Publicação

Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI

Ver documento

Detalhes bibliográficos
Resumo:Atrial fibrillation (AF), is the most frequent sustained cardiac arrhythmia, described by an irregular and rapid contraction of the two upper chambers of the heart (the atria). AF development is promoted and predisposed by atrial dilation, which is a consequence of atria adaptation to AF. However, it is not clear whether atrial dilation appears similarly over the cardiac cycle and how it affects ventricular volumes. Catheter ablation is arguably the AF gold standard treatment. In their current form, ablations are capable of directly terminating AF in selected patients but are only first-time effective in approximately 50% of the cases. In the first part of this work, volumetric functional markers of the left atrium (LA) and left ventricle (LV) of AF patients were studied. More precisely, a customised convolutional neural network (CNN) was proposed to segment, across the cardiac cycle, the LA from short axis CINE MRI images acquired with full cardiac coverage in AF patients. Using the proposed automatic LA segmentation, volumetric time curves were plotted and ejection fractions (EF) were automatically calculated for both chambers. The second part of the project was dedicated to developing classification models based on cardiac MR images. The EMIDEC STACOM 2020 challenge was used as an initial project and basis to create binary classifiers based on fully automatic classification neural networks (NNs), since it presented a relatively simple binary classification task (presence/absence of disease) and a large dataset. For the challenge, a deep learning NN was proposed to automatically classify myocardial disease from delayed enhancement cardiac MR (DE-CMR) and patient clinical information. The highest classification accuracy (100%) was achieved with Clinic-NET+, a NN that used information from images, segmentations and clinical annotations. For the final goal of this project, the previously referred NNs were re-trained to predict AF recurrence after catheter ablation (CA) in AF patients using pre-ablation LA short axis in CINE MRI images. In this task, the best overall performance was achieved by Clinic-NET+ with a test accuracy of 88%. This work shown the potential of NNs to interpret and extract clinical information from cardiac MRI. If more data is available, in the future, these methods can potentially be used to help and guide clinical AF prognosis and diagnosis.
Autores principais:Lourenço, Ana Catarina Feliciano
Assunto:Fibrilhação Auricular Enfarte do miocárdio Ressonância Magnética Cardíaca Deep learning Classificação/ Segmentação Teses de mestrado - 2021
Ano:2021
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade de Lisboa
Idioma:inglês
Origem:Repositório da Universidade de Lisboa
_version_ 1865920832600014849
author Lourenço, Ana Catarina Feliciano
author_facet Lourenço, Ana Catarina Feliciano
Lourenço, Ana Catarina Feliciano
author_role author
contributor_name_str_mv Caetano, Gina Maria Costa
Varela, Marta
Repositório Científico de Acesso Aberto da ULisboa
country_str PT
creators_json_str [{\"Person.name\":\"Lourenço, Ana Catarina Feliciano\"}]
datacite.contributors.contributor.contributorName.fl_str_mv Caetano, Gina Maria Costa
Varela, Marta
Repositório Científico de Acesso Aberto da ULisboa
datacite.creators.creator.creatorName.fl_str_mv Lourenço, Ana Catarina Feliciano
datacite.date.Accepted.fl_str_mv 2021-01-01T00:00:00Z
datacite.date.available.fl_str_mv 2021-06-11T15:02:52Z
datacite.date.embargoed.fl_str_mv 2021-06-11T15:02:52Z
datacite.rights.fl_str_mv http://purl.org/coar/access_right/c_abf2
datacite.subjects.subject.fl_str_mv Fibrilhação Auricular
Enfarte do miocárdio
Ressonância Magnética Cardíaca
Deep learning
Classificação/ Segmentação
Teses de mestrado - 2021
datacite.titles.title.fl_str_mv Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
dc.contributor.none.fl_str_mv Caetano, Gina Maria Costa
Varela, Marta
Repositório Científico de Acesso Aberto da ULisboa
dc.creator.none.fl_str_mv Lourenço, Ana Catarina Feliciano
dc.date.Accepted.fl_str_mv 2021-01-01T00:00:00Z
dc.date.available.fl_str_mv 2021-06-11T15:02:52Z
dc.date.embargoed.fl_str_mv 2021-06-11T15:02:52Z
dc.format.none.fl_str_mv application/pdf
dc.identifier.none.fl_str_mv http://hdl.handle.net/10451/48465
dc.language.none.fl_str_mv eng
dc.rights.none.fl_str_mv http://purl.org/coar/access_right/c_abf2
dc.subject.none.fl_str_mv Fibrilhação Auricular
Enfarte do miocárdio
Ressonância Magnética Cardíaca
Deep learning
Classificação/ Segmentação
Teses de mestrado - 2021
dc.title.fl_str_mv Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
dc.type.none.fl_str_mv http://purl.org/coar/resource_type/c_bdcc
description Atrial fibrillation (AF), is the most frequent sustained cardiac arrhythmia, described by an irregular and rapid contraction of the two upper chambers of the heart (the atria). AF development is promoted and predisposed by atrial dilation, which is a consequence of atria adaptation to AF. However, it is not clear whether atrial dilation appears similarly over the cardiac cycle and how it affects ventricular volumes. Catheter ablation is arguably the AF gold standard treatment. In their current form, ablations are capable of directly terminating AF in selected patients but are only first-time effective in approximately 50% of the cases. In the first part of this work, volumetric functional markers of the left atrium (LA) and left ventricle (LV) of AF patients were studied. More precisely, a customised convolutional neural network (CNN) was proposed to segment, across the cardiac cycle, the LA from short axis CINE MRI images acquired with full cardiac coverage in AF patients. Using the proposed automatic LA segmentation, volumetric time curves were plotted and ejection fractions (EF) were automatically calculated for both chambers. The second part of the project was dedicated to developing classification models based on cardiac MR images. The EMIDEC STACOM 2020 challenge was used as an initial project and basis to create binary classifiers based on fully automatic classification neural networks (NNs), since it presented a relatively simple binary classification task (presence/absence of disease) and a large dataset. For the challenge, a deep learning NN was proposed to automatically classify myocardial disease from delayed enhancement cardiac MR (DE-CMR) and patient clinical information. The highest classification accuracy (100%) was achieved with Clinic-NET+, a NN that used information from images, segmentations and clinical annotations. For the final goal of this project, the previously referred NNs were re-trained to predict AF recurrence after catheter ablation (CA) in AF patients using pre-ablation LA short axis in CINE MRI images. In this task, the best overall performance was achieved by Clinic-NET+ with a test accuracy of 88%. This work shown the potential of NNs to interpret and extract clinical information from cardiac MRI. If more data is available, in the future, these methods can potentially be used to help and guide clinical AF prognosis and diagnosis.
dirty 0
eu_rights_str_mv openAccess
format masterThesis
fulltext.url.fl_str_mv https://repositorio.ulisboa.pt/bitstreams/b9009fa0-5a90-4493-b6ea-1c8c8be7cab6/download
id ul_c170f090910fca8293e1b4c5de97bc21
identifier.url.fl_str_mv http://hdl.handle.net/10451/48465
instacron_str ul
institution Universidade de Lisboa
instname_str Universidade de Lisboa
language eng
network_acronym_str ul
network_name_str Repositório da Universidade de Lisboa
oai_identifier_str oai:repositorio.ulisboa.pt:10451/48465
organization_str_mv urn:organizationAcronym:ul
person_str_mv Lourenço, Ana Catarina Feliciano
publishDate 2021
reponame_str Repositório da Universidade de Lisboa
repository_id_str urn:repositoryAcronym:ul
service_str_mv urn:repositoryAcronym:ul
spelling engpt_PTAtrial fibrillation (AF), is the most frequent sustained cardiac arrhythmia, described by an irregular and rapid contraction of the two upper chambers of the heart (the atria). AF development is promoted and predisposed by atrial dilation, which is a consequence of atria adaptation to AF. However, it is not clear whether atrial dilation appears similarly over the cardiac cycle and how it affects ventricular volumes. Catheter ablation is arguably the AF gold standard treatment. In their current form, ablations are capable of directly terminating AF in selected patients but are only first-time effective in approximately 50% of the cases. In the first part of this work, volumetric functional markers of the left atrium (LA) and left ventricle (LV) of AF patients were studied. More precisely, a customised convolutional neural network (CNN) was proposed to segment, across the cardiac cycle, the LA from short axis CINE MRI images acquired with full cardiac coverage in AF patients. Using the proposed automatic LA segmentation, volumetric time curves were plotted and ejection fractions (EF) were automatically calculated for both chambers. The second part of the project was dedicated to developing classification models based on cardiac MR images. The EMIDEC STACOM 2020 challenge was used as an initial project and basis to create binary classifiers based on fully automatic classification neural networks (NNs), since it presented a relatively simple binary classification task (presence/absence of disease) and a large dataset. For the challenge, a deep learning NN was proposed to automatically classify myocardial disease from delayed enhancement cardiac MR (DE-CMR) and patient clinical information. The highest classification accuracy (100%) was achieved with Clinic-NET+, a NN that used information from images, segmentations and clinical annotations. For the final goal of this project, the previously referred NNs were re-trained to predict AF recurrence after catheter ablation (CA) in AF patients using pre-ablation LA short axis in CINE MRI images. In this task, the best overall performance was achieved by Clinic-NET+ with a test accuracy of 88%. This work shown the potential of NNs to interpret and extract clinical information from cardiac MRI. If more data is available, in the future, these methods can potentially be used to help and guide clinical AF prognosis and diagnosis.application/pdfpt_PTDeep learning tools for outcome prediction in a trial fibrilation from cardiac MRILourenço, Ana Catarina FelicianoCaetano, Gina Maria CostaVarela, MartaHostingInstitutionOrganizationalRepositório Científico de Acesso Aberto da ULisboae-mailmailto:repositorio@reitoria.ulisboa.ptrepositorio@reitoria.ulisboa.ptURNurn:tid:2029339622021-06-11T15:02:52Z202120212021-01-01T00:00:00ZHandlehttp://hdl.handle.net/10451/48465http://purl.org/coar/access_right/c_abf2open accessFibrilhação AuricularEnfarte do miocárdioRessonância Magnética CardíacaDeep learningClassificação/ SegmentaçãoTeses de mestrado - 20218025894 bytesliteraturehttp://purl.org/coar/resource_type/c_bdccmaster thesishttp://purl.org/coar/access_right/c_abf2application/pdffulltexthttps://repositorio.ulisboa.pt/bitstreams/b9009fa0-5a90-4493-b6ea-1c8c8be7cab6/download
spellingShingle Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
Lourenço, Ana Catarina Feliciano
Fibrilhação Auricular
Enfarte do miocárdio
Ressonância Magnética Cardíaca
Deep learning
Classificação/ Segmentação
Teses de mestrado - 2021
Lourenço, Ana Catarina Feliciano
Fibrilhação Auricular
Enfarte do miocárdio
Ressonância Magnética Cardíaca
Deep learning
Classificação/ Segmentação
Teses de mestrado - 2021
status SINGLETON
subject.fl_str_mv Fibrilhação Auricular
Enfarte do miocárdio
Ressonância Magnética Cardíaca
Deep learning
Classificação/ Segmentação
Teses de mestrado - 2021
title Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
title_full Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
title_fullStr Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
title_full_unstemmed Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
title_short Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
title_sort Deep learning tools for outcome prediction in a trial fibrilation from cardiac MRI
topic Fibrilhação Auricular
Enfarte do miocárdio
Ressonância Magnética Cardíaca
Deep learning
Classificação/ Segmentação
Teses de mestrado - 2021
topic_facet Fibrilhação Auricular
Enfarte do miocárdio
Ressonância Magnética Cardíaca
Deep learning
Classificação/ Segmentação
Teses de mestrado - 2021
url http://hdl.handle.net/10451/48465
visible 1