The training of advanced deep learning algorithms for medical image interpretation requires precisely annotated datasets, which is laborious and expensive. Therefore, this research investigates state-of-the-art active learning methods for utilising limited annotations when performing automated left ventricle segmentation in echocardiography. Our experiments reveal that the performance of different sampling strategies varies between datasets from the same domain. Further, an optimised method for representativeness sampling is introduced, combining images from feature-based outliers to the most representative samples for label acquisition. The proposed method significantly outperforms the current literature and demonstrates convergence with minimal annotations. We demonstrate that careful selection of images can reduce the number of images needed to be annotated by up to 70%. This research can therefore present a cost-effective approach to handling datasets with limited expert annotations in echocardiography.
CITATION STYLE
Alajrami, E., Naidoo, P., Jevsikov, J., Lane, E., Pordoy, J., Serej, N. D., … Zolgharni, M. (2023). Deep Active Learning for Left Ventricle Segmentation in Echocardiography. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 13958 LNCS, pp. 283–291). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-35302-4_29
Mendeley helps you to discover research relevant for your work.