Deep Learning for assessing rotational misalignment in echocardiographic imaging

Fernandes, Patricia ORCID logoORCID: https://orcid.org/0009-0000-9720-2829, Naidoo, Preshen ORCID logoORCID: https://orcid.org/0009-0004-8328-6194, Ufumaka, Isreal ORCID logoORCID: https://orcid.org/0000-0002-8906-4822, Adibzadeh, Sara ORCID logoORCID: https://orcid.org/0009-0003-3656-916X, Alajrami, Eman ORCID logoORCID: https://orcid.org/0000-0002-8656-5541, Jevsikov, Jevgeni ORCID logoORCID: https://orcid.org/0009-0004-2070-7289, Dadashiserej, Nasim ORCID logoORCID: https://orcid.org/0000-0002-2898-1926, Howard, James ORCID logoORCID: https://orcid.org/0000-0002-9989-6331, Shun-Shin, Matthew ORCID logoORCID: https://orcid.org/0000-0002-1179-0867, Manisty, Charlotte ORCID logoORCID: https://orcid.org/0000-0003-0245-7090, Francis, Darrel ORCID logoORCID: https://orcid.org/0000-0002-3410-0814 and Zolgharni, Massoud ORCID logoORCID: https://orcid.org/0000-0003-0904-2904 (2025) Deep Learning for assessing rotational misalignment in echocardiographic imaging. In: International Conference on AI in Healthcare, 8-10 September 2025, Cambridge, United Kingdom.

[thumbnail of 278-291.pdf] PDF
278-291.pdf - Accepted Version
Restricted to Repository staff only until 25 August 2026.

Download (2MB) | Request a copy

Abstract

Accurate assessment of image quality in echocardiography is essential for both clinical interpretation and the performance of automated diagnostic tools. Rotational misalignment in apical four-chamber views is a common yet underexplored quality issue that can significantly impair anatomical interpretation and quantitative analysis. In this study, we propose a deep learning based framework for automated evaluation of rotational image quality in echocardiographic images. Leveraging a multi-image ranking annotation strategy, we trained a regression model on expert-annotated data. The model exhibited strong alignment with expert consensus, achieving Spearman’s correlation coefficients exceeding 0.88 across multiple validation sets. Comparative analysis demonstrated that model performance was on par with individual expert assessments. Additionally, a training set size analysis revealed performance plateauing beyond approximately 1,000 labelled samples, offering practical guidance for efficient annotation. These findings highlight the feasibility of scalable, objective, and clinically meaningful rotational quality assessment, with promising applications in real-time feedback, acquisition guidance, and automated quality control in echocardiographic workflows.

Item Type: Conference or Workshop Item (Paper)
ISBN: 9783032006561
Identifier: 10.1007/978-3-032-00656-1_20
Page Range: pp. 269-282
Identifier: 10.1007/978-3-032-00656-1_20
Keywords: Image quality assessment; rotational misalignment; expert consensus; echocardiography; deep learning
Subjects: Computing > Intelligent systems
Date Deposited: 03 Oct 2025 10:15
Last Modified: 03 Oct 2025 15:00
URI: https://repository.uwl.ac.uk/id/eprint/14143
Sustainable Development Goals: Goal 3: Good Health and Well-Being

Actions (login required)

View Item View Item

Menu