Cargando…

HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution

the automatic segmentation of lung infections in CT slices provides a rapid and effective strategy for diagnosing, treating, and assessing COVID-19 cases. However, the segmentation of the infected areas presents several difficulties, including high intraclass variability and interclass similarity am...

Descripción completa

Detalles Bibliográficos
Autores principales: Chen, Ying, Zhou, Taohui, Chen, Yi, Feng, Longfeng, Zheng, Cheng, Liu, Lan, Hu, Liping, Pan, Bujian
Formato: Online Artículo Texto
Lenguaje:English
Publicado: Elsevier Ltd. 2022
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9391231/
https://www.ncbi.nlm.nih.gov/pubmed/36029749
http://dx.doi.org/10.1016/j.compbiomed.2022.105981
_version_ 1784770828348424192
author Chen, Ying
Zhou, Taohui
Chen, Yi
Feng, Longfeng
Zheng, Cheng
Liu, Lan
Hu, Liping
Pan, Bujian
author_facet Chen, Ying
Zhou, Taohui
Chen, Yi
Feng, Longfeng
Zheng, Cheng
Liu, Lan
Hu, Liping
Pan, Bujian
author_sort Chen, Ying
collection PubMed
description the automatic segmentation of lung infections in CT slices provides a rapid and effective strategy for diagnosing, treating, and assessing COVID-19 cases. However, the segmentation of the infected areas presents several difficulties, including high intraclass variability and interclass similarity among infected areas, as well as blurred edges and low contrast. Therefore, we propose HADCNet, a deep learning framework that segments lung infections based on a dual hybrid attention strategy. HADCNet uses an encoder hybrid attention module to integrate feature information at different scales across the peer hierarchy to refine the feature map. Furthermore, a decoder hybrid attention module uses an improved skip connection to embed the semantic information of higher-level features into lower-level features by integrating multi-scale contextual structures and assigning the spatial information of lower-level features to higher-level features, thereby capturing the contextual dependencies of lesion features across levels and refining the semantic structure, which reduces the semantic gap between feature maps at different levels and improves the model segmentation performance. We conducted fivefold cross-validations of our model on four publicly available datasets, with final mean Dice scores of 0.792, 0.796, 0.785, and 0.723. These results show that the proposed model outperforms popular state-of-the-art semantic segmentation methods and indicate its potential use in the diagnosis and treatment of COVID-19.
format Online
Article
Text
id pubmed-9391231
institution National Center for Biotechnology Information
language English
publishDate 2022
publisher Elsevier Ltd.
record_format MEDLINE/PubMed
spelling pubmed-93912312022-08-22 HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution Chen, Ying Zhou, Taohui Chen, Yi Feng, Longfeng Zheng, Cheng Liu, Lan Hu, Liping Pan, Bujian Comput Biol Med Article the automatic segmentation of lung infections in CT slices provides a rapid and effective strategy for diagnosing, treating, and assessing COVID-19 cases. However, the segmentation of the infected areas presents several difficulties, including high intraclass variability and interclass similarity among infected areas, as well as blurred edges and low contrast. Therefore, we propose HADCNet, a deep learning framework that segments lung infections based on a dual hybrid attention strategy. HADCNet uses an encoder hybrid attention module to integrate feature information at different scales across the peer hierarchy to refine the feature map. Furthermore, a decoder hybrid attention module uses an improved skip connection to embed the semantic information of higher-level features into lower-level features by integrating multi-scale contextual structures and assigning the spatial information of lower-level features to higher-level features, thereby capturing the contextual dependencies of lesion features across levels and refining the semantic structure, which reduces the semantic gap between feature maps at different levels and improves the model segmentation performance. We conducted fivefold cross-validations of our model on four publicly available datasets, with final mean Dice scores of 0.792, 0.796, 0.785, and 0.723. These results show that the proposed model outperforms popular state-of-the-art semantic segmentation methods and indicate its potential use in the diagnosis and treatment of COVID-19. Elsevier Ltd. 2022-10 2022-08-20 /pmc/articles/PMC9391231/ /pubmed/36029749 http://dx.doi.org/10.1016/j.compbiomed.2022.105981 Text en © 2022 Elsevier Ltd. All rights reserved. Since January 2020 Elsevier has created a COVID-19 resource centre with free information in English and Mandarin on the novel coronavirus COVID-19. The COVID-19 resource centre is hosted on Elsevier Connect, the company's public news and information website. Elsevier hereby grants permission to make all its COVID-19-related research that is available on the COVID-19 resource centre - including this research content - immediately available in PubMed Central and other publicly funded repositories, such as the WHO COVID database with rights for unrestricted research re-use and analyses in any form or by any means with acknowledgement of the original source. These permissions are granted for free by Elsevier for as long as the COVID-19 resource centre remains active.
spellingShingle Article
Chen, Ying
Zhou, Taohui
Chen, Yi
Feng, Longfeng
Zheng, Cheng
Liu, Lan
Hu, Liping
Pan, Bujian
HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution
title HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution
title_full HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution
title_fullStr HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution
title_full_unstemmed HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution
title_short HADCNet: Automatic segmentation of COVID-19 infection based on a hybrid attention dense connected network with dilated convolution
title_sort hadcnet: automatic segmentation of covid-19 infection based on a hybrid attention dense connected network with dilated convolution
topic Article
url https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9391231/
https://www.ncbi.nlm.nih.gov/pubmed/36029749
http://dx.doi.org/10.1016/j.compbiomed.2022.105981
work_keys_str_mv AT chenying hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution
AT zhoutaohui hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution
AT chenyi hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution
AT fenglongfeng hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution
AT zhengcheng hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution
AT liulan hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution
AT huliping hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution
AT panbujian hadcnetautomaticsegmentationofcovid19infectionbasedonahybridattentiondenseconnectednetworkwithdilatedconvolution