Please use this identifier to cite or link to this item: https://scidar.kg.ac.rs/handle/123456789/12820
Full metadata record
DC FieldValueLanguage
dc.contributor.authorSustersic, Tijana-
dc.contributor.authorAnic, Milos-
dc.contributor.authorFilipovic, Nenad-
dc.date.accessioned2021-04-20T21:49:12Z-
dc.date.available2021-04-20T21:49:12Z-
dc.date.issued2020-
dc.identifier.urihttps://scidar.kg.ac.rs/handle/123456789/12820-
dc.description.abstract© 2020 IEEE. Automatic segmentation of the heart left ventricle (LV) is an important step in setting an adequate diagnostic in echocardiography. Some of the state-of-the-art methods for 2D segmentation include traditional methods like active shape models, active contours, level sets, Kalman filter etc., but also deep modern learning methods (i.e. convolutional neural networks), where accuracy usually surpasses the accuracy of traditional methods. Due to the promising results of convolutional neural network called U-net in different segmentation problems, we propose it for the extraction of the left heart ventricle. The results show that the network has been able to segment the left ventricle with the accuracy of around 83.5% on unseen data which surpasses the reported state-of-the-art results, even with a smaller database. Larger database will enable better learning that we are confident will contribute to even higher accuracy. Future work will include testing on larger databases in order to meet the needs for Big Data analysis, but pertain the accuracy and reduce the time necessary for manual analysis of images.-
dc.rightsrestrictedAccess-
dc.source20th IEEE Mediterranean Electrotechnical Conference, MELECON 2020 - Proceedings-
dc.titleHeart left ventricle segmentation in ultrasound images using deep learning-
dc.typeconferenceObject-
dc.identifier.doi10.1109/MELECON48756.2020.9140527-
dc.identifier.scopus2-s2.0-85089279522-
Appears in Collections:Faculty of Engineering, Kragujevac

Page views(s)

130

Downloads(s)

6

Files in This Item:
File Description SizeFormat 
PaperMissing.pdf
  Restricted Access
29.86 kBAdobe PDFThumbnail
View/Open


Items in SCIDAR are protected by copyright, with all rights reserved, unless otherwise indicated.