Use this link to cite:
https://hdl.handle.net/2183/50153 HMSANet: A Hierarchical Multi-Scale Attention Network for Precise Retinal Layer Segmentation in COVID-19 OCT Images
Loading...
Identifiers
Publication date
Authors
Saleh, Radhwan A.A.
Álvarez-Rodríguez, Lorena
Cordón Ciordia, Beatriz
García-Martín, Elena
Aptoula, Erchan
Advisors
Other responsabilities
Journal Title
Bibliographic citation
R. A. A. Saleh, L. Álvarez-Rodríguez, B. Cordon, E. García-Martín, E. Aptoula, and J. De Moura., "HMSANet: A Hierarchical Multi-Scale Attention Network for Precise Retinal Layer Segmentation in COVID-19 OCT Images," 2026 International Joint Conference on Neural Networks (IJCNN), Maastricht, Netherlands, 2026, pp. 1-8, doi: 10.1109/IJCNN69854.2026.11728446
Type of academic work
Academic degree
Abstract
[Abstract] The COVID-19 pandemic has underscored the systemic impact of the virus, including retinal abnormalities that may affect vision. Precise segmentation of retinal layers in optical coherence tomography (OCT) images is essential for reliable diagnosis and monitoring. We present the Hierarchical Multi-Scale Axial Attention Network (HMSANet), a novel deep learning architecture specifically designed for retinal layer segmentation in COVID-19 OCT data. HMSANet integrates two main modules: the Multi-Path Inception Attention Module (MPIAM) for multi-scale feature extraction, and the Multi-Scale Attention Module (MSAM) for adaptive attention weighting across spatial scales. This design enables the model to capture both fine-grained and large-scale retinal changes associated with COVID-19. Experimental results show that HMSANet consistently outperforms U-Net and SCAttNet by at least 1.2% in several retinal layers, demonstrating improved sensitivity to subtle structural alterations. By automating retinal layer segmentation, HMSANet enhances diagnostic speed and consistency, with potential applicability to other retinal diseases.
Description
Presented at: 2026 International Joint Conference on Neural Networks (IJCNN), 22-26 June 2026, Maastricht, Netherlands
© 2026 IEEE. This version of the article has been accepted for publication, after peer review. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. The Version of Record is available online at: https://doi.org/10.1109/IJCNN69854.2026.11728446
Editor version
Rights
Copyright © 2026, IEEE







