Fully Automated Aortic Root Localization and Tilt Alignment in Cardiac Computed Tomography

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Automated analysis of cardiac computed tomography (CCT) aids personalized management and outcome prediction in transcatheter aortic valve replacement (TAVR).
• Current CCT analysis methods often require manual selection of the region of interest, limiting efficiency.
• An object-oriented aortic root detection pipeline is needed to address these limitations.

Purpose of the Study:

• To develop and evaluate a fully automated object-oriented pipeline for aortic root detection in pre-TAVR CCT studies.
• To assess the performance of a convolutional neural network for accurate aortic root identification.
• To evaluate an automated method for tilt angle prediction for improved procedural planning.

Main Methods:

• Retrospective collection of CCT data from 179 eligible TAVR patients.
• Utilized a pretrained convolutional neural network for automated aortic root detection.
• Employed intensity thresholding, connected component, and principal component analyses for tilt alignment.

Main Results:

• The automated pipeline achieved high detection performance with recall, precision, and F1 scores of 99.0%.
• Mean average precision (mAP) at 50% overlap was 99.5%, with mAP 50%-95% at 60.4%.
• The tilt prediction algorithm demonstrated a mean error of 7.9°, comparable to interobserver variability.

Conclusions:

• A fully automated pipeline for aortic root detection and analysis in pre-TAVR CCTs shows robust performance.
• The developed pipeline can enhance personalized patient management and outcome prediction for TAVR.
• Further prospective studies are warranted to integrate this technology into clinical practice.