Automated MSCT Analysis for Planning Left Atrial Appendage Occlusion Using Artificial Intelligence.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Device Technology

Background:

• Increasing volume of multislice computed tomography (MSCT) analyses for structural heart interventions necessitates automation.
• Current manual analysis methods are time-consuming, lack standardization, and have a steep learning curve.

Purpose of the Study:

• To assess the feasibility of a fully automated artificial intelligence (AI)-based MSCT analysis system.
• To specifically apply and evaluate this system for planning left atrial appendage occlusion (LAAO) procedures.

Main Methods:

• Deep learning models were trained and validated on 583 patient MSCT datasets with manual annotations.
• Models were developed to detect key anatomical structures (ostium, landing zones, mitral valve annulus, fossa ovalis) and segment the left atrium and appendage.
• Accuracy was evaluated against manual measurements and segmentations.

Main Results:

• Automated segmentation of the left atrium/appendage showed high similarity to manual segmentation (Dice score 0.94 ± 0.02).
• Measurements of anatomical ostium and landing zones demonstrated minimal differences compared to manual measurements, comparable to operator variability.
• Detection of the mitral valve annulus and fossa ovalis was accurate, with low Hausdorff distances.
• The complete automated workflow, including pre- and post-processing, averaged 57.5 ± 34.5 seconds per patient.

Conclusions:

• A fast and accurate AI-based workflow for automated MSCT analysis in LAAO planning has been developed.
• This automated approach can significantly aid in managing the increasing patient caseload for structural heart interventions.
• The AI system's capabilities are potentially extensible to other structural heart interventions.