procedure for severe aortic stenosis.
• Careful preoperative planning is crucial to mitigate procedural risks associated with TAVI.

Purpose of the Study:

• To introduce TAVI-PREP, an automated deep learning system for pre-TAVI planning.
• To extract key measurements of the aortic valvular complex from CT scans.

Main Methods:

• TAVI-PREP utilizes MeshDeformNet for 3D mesh generation and 3D Residual U-Net for landmark detection.
• The algorithm was trained on public and private datasets, analyzing 200 CT scans.
• Measurements were compared against manual annotations by expert cardiologists, with inter-operator variability assessed.

Main Results:

• High correlation (0.90-0.97) was found between TAVI-PREP and expert measurements for most parameters.
• Mean absolute relative error was below 5% for most measurements, with exceptions for coronary height.
• The algorithm demonstrated no bias, and expert consensus exceeded agreement with the automated approach.

Conclusions:

• TAVI-PREP delivers dependable and rapid measurements for aortic valvular complex assessment.
• The system can significantly aid clinicians in the preprocedural planning of TAVI.