Naidoo, Preshen ORCID: https://orcid.org/0009-0004-8328-6194, Fernandes, Patricia ORCID: https://orcid.org/0009-0000-9720-2829, Dadashi Serej, Nasim ORCID: https://orcid.org/0000-0002-2898-1926, Manisty, Charlotte H. ORCID: https://orcid.org/0000-0003-0245-7090, Shun-Shin, Matthew J. ORCID: https://orcid.org/0000-0002-1179-0867, Francis, Darrel P. ORCID: https://orcid.org/0000-0002-3410-0814 and Zolgharni, Massoud ORCID: https://orcid.org/0000-0003-0904-2904 (2025) Consensus-guided evaluation of self-supervised learning in echocardiographic segmentation. Computers in Biology and Medicine, 198. p. 111148. ISSN 00104825

PDF 2025-MultiX.pdf - Published Version Restricted to Repository staff only Available under License Creative Commons Attribution. Download (3MB)

Abstract

Background:
Left ventricle segmentation is a fundamental task in echocardiography, essential for assessing cardiac function. However, deep learning models for segmentation rely on large labelled datasets, which are expensive and time-consuming to annotate. Self-supervised learning has emerged as a promising approach to leverage unlabelled data, but its effectiveness for left ventricle segmentation remains underexplored.
Methods:
This study investigates self-supervised learning for echocardiographic segmentation, comparing various pretext tasks. The impact of dataset size and distribution on pre-training is examined, revealing that excessive unlabelled data can degrade performance due to redundancy and low variability. A novel multi-expert labelled dataset is introduced to enhance segmentation evaluation, using consensus-based annotations to reduce annotation noise and improve reliability.
Results:
Among the self-supervised learning methods evaluated, contrastive learning consistently outperforms other approaches, particularly in low-label settings. The study demonstrates that AI models pre-trained using self-supervised learning and fine-tuned with only 15% of labelled data achieve stronger alignment with multi-expert consensus than any individual expert.
Conclusion:
The findings suggest that AI models can generalise well across expert annotations, providing more reliable and reproducible assessments.

Item Type:	Article
Identifier:	10.1016/j.compbiomed.2025.111148
Keywords:	Self-supervised learning; Contrastive learning; Left ventricle segmentation; Ejection fraction
Subjects:	Medicine and health
Date Deposited:	02 Oct 2025 16:45
Last Modified:	02 Oct 2025 17:00
URI:	https://repository.uwl.ac.uk/id/eprint/14141
Sustainable Development Goals:	Goal 3: Good Health and Well-Being

Actions (login required)

View Item