Volume 8, Issue 3 (12-2021)
jhbmi 2021, 8(3): 315-325 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Ghayoumi Zadeh H, Fayazi A, Rezaee K, Gholizadeh M H, Eskandari M. Segmentation of the Left Atrial Appendage in the Echocardiographic Images of the Heart Using a Deep Neural Network. jhbmi 2021; 8 (3) :315-325
URL: http://jhbmi.ir/article-1-612-en.html
URL: http://jhbmi.ir/article-1-612-en.html
Hossein Ghayoumi Zadeh * 
, Ali Fayazi , Khosro Rezaee , Mohammad Hossein Gholizadeh , Mehdi Eskandari
, Ali Fayazi , Khosro Rezaee , Mohammad Hossein Gholizadeh , Mehdi Eskandari
PhD. in Biomedical Engineering, Assistant Professor, Department of Electrical Engineering, Vali-e-Asr University of Rafsanjan, Rafsanjan, Iran
Abstract: (2089 Views)
Introduction: Cardiovascular diseases are one of the leading causes of mortality in today’s industrial world. Occlusion of left atrial appendage (LAA) using the manufactured devices is a growing trend. The objective of this study was to develop a computer-aided diagnosis system for the identification of LAA in echocardiographic images.
Method: The data used in this descriptive analytical study included 3D echocardiographic images of the heart of 32 patients in King’s College Hospital in London. All patients were treated successfully using the LAA closure device. A total of 208 two-dimensional images were first obtained from each 3D echocardiographic image data set. Then, 1914 images in which the LAA region was clearly recognizable were selected for this study. The proposed neural network was compiled based on the YOLOv3 algorithm. Finally, 1369 and 545 images were used for training and testing the algorithm, respectively.
Results: The performance of the algorithm in detecting the LAA on a set of 545 images was compared with the regions detected in similar images by an expert in echocardiography through intersection over union (IOU). The algorithm was able to correctly identify the LAA region in all 545 examined images with an average IOU of 99.37%.
Conclusion: The proposed image-based algorithm could detect LAA region in echocardiographic images with a high accuracy. This method can be used to develop algorithms for automatic analysis of the LAA region to determine the size of the closure device and to plan an efficient procedure in LAA occlusion methods.
Method: The data used in this descriptive analytical study included 3D echocardiographic images of the heart of 32 patients in King’s College Hospital in London. All patients were treated successfully using the LAA closure device. A total of 208 two-dimensional images were first obtained from each 3D echocardiographic image data set. Then, 1914 images in which the LAA region was clearly recognizable were selected for this study. The proposed neural network was compiled based on the YOLOv3 algorithm. Finally, 1369 and 545 images were used for training and testing the algorithm, respectively.
Results: The performance of the algorithm in detecting the LAA on a set of 545 images was compared with the regions detected in similar images by an expert in echocardiography through intersection over union (IOU). The algorithm was able to correctly identify the LAA region in all 545 examined images with an average IOU of 99.37%.
Conclusion: The proposed image-based algorithm could detect LAA region in echocardiographic images with a high accuracy. This method can be used to develop algorithms for automatic analysis of the LAA region to determine the size of the closure device and to plan an efficient procedure in LAA occlusion methods.
Keywords: Echocardiography, Deep Neural Network, Convolutional Neural Network (CNN), Atrial Fibrillation (AF), Left Atrial Appendages (LAA)
Type of Study: Original Article |
Subject:
Artificial Intelligence in Healthcare
Received: 2021/07/9 | Accepted: 2021/11/15
Received: 2021/07/9 | Accepted: 2021/11/15
Audio File [MP3 4970 KB] (33 Download)
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
