Predicting mortality after transcatheter aortic valve replacement using preprocedural CT.

Area of Science:

• Cardiology
• Radiology
• Artificial Intelligence

Background:

• Transcatheter aortic valve replacement (TAVR) is a key treatment for severe aortic stenosis.
• Identifying patients at high risk for post-TAVR complications is essential.
• Current risk assessment relies on manual clinical and radiological evaluations of CT images, which are time-consuming.

Purpose of the Study:

• To develop and validate an automated probabilistic model for predicting post-TAVR mortality.
• To integrate preprocedural CT image data with patient characteristics for improved risk stratification.
• To address challenges posed by missing data in TAVR planning.

Main Methods:

• A 3D deep neural network was used to extract features from CT volumes of the aortic root and ascending aorta.
• The model automatically localized regions of interest within CT scans.
• A probabilistic structure was implemented to handle missing CT images or measurements, integrating these with 25 baseline patient characteristics.

Main Results:

• The model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.725 in predicting all-cause mortality.
• Performance was comparable to expert radiological assessments.
• The study analyzed a cohort of 1449 TAVR patients.

Conclusions:

• The developed AI model can automatically analyze CT scans and patient data to predict TAVR mortality.
• This automated approach shows potential for efficient and accurate risk assessment in TAVR patients.
• Findings suggest a valuable tool for clinical decision-making in TAVR procedures.