Automatic aortic valve area detection in echocardiography images using convolutional neural networks and U-net architecture for bicuspid aortic valve recognition

Automatic methods for heart disease recognition are a promising asset in precise diagnosis and prevention of complications. Regarding bicuspid aortic valve, for which this field is still limited, accurate aortic valve detection would be an essential step in the procedure of using the most common tes...

Full description

Saved in:
Bibliographic Details
Published in2021 IEEE International Conference on Imaging Systems and Techniques (IST) pp. 1 - 6
Main Authors Giannakaki, Katerina, Moirogiorgou, Konstantia, Zervakis, Michalis, Anousakis-Vlachochristou, Nikolaos, Matsopoulos, George K., Komporozos, Christoforos, Sourides, Vasileios, Katsimagklis, Georgios, Drakopoulou, Maria, Toutouzas, Konstantinos, Avgeropoulou, Catherine, Androulakis, Aristeidis
Format Conference Proceeding
LanguageEnglish
Published IEEE 24.08.2021
Subjects
aortic valve
bicuspid aortic valve
convolutional neural networks
Echocardiography
Heart
Image recognition
Imaging
Predictive models
Sensitivity
U-net
Valves
Online AccessGet full text

Cover

More Information
Summary:Automatic methods for heart disease recognition are a promising asset in precise diagnosis and prevention of complications. Regarding bicuspid aortic valve, for which this field is still limited, accurate aortic valve detection would be an essential step in the procedure of using the most common testing method, echocardiography, to automatically detect this malformation. In this study, we propose using a convolutional neural network with U-net architecture for demarcating the aortic valve area in echocardiography images, as an initial step in automatic bicuspid aortic valve detection. Our model achieved a prediction accuracy of 97%, sensitivity 94%, specificity 98% and Intersection over Union 87%.
DOI:10.1109/IST50367.2021.9651398