FRNet: A Full-Resolution Convolutional Neural Network for OCTA Vascular Segmentation

Liang Wang, Dongxu Gao*, Hongwei Gao, Guozhang Jiang, Zhaojie Ju

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

14 Downloads (Pure)

Abstract

Optical coherence tomography angiography (OCTA) is an advanced, non-invasive imaging technique. Automatic segmentation of vascular networks in retinal OCTA images is essential for the early diagnosis and progression assessment of various vision-related diseases. However, most existing methods for OCTA image vessel segmentation rely on encoder-decoder architectures, which typically involve a high number of parameters and result in slower inference speeds. In this paper, we introduce FRNet V2, an accurate and efficient neural network specifically designed for retinal vessel segmentation in OCTA images. This is achieved by integrating a modified recursive ConvNeXt V2 block and a recursive HAAM attention mechanism into a full-resolution convolutional network framework. We evaluate our proposed method on two large public datasets: ROSSA and OCTA-500. Experimental results demonstrate that our network achieves performance on par with other methods, while boasting significantly fewer parameters and faster inference speeds. This makes it highly suitable for practical industrial applications.
Original languageEnglish
Title of host publication2024 30th International Conference on Mechatronics and Machine Vision in Practice (M2VIP)
PublisherInstitute of Electrical and Electronics Engineers Inc.
Number of pages6
ISBN (Electronic)9798350391916
ISBN (Print)9798350391923
DOIs
Publication statusPublished - 12 Nov 2024
Event2024 30th International Conference on Mechatronics and Machine Vision in Practice (M2VIP) - Leeds, United Kingdom
Duration: 3 Oct 20245 Oct 2024

Publication series

NameProceedings of the International Conference on Mechatronics and Machine Vision in Practice (M2VIP)
PublisherIEEE
ISSN (Print)2996-4156
ISSN (Electronic)2996-4164

Conference

Conference2024 30th International Conference on Mechatronics and Machine Vision in Practice (M2VIP)
Period3/10/245/10/24

Keywords

  • Optical coherence tomography angiography
  • Vascular segmentation
  • Neural networks
  • ConvNeXt V2
  • Dataset

Fingerprint

Dive into the research topics of 'FRNet: A Full-Resolution Convolutional Neural Network for OCTA Vascular Segmentation'. Together they form a unique fingerprint.

Cite this