Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation

Zhengji Liu; Tsz-Kin Law; Jizhou Li; Chi-Ho To; Rachel Ka-Man Chun

doi:10.1109/ICIP49359.2023.10223125

Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation

Zhengji Liu, Tsz-Kin Law, Jizhou Li, Chi-Ho To, Rachel Ka-Man Chun^*

^*Corresponding author for this work

Department of Data Science

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

Abstract

Spectral-domain optical coherence tomography (SD-OCT) is a high-speed ocular imaging technology that is commonly employed in eye examinations to visualize the back structures of the eyes. OCT volume containing a sequence of cross-sectional images can be captured in seconds. However, the low signal-to-noise ratio (SNR) prevents accurate result interpretation. To obtain a high SNR OCT volume, numerous images must be averaged at each imaging depth, which is time-consuming. Subjects, especially children, who have short attention spans, may significantly hinder the data collection procedure. Most of the current algorithms focus on single-frame processing without using inter-frame information. Here we developed a lightweight 3D-UNet with a self-supervised strategy to denoise the low SNR OCT volume. This method does not require noisy-clean pairs and can be accomplished by simply measuring a volume containing multiple OCT images. The proposed method improves image quality with structural details preserved and achieves state-of-the-art performance on real OCT datasets. © 2023 IEEE.

Original language	English
Title of host publication	2023 IEEE International Conference on Image Processing, ICIP 2023 - Proceedings
Publisher	IEEE
Pages	3334-3338
ISBN (Electronic)	9781728198354
ISBN (Print)	978-1-7281-9836-1
DOIs	https://doi.org/10.1109/ICIP49359.2023.10223125
Publication status	Published - 2023
Event	30th IEEE International Conference on Image Processing (ICIP 2023) - Kuala Lumpur Convention Centre, Kuala Lumpur, Malaysia Duration: 8 Oct 2023 → 11 Oct 2023 https://2023.ieeeicip.org/

Publication series

Name	Proceedings - International Conference on Image Processing, ICIP
ISSN (Print)	1522-4880

Conference

Conference	30th IEEE International Conference on Image Processing (ICIP 2023)
Abbreviated title	IEEE ICIP 2023
Place	Malaysia
City	Kuala Lumpur
Period	8/10/23 → 11/10/23
Internet address	https://2023.ieeeicip.org/

Funding

This work was supported by grants from the CityU grant (9229120), PolyU grants (P0034099, P0035514), Health and Medical Research Fund (P0036308), RCSV (P0039545), InnoHK Initiative and Hong Kong Special Administrative Region Government.

Research Keywords

3D-UNet
multi-frames image denoising
Noise2Noise
Spectral-domain Optical Coherence Tomography

Access Output

10.1109/ICIP49359.2023.10223125

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

Liu, Z., Law, T.-K., Li, J., To, C.-H., & Chun, R. K.-M. (2023). Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation. In 2023 IEEE International Conference on Image Processing, ICIP 2023 - Proceedings (pp. 3334-3338). (Proceedings - International Conference on Image Processing, ICIP). IEEE. https://doi.org/10.1109/ICIP49359.2023.10223125

@inproceedings{c174f7e83ef54b38a8257e98ec241f40,

title = "Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation",

abstract = "Spectral-domain optical coherence tomography (SD-OCT) is a high-speed ocular imaging technology that is commonly employed in eye examinations to visualize the back structures of the eyes. OCT volume containing a sequence of cross-sectional images can be captured in seconds. However, the low signal-to-noise ratio (SNR) prevents accurate result interpretation. To obtain a high SNR OCT volume, numerous images must be averaged at each imaging depth, which is time-consuming. Subjects, especially children, who have short attention spans, may significantly hinder the data collection procedure. Most of the current algorithms focus on single-frame processing without using inter-frame information. Here we developed a lightweight 3D-UNet with a self-supervised strategy to denoise the low SNR OCT volume. This method does not require noisy-clean pairs and can be accomplished by simply measuring a volume containing multiple OCT images. The proposed method improves image quality with structural details preserved and achieves state-of-the-art performance on real OCT datasets. {\textcopyright} 2023 IEEE.",

keywords = "3D-UNet, multi-frames image denoising, Noise2Noise, Spectral-domain Optical Coherence Tomography",

author = "Zhengji Liu and Tsz-Kin Law and Jizhou Li and Chi-Ho To and Chun, {Rachel Ka-Man}",

year = "2023",

doi = "10.1109/ICIP49359.2023.10223125",

language = "English",

isbn = "978-1-7281-9836-1",

series = "Proceedings - International Conference on Image Processing, ICIP",

publisher = "IEEE",

pages = "3334--3338",

booktitle = "2023 IEEE International Conference on Image Processing, ICIP 2023 - Proceedings",

address = "United States",

note = "30th IEEE International Conference on Image Processing (ICIP 2023), IEEE ICIP 2023 ; Conference date: 08-10-2023 Through 11-10-2023",

url = "https://2023.ieeeicip.org/",

}

Liu, Z, Law, T-K, Li, J, To, C-H & Chun, RK-M 2023, Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation. in 2023 IEEE International Conference on Image Processing, ICIP 2023 - Proceedings. Proceedings - International Conference on Image Processing, ICIP, IEEE, pp. 3334-3338, 30th IEEE International Conference on Image Processing (ICIP 2023), Kuala Lumpur, Malaysia, 8/10/23. https://doi.org/10.1109/ICIP49359.2023.10223125

Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation. / Liu, Zhengji; Law, Tsz-Kin; Li, Jizhou et al.
2023 IEEE International Conference on Image Processing, ICIP 2023 - Proceedings. IEEE, 2023. p. 3334-3338 (Proceedings - International Conference on Image Processing, ICIP).

Research output: Chapters, Conference Papers, Creative and Literary Works › RGC 32 - Refereed conference paper (with host publication) › peer-review

TY - GEN

T1 - Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation

AU - Liu, Zhengji

AU - Law, Tsz-Kin

AU - Li, Jizhou

AU - To, Chi-Ho

AU - Chun, Rachel Ka-Man

PY - 2023

Y1 - 2023

N2 - Spectral-domain optical coherence tomography (SD-OCT) is a high-speed ocular imaging technology that is commonly employed in eye examinations to visualize the back structures of the eyes. OCT volume containing a sequence of cross-sectional images can be captured in seconds. However, the low signal-to-noise ratio (SNR) prevents accurate result interpretation. To obtain a high SNR OCT volume, numerous images must be averaged at each imaging depth, which is time-consuming. Subjects, especially children, who have short attention spans, may significantly hinder the data collection procedure. Most of the current algorithms focus on single-frame processing without using inter-frame information. Here we developed a lightweight 3D-UNet with a self-supervised strategy to denoise the low SNR OCT volume. This method does not require noisy-clean pairs and can be accomplished by simply measuring a volume containing multiple OCT images. The proposed method improves image quality with structural details preserved and achieves state-of-the-art performance on real OCT datasets. © 2023 IEEE.

AB - Spectral-domain optical coherence tomography (SD-OCT) is a high-speed ocular imaging technology that is commonly employed in eye examinations to visualize the back structures of the eyes. OCT volume containing a sequence of cross-sectional images can be captured in seconds. However, the low signal-to-noise ratio (SNR) prevents accurate result interpretation. To obtain a high SNR OCT volume, numerous images must be averaged at each imaging depth, which is time-consuming. Subjects, especially children, who have short attention spans, may significantly hinder the data collection procedure. Most of the current algorithms focus on single-frame processing without using inter-frame information. Here we developed a lightweight 3D-UNet with a self-supervised strategy to denoise the low SNR OCT volume. This method does not require noisy-clean pairs and can be accomplished by simply measuring a volume containing multiple OCT images. The proposed method improves image quality with structural details preserved and achieves state-of-the-art performance on real OCT datasets. © 2023 IEEE.

KW - 3D-UNet

KW - multi-frames image denoising

KW - Noise2Noise

KW - Spectral-domain Optical Coherence Tomography

UR - http://www.scopus.com/inward/record.url?scp=85180773413&partnerID=8YFLogxK

UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-85180773413&origin=recordpage

U2 - 10.1109/ICIP49359.2023.10223125

DO - 10.1109/ICIP49359.2023.10223125

M3 - RGC 32 - Refereed conference paper (with host publication)

SN - 978-1-7281-9836-1

T3 - Proceedings - International Conference on Image Processing, ICIP

SP - 3334

EP - 3338

BT - 2023 IEEE International Conference on Image Processing, ICIP 2023 - Proceedings

PB - IEEE

T2 - 30th IEEE International Conference on Image Processing (ICIP 2023)

Y2 - 8 October 2023 through 11 October 2023

ER -

Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation

Abstract

Publication series

Conference

Funding

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

DON_RMG: Physics-driven Machine Learning Accelerates Multi-frame Imaging: Algorithms and Applications - RMGS

Cite this

Self-Supervised Denoising of Optical Coherence Tomography with Inter-Frame Representation

Abstract

Publication series

Conference

Funding

Research Keywords

Access Output

Other Access Options

Permanent Link

Other Files and Links

Fingerprint

Projects

DON_RMG: Physics-driven Machine Learning Accelerates Multi-frame Imaging: Algorithms and Applications - RMGS

Cite this