Related Experiment Video
Updated: Aug 7, 2025
Author Spotlight: Advancements in Intracardiac Echocardiography for Atrial Anatomy Assessment
Published on: June 30, 2023
Automatic Segmentation of the Left Atrium from Computed Tomography Angiography Images
Amaan Kazi1,2, Sage Betko3,2, Anish Salvi1,2
1University of Pittsburgh, 302 Benedum Hall, 3700 O'Hara Street, Pittsburgh, PA, 15213, USA.
Insights
Automated segmentation of the left atrium and appendage (LAA) using deep learning significantly improves stroke risk assessment in atrial fibrillation patients. This AI approach reduces variability and speeds up analysis of LAA geometry from CT scans.
Area of Science:
- Medical imaging analysis
- Cardiovascular imaging
- Artificial intelligence in medicine
Background:
- The left atrial appendage (LAA) is a primary source of thrombi in atrial fibrillation (AF) patients, increasing stroke risk.
- Accurate segmentation of the left atrium (LA) and LAA from computed tomography angiography (CTA) is crucial for stroke risk stratification but is time-consuming and prone to inter-observer variability.
Purpose of the Study:
- To develop and evaluate automated 3D U-Net models for segmenting the LA and LAA from CTA images.
- To compare the performance of a unified-image-volume U-Net model with a patch-volume U-Net model for LA/LAA segmentation.
Main Methods:
- Training two distinct 3D U-Net models: one on the entire image volume and another on regional patch-volumes.
- Utilizing binary masks of the LA and corresponding CTA images for model training and testing.
- Evaluating model performance using Dice Similarity Coefficients (DSC) on training and testing datasets.
Main Results:
- The unified-image-volume U-Net achieved median DSCs of 0.92 (train) and 0.88 (test).
- The patch-volume U-Net achieved median DSCs of 0.90 (train) and 0.89 (test).
- Both models demonstrated high accuracy in capturing the complex boundaries of the LA/LAA, with the patch-volume model capturing 89% of regional complexity.
Conclusions:
- Deep learning models can automate LA/LAA segmentation, significantly reducing analysis time and inter-observer variability.
- Automated segmentation facilitates rapid assessment of LA/LAA shape, aiding in more efficient stroke risk stratification for AF patients.
- The developed AI models show high fidelity in segmenting LA/LAA structures, supporting clinical decision-making.
Abstract:
The left atrial appendage (LAA) causes 91% of thrombi in atrial fibrillation patients, a potential harbinger of stroke. Leveraging computed tomography angiography (CTA) images, radiologists interpret the left atrium (LA) and LAA geometries to stratify stroke risk. Nevertheless, accurate LA segmentation remains a time-consuming task with high inter-observer variability. Binary masks of the LA and their corresponding CTA images were used to train and test a 3D U-Net to automate LA segmentation. One model was trained using the entire unified-image-volume while a second model was trained on regional patch-volumes which were run for inference and then assimilated back into the full volume. The unified-image-volume U-Net achieved median DSCs of 0.92 and 0.88 for the train and test sets, respectively; the patch-volume U-Net achieved median DSCs of 0.90 and 0.89 for the train and test sets, respectively. This indicates that the unified-image-volume and patch-volume U-Net models captured up to 88 and 89% of the LA/LAA boundary's regional complexity, respectively. Additionally, the results indicate that the LA/LAA were fully captured in most of the predicted segmentations. By automating the segmentation process, our deep learning model can expedite LA/LAA shape, informing stratification of stroke risk.
