Related Experiment Video
Updated: Jul 8, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Deep learning networks in the segmentation of the left atrial appendage in 2D ultrasound: A comparative analysis
Insights
Deep learning models like AttUnet and TransAttUnet show promise for segmenting the left atrial appendage (LAA) in 2D ultrasound, improving accuracy for crucial device sizing in atrial fibrillation treatment.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Medicine
- Cardiovascular Interventions
Background:
- The left atrial appendage (LAA) is a primary source of thromboembolism in non-valvular atrial fibrillation patients.
- Percutaneous LAA occlusion is a treatment option, but accurate device sizing relies on complex, variable manual image analysis.
- Current methods lack efficient solutions for 2D ultrasound, the standard for intervention planning.
Purpose of the Study:
- To evaluate the performance of deep learning (DL) methods for segmenting the LAA in 2D ultrasound images.
- To compare the efficacy of different DL architectures (Unet, UnetR, AttUnet, TransAttUnet) for this task.
- To assess the feasibility of DL for improving LAA anatomical analysis in clinical practice.
Main Methods:
- A dedicated 2D ultrasound database for LAA segmentation was created.
- Four DL networks (Unet, UnetR, AttUnet, TransAttUnet) were trained and evaluated.
- Performance was quantified using Dice coefficient, Accuracy, Recall, Specificity, Precision, Hausdorff distance, and Average distance error.
Main Results:
- AttUnet and TransAttUnet demonstrated high performance, achieving Dice scores of 88.62% and 89.28%, respectively.
- Accuracy for AttUnet and TransAttUnet was recorded at 88.25% and 86.30%.
- The study confirmed the effectiveness of DL for LAA segmentation in 2D ultrasound.
Conclusions:
- Deep learning methods, particularly AttUnet and TransAttUnet, are effective for LAA segmentation in 2D ultrasound.
- These findings highlight the clinical potential of DL for precise LAA anatomical analysis and intervention planning.
- DL-based segmentation offers a more efficient and potentially less variable alternative to manual analysis for LAA occlusion procedures.
Abstract:
Left atrial appendage (LAA) is the major source of thromboembolism in patients with non-valvular atrial fibrillation. Currently, LAA occlusion can be offered as a treatment for these patients, obstructing the LAA through a percutaneously delivered device. Nevertheless, correct device sizing is a complex task, requiring manual analysis of medical images. This approach is sub-optimal, time-demanding, and highly variable between experts. Different solutions were proposed to improve intervention planning, but, no efficient solution is available to 2D ultrasound, which is the most used imaging modality for intervention planning and guidance. In this work, we studied the performance of recently proposed deep learning methods when applied for the LAA segmentation in 2D ultrasound. For that, it was created a 2D ultrasound database. Then, the performance of different deep learning methods, namely Unet, UnetR, AttUnet, TransAttUnet was assessed. All networks were compared using seven metrics: i) Dice coefficient; ii) Accuracy iii) Recall; iv) Specificity; v) Precision; vi) Hausdorff distance and vii) Average distance error. Overall, the results demonstrate the efficiency of AttUnet and TransAttUnet with dice scores of 88.62% and 89.28%, and accuracy of 88.25% and 86.30%, respectively. The current results demonstrate the feasibility of deep learning methods for LAA segmentation in 2D ultrasound.Clinical relevance- Our results proved the clinical potential of deep neural networks for the LAA anatomical analysis.
