File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Conference Paper: Prediction of large vessel occlusion on non-contrast CT brain using deep machine learning with Siamese neural network

TitlePrediction of large vessel occlusion on non-contrast CT brain using deep machine learning with Siamese neural network
Authors
Issue Date2020
PublisherSage Publications Ltd. The Journal's web site is located at http://www.sagepub.in/journals/Journal202429
Citation
Joint European Stroke Organisation and World Stroke Organisation Conference (ESO-WSO 2020), Virtual Conference, 7-9 November 2020. In International Journal of Stroke, 2020, v. 15 n. 1, Suppl., p. 732 How to Cite?
AbstractBackground And Aims: The diagnosis of acute large vessel occlusion (LVO) requires timely neuroimaging and interpretation of CT angiography, which may not be readily available in resourcelimited healthcare settings. The development of artificial intelligence in neuroimaging analysis may provide a preliminary analysis and prediction of the likelihood of large vessel occlusion in noncontrast CT imaging. This can potentially triage patients with high probability of LVO to appropriate thrombectomy center and streamline the stroke care pathway, and improve the utilization of CT angiography. Methods: Acute ischemic stroke patients from 2016 to 2018 with CT and CT angiography performed within 3 hours of admission were retrospectively identified from the territory-wide public hospital database. Those with large vessel occlusion stroke (ground truth) were based on clinical history and CT angiography findings, verified by 2 experienced neuroradiologists. Dense MCA signs and early ischemic change on CT were demarcated when present. A computer model was built using the Siamese neural network technique to analyze the initial non-contrast CT images to predict large vessel occlusion. Results: Amongst 324 acute ischemic stroke patients with complete clinical, CT and CT angiography data, 165 were large vessel occlusion stroke and 121 had dense MCA sign on the noncontrast CT. Using Siamese neural network model, the algorithm predicted LVO based only on the non-contrast CT images with an AUC of 0.90, sensitivity of 88%, specificity of 67% and accuracy of 77%. Conclusions: Machine learning with Siamese neural network can predict LVO stroke based on noncontrast CT with high accuracy, and can potentially be applied clinically after prospective validation.
Description02776 / #2159: E-Poster Viewing - AS38. Technology Innovations in Stroke
Persistent Identifierhttp://hdl.handle.net/10722/308197
ISSN
2023 Impact Factor: 6.3
2023 SCImago Journal Rankings: 1.800

 

DC FieldValueLanguage
dc.contributor.authorTsang, COA-
dc.contributor.authorYou, J-
dc.contributor.authorYu, PLH-
dc.contributor.authorTsui, E-
dc.contributor.authorLui, WM-
dc.contributor.authorLeung, G-
dc.date.accessioned2021-11-12T13:43:51Z-
dc.date.available2021-11-12T13:43:51Z-
dc.date.issued2020-
dc.identifier.citationJoint European Stroke Organisation and World Stroke Organisation Conference (ESO-WSO 2020), Virtual Conference, 7-9 November 2020. In International Journal of Stroke, 2020, v. 15 n. 1, Suppl., p. 732-
dc.identifier.issn1747-4930-
dc.identifier.urihttp://hdl.handle.net/10722/308197-
dc.description02776 / #2159: E-Poster Viewing - AS38. Technology Innovations in Stroke-
dc.description.abstractBackground And Aims: The diagnosis of acute large vessel occlusion (LVO) requires timely neuroimaging and interpretation of CT angiography, which may not be readily available in resourcelimited healthcare settings. The development of artificial intelligence in neuroimaging analysis may provide a preliminary analysis and prediction of the likelihood of large vessel occlusion in noncontrast CT imaging. This can potentially triage patients with high probability of LVO to appropriate thrombectomy center and streamline the stroke care pathway, and improve the utilization of CT angiography. Methods: Acute ischemic stroke patients from 2016 to 2018 with CT and CT angiography performed within 3 hours of admission were retrospectively identified from the territory-wide public hospital database. Those with large vessel occlusion stroke (ground truth) were based on clinical history and CT angiography findings, verified by 2 experienced neuroradiologists. Dense MCA signs and early ischemic change on CT were demarcated when present. A computer model was built using the Siamese neural network technique to analyze the initial non-contrast CT images to predict large vessel occlusion. Results: Amongst 324 acute ischemic stroke patients with complete clinical, CT and CT angiography data, 165 were large vessel occlusion stroke and 121 had dense MCA sign on the noncontrast CT. Using Siamese neural network model, the algorithm predicted LVO based only on the non-contrast CT images with an AUC of 0.90, sensitivity of 88%, specificity of 67% and accuracy of 77%. Conclusions: Machine learning with Siamese neural network can predict LVO stroke based on noncontrast CT with high accuracy, and can potentially be applied clinically after prospective validation.-
dc.languageeng-
dc.publisherSage Publications Ltd. The Journal's web site is located at http://www.sagepub.in/journals/Journal202429-
dc.relation.ispartofInternational Journal of Stroke-
dc.relation.ispartofJoint European Stroke Organisation and World Stroke Organisation Conference (ESO-WSO 2020)-
dc.titlePrediction of large vessel occlusion on non-contrast CT brain using deep machine learning with Siamese neural network-
dc.typeConference_Paper-
dc.identifier.emailTsang, COA: acotsang@hku.hk-
dc.identifier.emailYu, PLH: plhyu@hku.hk-
dc.identifier.authorityTsang, COA=rp01519-
dc.identifier.authorityYu, PLH=rp00835-
dc.description.natureabstract-
dc.identifier.hkuros329409-
dc.identifier.volume15-
dc.identifier.issue1, Suppl.-
dc.identifier.spage732-
dc.identifier.epage732-
dc.publisher.placeUnited Kingdom-
dc.identifier.partofdoi10.1177/1747493020963387-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats