Context-Based Masking for Spontaneous Venous Pulsations Detection

Sheng, H; Yu, X; Li, X; Golzan, M

Context-Based Masking for Spontaneous Venous Pulsations Detection

Sheng, H Yu, X Li, X Golzan, M

Permalink

Publisher:: SPRINGER-VERLAG SINGAPORE PTE LTD
Publication Type:: Chapter
Citation:: AI 2023: Advances in Artificial Intelligence, 2024, 14471 LNAI, pp. 520-532
Issue Date:: 2024-01-01

Recently Added

	Filename	Description	Size
	venous.pdf	Published version	3.94 MB	Adobe PDF	View/Open

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is new to OPUS and is not currently available.

Full metadata record

Field	Value	Language
dc.contributor.author	Sheng, H
dc.contributor.author	Yu, X
dc.contributor.author	Li, X
dc.contributor.author	Golzan, M https://orcid.org/0000-0002-4479-3917
dc.contributor.editor	Liu, T
dc.contributor.editor	Yue, L
dc.contributor.editor	Webb, G
dc.contributor.editor	Wang, D
dc.date.accessioned	2024-07-19T03:11:26Z
dc.date.available	2024-07-19T03:11:26Z
dc.date.issued	2024-01-01
dc.identifier.citation	AI 2023: Advances in Artificial Intelligence, 2024, 14471 LNAI, pp. 520-532
dc.identifier.isbn	978-981-99-8387-2
dc.identifier.uri	http://hdl.handle.net/10453/179797
dc.description.abstract	Spontaneous retinal venous pulsations (SVP) serve as vital dynamic biomarkers, representing rhythmic changes of the central retinal vein observed at the optic disc region (ODR) within an eye. SVPs serve as vital dynamic biomarkers, representing rhythmic changes of the central retinal vein observed at the optic disc region (ODR) within an eye. In light of their crucial clinical role, automatic detection of SVPs from fundus videos has become an area of burgeoning research. However, the inherent eye movements and the variability in retinal video quality present significant challenges to direct SVP detection via existing deep learning models. In response, we devise a spatio-temporal context-based masking approach (STC Masking), exploiting the spatiotemporal characteristics of SVPs to enhance their detection in retinal videos. We first apply a spatio-temporal mask to clip the video into an ODR-focused video tube. Diverging from conventional masking with gray or black blocks, we then employ a context masking method which using the original pixel values from video frames as the mask fill-in. The context mask map temporally transforms the dynamic video tubes into static tubes, thus changing the pulsation status of SVPs. Correspondingly, we adjust the SVP video labels based on the changing extent of masked regions to avoid ambiguity in data labelling. This innovative strategy provides more vivid videos which are similar to unmasked videos pixel-wise but having contrast semantics in SVP presenting regions. This enables network to capture the most discriminating regions through spatio-temporal variations, allowing explicit detection on SVP existence in the video. Our experiments illustrate the efficacy of our STC masking strategy, outperforming baseline methods. This work, thereby, underscores the potential of grid context-based masking for more accurate SVP detection in retinal video analysis.
dc.language	en
dc.publisher	SPRINGER-VERLAG SINGAPORE PTE LTD
dc.relation	http://purl.org/au-research/grants/arc/DP220100800
dc.relation	http://purl.org/au-research/grants/arc/DE230100477
dc.relation.ispartof	AI 2023: Advances in Artificial Intelligence
dc.relation.ispartofseries	Lecture Notes in Artificial Intelligence
dc.relation.isbasedon	10.1007/978-981-99-8388-9_42
dc.rights	info:eu-repo/semantics/restrictedAccess
dc.subject.classification	Artificial Intelligence & Image Processing
dc.subject.classification	46 Information and computing sciences
dc.title	Context-Based Masking for Spontaneous Venous Pulsations Detection
dc.type	Chapter
utslib.citation.volume	14471 LNAI
pubs.organisational-group	University of Technology Sydney
pubs.organisational-group	University of Technology Sydney/Faculty of Health
pubs.organisational-group	University of Technology Sydney/Strength - CHT - Health Technologies
utslib.copyright.status	recently_added	*
dc.date.updated	2024-07-19T03:11:23Z
pubs.publication-status	Published
pubs.volume	14471 LNAI

Abstract:

Spontaneous retinal venous pulsations (SVP) serve as vital dynamic biomarkers, representing rhythmic changes of the central retinal vein observed at the optic disc region (ODR) within an eye. SVPs serve as vital dynamic biomarkers, representing rhythmic changes of the central retinal vein observed at the optic disc region (ODR) within an eye. In light of their crucial clinical role, automatic detection of SVPs from fundus videos has become an area of burgeoning research. However, the inherent eye movements and the variability in retinal video quality present significant challenges to direct SVP detection via existing deep learning models. In response, we devise a spatio-temporal context-based masking approach (STC Masking), exploiting the spatiotemporal characteristics of SVPs to enhance their detection in retinal videos. We first apply a spatio-temporal mask to clip the video into an ODR-focused video tube. Diverging from conventional masking with gray or black blocks, we then employ a context masking method which using the original pixel values from video frames as the mask fill-in. The context mask map temporally transforms the dynamic video tubes into static tubes, thus changing the pulsation status of SVPs. Correspondingly, we adjust the SVP video labels based on the changing extent of masked regions to avoid ambiguity in data labelling. This innovative strategy provides more vivid videos which are similar to unmasked videos pixel-wise but having contrast semantics in SVP presenting regions. This enables network to capture the most discriminating regions through spatio-temporal variations, allowing explicit detection on SVP existence in the video. Our experiments illustrate the efficacy of our STC masking strategy, outperforming baseline methods. This work, thereby, underscores the potential of grid context-based masking for more accurate SVP detection in retinal video analysis.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/179797