Publicação
Deep Learning in COVID-19 lung ultrasound analysis
| Resumo: | One of the determining factors for the COVID-19 course of treatment is disease severity. Medical imaging, particularly CT scans and X-rays, is frequently used in this assessment. However, these methods have significant downsides. Lung ultrasound has recently gained recognition as an alternative to CT scans and X-rays due to the advantages presented by the exam. Lung ultrasounds are cost-effective, portable, fast, non-invasive, easily repeat- able, and do not expose patients to radiation. These aspects are particularly relevant in the context of low-resource health care providers. This study proposes a framework for automated severity analysis of COVID-19 lung ultrasounds, intended to detect frames presenting alterations indicative of the disease and provide its subsequent severity classification. The system could be used to ease the learning process of health care providers and enable them to obtain expert-level COVID- 19 severity assessments even in low-resource situations. The pipeline is comprised of two main modules, the frame selection phase, and a frame classification phase. Due to dataset limitations, the demonstration of the entire se- quence of the pipeline is unachievable. Thus, both phases of the project were approached separately. The frame selection phase consisted of unsupervised learning methods and thus couldn’t be evaluated in this study’s context. Instead, the study offers a demonstration of this step’s procedures. As for the frame classification phase, the study optimized four CNN architectures, one for each anomaly considered for severity classification. All four models were able to accurately determine severity. Based on these findings, the study concluded that by having pre-selected informative frames, it’s possible to train a classifier to reproduce the doctor-attributed scores, which suggests the feasibility of the proposed pipeline. |
|---|---|
| Autores principais: | Cerqueira, Marta de Oliveira |
| Assunto: | COVID-19 SARS-CoV-2 Lung ultrasound Dimensionality Reduction Clustering Convolutional Neural Network (CNN) |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| _version_ | 1868414974795186176 |
|---|---|
| author | Cerqueira, Marta de Oliveira |
| author_facet | Cerqueira, Marta de Oliveira |
| author_role | author |
| contributor_name_str_mv | Krippahl, Ludwig Bispo, Regina RUN |
| country_str | PT |
| creators_json_txt | [{\"Person.name\":\"Cerqueira, Marta de Oliveira\"}] |
| datacite.contributors.contributor.contributorName.fl_str_mv | Krippahl, Ludwig Bispo, Regina RUN |
| datacite.creators.creator.creatorName.fl_str_mv | Cerqueira, Marta de Oliveira |
| datacite.date.Accepted.fl_str_mv | 2022-12-01T00:00:00Z |
| datacite.date.available.fl_str_mv | 2024-05-14T14:11:04Z |
| datacite.date.embargoed.fl_str_mv | 2024-05-14T14:11:04Z |
| datacite.rights.fl_str_mv | http://purl.org/coar/access_right/c_abf2 |
| datacite.subjects.subject.fl_str_mv | COVID-19 SARS-CoV-2 Lung ultrasound Dimensionality Reduction Clustering Convolutional Neural Network (CNN) |
| datacite.titles.title.fl_str_mv | Deep Learning in COVID-19 lung ultrasound analysis |
| dc.contributor.none.fl_str_mv | Krippahl, Ludwig Bispo, Regina RUN |
| dc.creator.none.fl_str_mv | Cerqueira, Marta de Oliveira |
| dc.date.Accepted.fl_str_mv | 2022-12-01T00:00:00Z |
| dc.date.available.fl_str_mv | 2024-05-14T14:11:04Z |
| dc.date.embargoed.fl_str_mv | 2024-05-14T14:11:04Z |
| dc.format.none.fl_str_mv | application/pdf |
| dc.identifier.none.fl_str_mv | http://hdl.handle.net/10362/167374 |
| 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 | COVID-19 SARS-CoV-2 Lung ultrasound Dimensionality Reduction Clustering Convolutional Neural Network (CNN) |
| dc.title.fl_str_mv | Deep Learning in COVID-19 lung ultrasound analysis |
| dc.type.none.fl_str_mv | http://purl.org/coar/resource_type/c_bdcc |
| description | One of the determining factors for the COVID-19 course of treatment is disease severity. Medical imaging, particularly CT scans and X-rays, is frequently used in this assessment. However, these methods have significant downsides. Lung ultrasound has recently gained recognition as an alternative to CT scans and X-rays due to the advantages presented by the exam. Lung ultrasounds are cost-effective, portable, fast, non-invasive, easily repeat- able, and do not expose patients to radiation. These aspects are particularly relevant in the context of low-resource health care providers. This study proposes a framework for automated severity analysis of COVID-19 lung ultrasounds, intended to detect frames presenting alterations indicative of the disease and provide its subsequent severity classification. The system could be used to ease the learning process of health care providers and enable them to obtain expert-level COVID- 19 severity assessments even in low-resource situations. The pipeline is comprised of two main modules, the frame selection phase, and a frame classification phase. Due to dataset limitations, the demonstration of the entire se- quence of the pipeline is unachievable. Thus, both phases of the project were approached separately. The frame selection phase consisted of unsupervised learning methods and thus couldn’t be evaluated in this study’s context. Instead, the study offers a demonstration of this step’s procedures. As for the frame classification phase, the study optimized four CNN architectures, one for each anomaly considered for severity classification. All four models were able to accurately determine severity. Based on these findings, the study concluded that by having pre-selected informative frames, it’s possible to train a classifier to reproduce the doctor-attributed scores, which suggests the feasibility of the proposed pipeline. |
| dirty | 0 |
| eu_rights_str_mv | openAccess |
| format | masterThesis |
| fulltext.url.fl_str_mv | https://run.unl.pt/bitstreams/26947897-eb1c-47ed-8714-8772979c6c97/download |
| id | run_54c03dfd4d6c54ac5eb3a8bc910e6b53 |
| identifier.url.fl_str_mv | http://hdl.handle.net/10362/167374 |
| instacron_str | unl |
| institution | Universidade Nova de Lisboa |
| instname_str | Universidade Nova de Lisboa |
| language | eng |
| network_acronym_str | run |
| network_name_str | Repositório Institucional da UNL |
| oai_identifier_str | oai:run.unl.pt:10362/167374 |
| organization_str_mv | urn:organizationAcronym:unl |
| person_str_mv | Cerqueira, Marta de Oliveira |
| publishDate | 2022 |
| reponame_str | Repositório Institucional da UNL |
| repository_id_str | urn:repositoryAcronym:run |
| service_str_mv | urn:repositoryAcronym:run |
| spelling | engpt_PTOne of the determining factors for the COVID-19 course of treatment is disease severity. Medical imaging, particularly CT scans and X-rays, is frequently used in this assessment. However, these methods have significant downsides. Lung ultrasound has recently gained recognition as an alternative to CT scans and X-rays due to the advantages presented by the exam. Lung ultrasounds are cost-effective, portable, fast, non-invasive, easily repeat- able, and do not expose patients to radiation. These aspects are particularly relevant in the context of low-resource health care providers. This study proposes a framework for automated severity analysis of COVID-19 lung ultrasounds, intended to detect frames presenting alterations indicative of the disease and provide its subsequent severity classification. The system could be used to ease the learning process of health care providers and enable them to obtain expert-level COVID- 19 severity assessments even in low-resource situations. The pipeline is comprised of two main modules, the frame selection phase, and a frame classification phase. Due to dataset limitations, the demonstration of the entire se- quence of the pipeline is unachievable. Thus, both phases of the project were approached separately. The frame selection phase consisted of unsupervised learning methods and thus couldn’t be evaluated in this study’s context. Instead, the study offers a demonstration of this step’s procedures. As for the frame classification phase, the study optimized four CNN architectures, one for each anomaly considered for severity classification. All four models were able to accurately determine severity. Based on these findings, the study concluded that by having pre-selected informative frames, it’s possible to train a classifier to reproduce the doctor-attributed scores, which suggests the feasibility of the proposed pipeline.application/pdfpt_PTDeep Learning in COVID-19 lung ultrasound analysisCerqueira, Marta de OliveiraKrippahl, LudwigBispo, ReginaHostingInstitutionOrganizationalRUNe-mailmailto:run@unl.ptrun@unl.pt2024-05-14T14:11:04Z2022-122022-12-01T00:00:00ZHandlehttp://hdl.handle.net/10362/167374http://purl.org/coar/access_right/c_abf2open accessCOVID-19SARS-CoV-2Lung ultrasoundDimensionality ReductionClusteringConvolutional Neural Network (CNN)7350545 bytesliteraturehttp://purl.org/coar/resource_type/c_bdccmaster thesishttp://purl.org/coar/access_right/c_abf2application/pdffulltexthttps://run.unl.pt/bitstreams/26947897-eb1c-47ed-8714-8772979c6c97/download |
| spellingShingle | Deep Learning in COVID-19 lung ultrasound analysis Cerqueira, Marta de Oliveira COVID-19 SARS-CoV-2 Lung ultrasound Dimensionality Reduction Clustering Convolutional Neural Network (CNN) |
| status | SINGLETON |
| subject.fl_str_mv | COVID-19 SARS-CoV-2 Lung ultrasound Dimensionality Reduction Clustering Convolutional Neural Network (CNN) |
| title | Deep Learning in COVID-19 lung ultrasound analysis |
| title_full | Deep Learning in COVID-19 lung ultrasound analysis |
| title_fullStr | Deep Learning in COVID-19 lung ultrasound analysis |
| title_full_unstemmed | Deep Learning in COVID-19 lung ultrasound analysis |
| title_short | Deep Learning in COVID-19 lung ultrasound analysis |
| title_sort | Deep Learning in COVID-19 lung ultrasound analysis |
| topic | COVID-19 SARS-CoV-2 Lung ultrasound Dimensionality Reduction Clustering Convolutional Neural Network (CNN) |
| topic_facet | COVID-19 SARS-CoV-2 Lung ultrasound Dimensionality Reduction Clustering Convolutional Neural Network (CNN) |
| url | http://hdl.handle.net/10362/167374 |
| visible | 1 |