Deep learning based segmentation of retinal layers from OCT Imaging

Alexander Baine; Abbas Haider; Ruth Hogg; Richard Gault

doi:10.21251/8043511f-bf93-4b36-9348-0726af0987f6

Deep learning based segmentation of retinal layers from OCT Imaging

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

20 Downloads (Pure)

Abstract

Optical Coherence Tomography (OCT) is a key modality for diagnosing retinal disease. Accurate segmentation of retinal layers is essential for quantifying biomarkers linked to age-related macular degeneration, diabetic retinopathy, and glaucoma. This study compares two deep learning architectures, U-Net and DeepLabV3+, for retinal layer segmentation. Both were trained on the OCT5k dataset using identical preprocessing and hyperparameters. Performance was assessed using standard segmentation metrics and visual inspection. DeepLabV3+ showed improved anatomical detail preservation, suggesting its suitability for precision tasks.

Original language	English
Title of host publication	Irish Machine Vision and Image Processing Conference Proceedings 2025
Publisher	Ulster University
Pages	125-218
Number of pages	4
ISBN (Electronic)	9780993420795
DOIs	https://doi.org/10.21251/8043511f-bf93-4b36-9348-0726af0987f6
Publication status	Published - 04 Sept 2025
Event	27th Irish Machine Vision and Image Processing Conference - Ulster University, Derry, United Kingdom Duration: 01 Sept 2025 → 03 Sept 2025 Conference number: 27 https://imvipconference.github.io/#

Publication series

Name	Proceedings of the Irish Machine Vision and Image Processing Conference
ISSN (Print)	2326-2354
ISSN (Electronic)	2732-4494

Conference

Conference	27th Irish Machine Vision and Image Processing Conference
Abbreviated title	IMVIP 2025
Country/Territory	United Kingdom
City	Derry
Period	01/09/2025 → 03/09/2025
Internet address	https://imvipconference.github.io/#

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being

Keywords

optical coherence tomography
OCT
segmentation
deep learning
U-net
DeepLabV3+

Access to Document

10.21251/8043511f-bf93-4b36-9348-0726af0987f6Licence: CC BY

Deep learning based segmentation of retinal layers from OCT Imaging
© 2025 The Authors. This is an open access article published under a Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium, provided the author and source are cited.
Final published version, 3.28 MBLicence: CC BY

1 Conference contribution
1 Poster

ConsensusXAI: a framework to examine class-wise agreement in medical imaging
Haider, A., Wright, D., Hogg, R., Wang, H., Peto, T. & Gault, R., 05 May 2026, 2026 IEEE Winter Conference on Applications of Computer Vision (WACV): Proceedings. IEEE, (IEEE/CVF Winter Conference on Applications of Computer Vision (WACV): Proceedings).
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Open Access
File
14 Downloads (Pure)
OCT-F: the lightweight, data-efficient, foundational model for OCT image analysis
Gault, R., Baine, A., Haider, A., Wright, D., Hogg, R., Wang, H. & Peto, T., 03 May 2026.
Research output: Contribution to conference › Poster

Cite this

@inproceedings{c8b07339bc5f475f9e7cbac9059a3923,

title = "Deep learning based segmentation of retinal layers from OCT Imaging",

abstract = "Optical Coherence Tomography (OCT) is a key modality for diagnosing retinal disease. Accurate segmentation of retinal layers is essential for quantifying biomarkers linked to age-related macular degeneration, diabetic retinopathy, and glaucoma. This study compares two deep learning architectures, U-Net and DeepLabV3+, for retinal layer segmentation. Both were trained on the OCT5k dataset using identical preprocessing and hyperparameters. Performance was assessed using standard segmentation metrics and visual inspection. DeepLabV3+ showed improved anatomical detail preservation, suggesting its suitability for precision tasks. ",

keywords = "optical coherence tomography, OCT, segmentation, deep learning, U-net, DeepLabV3+",

author = "Alexander Baine and Abbas Haider and Ruth Hogg and Richard Gault",

year = "2025",

month = sep,

day = "4",

doi = "10.21251/8043511f-bf93-4b36-9348-0726af0987f6",

language = "English",

series = "Proceedings of the Irish Machine Vision and Image Processing Conference",

publisher = "Ulster University",

pages = "125--218",

booktitle = "Irish Machine Vision and Image Processing Conference Proceedings 2025",

note = "27th Irish Machine Vision and Image Processing Conference, IMVIP 2025 ; Conference date: 01-09-2025 Through 03-09-2025",

url = "https://imvipconference.github.io/\#",

}

Baine, A, Haider, A, Hogg, R & Gault, R 2025, Deep learning based segmentation of retinal layers from OCT Imaging. in Irish Machine Vision and Image Processing Conference Proceedings 2025. Proceedings of the Irish Machine Vision and Image Processing Conference, Ulster University, pp. 125-218, 27th Irish Machine Vision and Image Processing Conference, Derry, United Kingdom, 01/09/2025. https://doi.org/10.21251/8043511f-bf93-4b36-9348-0726af0987f6

Deep learning based segmentation of retinal layers from OCT Imaging. / Baine, Alexander; Haider, Abbas; Hogg, Ruth et al.
Irish Machine Vision and Image Processing Conference Proceedings 2025. Ulster University, 2025. p. 125-218 (Proceedings of the Irish Machine Vision and Image Processing Conference).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Deep learning based segmentation of retinal layers from OCT Imaging

AU - Baine, Alexander

AU - Haider, Abbas

AU - Hogg, Ruth

AU - Gault, Richard

N1 - Conference code: 27

PY - 2025/9/4

Y1 - 2025/9/4

N2 - Optical Coherence Tomography (OCT) is a key modality for diagnosing retinal disease. Accurate segmentation of retinal layers is essential for quantifying biomarkers linked to age-related macular degeneration, diabetic retinopathy, and glaucoma. This study compares two deep learning architectures, U-Net and DeepLabV3+, for retinal layer segmentation. Both were trained on the OCT5k dataset using identical preprocessing and hyperparameters. Performance was assessed using standard segmentation metrics and visual inspection. DeepLabV3+ showed improved anatomical detail preservation, suggesting its suitability for precision tasks.

AB - Optical Coherence Tomography (OCT) is a key modality for diagnosing retinal disease. Accurate segmentation of retinal layers is essential for quantifying biomarkers linked to age-related macular degeneration, diabetic retinopathy, and glaucoma. This study compares two deep learning architectures, U-Net and DeepLabV3+, for retinal layer segmentation. Both were trained on the OCT5k dataset using identical preprocessing and hyperparameters. Performance was assessed using standard segmentation metrics and visual inspection. DeepLabV3+ showed improved anatomical detail preservation, suggesting its suitability for precision tasks.

KW - optical coherence tomography

KW - OCT

KW - segmentation

KW - deep learning

KW - U-net

KW - DeepLabV3+

U2 - 10.21251/8043511f-bf93-4b36-9348-0726af0987f6

DO - 10.21251/8043511f-bf93-4b36-9348-0726af0987f6

M3 - Conference contribution

T3 - Proceedings of the Irish Machine Vision and Image Processing Conference

SP - 125

EP - 218

BT - Irish Machine Vision and Image Processing Conference Proceedings 2025

PB - Ulster University

T2 - 27th Irish Machine Vision and Image Processing Conference

Y2 - 1 September 2025 through 3 September 2025

ER -

Deep learning based segmentation of retinal layers from OCT Imaging

Abstract

Publication series

Conference

UN SDGs

Keywords

Access to Document

Fingerprint

Research output

ConsensusXAI: a framework to examine class-wise agreement in medical imaging

OCT-F: the lightweight, data-efficient, foundational model for OCT image analysis

Cite this