# Deep learning models for segmentation and quantification of left atrial appendage volume using noncontrast cardiac computed tomography

## TL;DR

This paper evaluates deep learning models for accurately measuring left atrial appendage volume in cardiac CT scans without contrast, improving cardiovascular risk assessment.

## Contribution

The study introduces and compares novel U-Net-based deep learning models for semiautomated segmentation of the left atrial appendage in noncontrast CT scans.

## Key findings

- 3D Attention-UNet achieved the highest correlation (r = 0.800) between predicted and manual LAA volumes.
- All models showed strong segmentation accuracy with Dice coefficients above 77%.
- Bland–Altman analysis confirmed consistent reliability across all architectures.

## Abstract

The left atrial appendage (LAA) is a critical but frequently overlooked site of thrombus formation, reinforcing the need for accurate identification in routine cardiac imaging. This process is related to pathological dilation associated with endothelial injury and a proinflammatory status. This study assesses the performance of deep learning architectures based on U-Net, specifically UNet3D, Residual-UNet3D, 3D Attention-UNet, and Res16-PAC-UNet, in the semiautomated segmentation and volume measurement of LAA.

We retrospectively analyzed noncontrast cardiac computed tomography (NCCT) scans from 452 patients aged ≥ 60 years, acquired for chest pain evaluation, to compare the performance of four U-Net–based deep learning architectures (UNet3D, Residual-UNet3D, 3D Attention-UNet, and Res16-PAC-UNet) for semiautomated LAA segmentation and volume measurement. Segmentation accuracy was assessed with the Dice coefficient, and volumetric agreement with Pearson correlation and Bland–Altman analysis.

Dice coefficients were 78.44 ± 1.93 for UNet3D, 78.97 ± 0.79 for Residual-UNet3D, 79.07 ± 1.43 for 3D Attention-UNet, and 77.68 ± 1.47 for Res16-PAC-UNet. All models showed strong correlations between predicted and manual volumes (P < 0.001), with the highest in 3D Attention-UNet (r = 0.800). Bland–Altman analysis indicated minimal bias and narrow limits of agreement for all architectures, confirming consistent reliability.

Deep learning–based segmentation on NCCT enables accurate, reproducible LAA morphological and volumetric assessment without contrast, offering a rapid and reliable tool to support cardiovascular risk stratification and treatment planning.

The online version contains supplementary material available at 10.1186/s44348-025-00058-1.