{"id":{"repo_id":"cambridge","oai_identifier":"oai:www.repository.cam.ac.uk:1810/391708"},"canonical_url":"https://search.dev.ndltd.org/etd/cambridge/oai:www.repository.cam.ac.uk:1810/391708","repository":{"repo_id":"cambridge","name":"Cambridge University","base_url":"https://api.repository.cam.ac.uk/server/oai/request"},"display":{"title":"Investigating the Use of Artificial Intelligence in Chest Radiography: Efficacy, Bias, and Lessons from the COVID-19 Pandemic","abstract":"The Coronavirus Disease 2019 (COVID-19) pandemic has emerged as one of the most significant global health crises in history. By December 2024, COVID-19 had resulted in more than 776 million cases and 7 million deaths worldwide. Early detection, accurate diagnosis, and timely management of COVID-19 were critical in the early stages of the pandemic, especially given the initial testing shortages. Chest radiography played a key role in diagnosis and clinical management, and radiological findings were found to predict patient outcomes. The calls for artificial intelligence (AI) solutions to help clinicians resulted in unprecedented access to medical datasets. Large, novel public datasets were created, such as the National COVID-19 Chest Imaging Database (NCCID) in the UK and the Medical Imaging and Data Resource Center (MIDRC) in the US. Thousands of AI models have been developed to improve diagnostic accuracy and predict outcomes. However, these efforts highlighted challenges related to data quality, generalisability, and bias, with few models being adopted in clinical practice due to these issues. One concern is shortcut learning, where AI models rely on spuriously correlated features with the outcome rather than the underlying. Technical factors such as projection and positioning of the patient on chest radiography can introduce such biases. This thesis takes advantage of unprecedented data access and the unique circumstances of the pandemic to deepen our understanding of the importance of data in radiological AI. It explores how adopting a more data-centric approach can enhance reliability and effectiveness. Chapter 2 begins with a comprehensive background on COVID-19, radiography, neural networks, and data curation methods. Chapter 3 then delves into the importance of data standards and the critical role of expert knowledge in preparing clinical datasets for use in imaging. Using the NCCID as a case study, we explore why rigorous data curation is vital to ensuring the accuracy and generalisability of AI models. Chapter 4 applies the curated data to explore how AI can be used to study uncommon conditions, focusing specifically on COVID-19-related pneumothorax. The focus then shifts in Chapter 5 to developing automated quality control tools to help developers and radiologists efficiently manage and interpret their data. Using the extensive COVID-19 datasets, these tools aim to streamline the data review process and improve the overall quality of AI input. Chapter 6 evaluates the impact of these tools on the performance of an AI model, highlighting the crucial role of quality control in mitigating shortcut learning and improving model robustness and generalisability. Finally, Chapter 7 summarises the main findings of this thesis and outlines directions for future research.","abstract_html":"The Coronavirus Disease 2019 (COVID-19) pandemic has emerged as one of the most significant global health crises in history. By December 2024, COVID-19 had resulted in more than 776 million cases and 7 million deaths worldwide. Early detection, accurate diagnosis, and timely management of COVID-19 were critical in the early stages of the pandemic, especially given the initial testing shortages. Chest radiography played a key role in diagnosis and clinical management, and radiological findings were found to predict patient outcomes. The calls for artificial intelligence (AI) solutions to help clinicians resulted in unprecedented access to medical datasets. Large, novel public datasets were created, such as the National COVID-19 Chest Imaging Database (NCCID) in the UK and the Medical Imaging and Data Resource Center (MIDRC) in the US. Thousands of AI models have been developed to improve diagnostic accuracy and predict outcomes. However, these efforts highlighted challenges related to data quality, generalisability, and bias, with few models being adopted in clinical practice due to these issues. One concern is shortcut learning, where AI models rely on spuriously correlated features with the outcome rather than the underlying. Technical factors such as projection and positioning of the patient on chest radiography can introduce such biases. This thesis takes advantage of unprecedented data access and the unique circumstances of the pandemic to deepen our understanding of the importance of data in radiological AI. It explores how adopting a more data-centric approach can enhance reliability and effectiveness. Chapter 2 begins with a comprehensive background on COVID-19, radiography, neural networks, and data curation methods. Chapter 3 then delves into the importance of data standards and the critical role of expert knowledge in preparing clinical datasets for use in imaging. Using the NCCID as a case study, we explore why rigorous data curation is vital to ensuring the accuracy and generalisability of AI models. Chapter 4 applies the curated data to explore how AI can be used to study uncommon conditions, focusing specifically on COVID-19-related pneumothorax. The focus then shifts in Chapter 5 to developing automated quality control tools to help developers and radiologists efficiently manage and interpret their data. Using the extensive COVID-19 datasets, these tools aim to streamline the data review process and improve the overall quality of AI input. Chapter 6 evaluates the impact of these tools on the performance of an AI model, highlighting the crucial role of quality control in mitigating shortcut learning and improving model robustness and generalisability. Finally, Chapter 7 summarises the main findings of this thesis and outlines directions for future research.","abstract_has_math":false,"creators":["Selby, Ian Andrew"],"institution":"University of Cambridge","degree_name":"Doctor of Philosophy (PhD)","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Gilbert, Fiona"],"committee_chairs":[],"committee_members":[],"year":2024,"date_issued":"2024-12-13","date_published":"2024-12-13","updated_at":"2026-07-22T22:24:21Z","subjects":["Chest Radiography","Artificial Intelligence","Convolutional Neural Networks","COVID-19","Shortcut Learning"],"languages":["eng"],"rights":[],"rights_urls":["https://www.repository.cam.ac.uk/bitstreams/5ccaedbf-e4f5-472d-ba90-e32a6549e286/download","http://purl.org/NET/rdflicense/allrightsreserved"],"identifier_entries":[{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["0000000342448893"],"render_values":[{"text":"0000-0003-4244-8893","href":"https://orcid.org/0000-0003-4244-8893","code":true}]}]},"links":{"outbound_url":"https://doi.org/10.17863/CAM.122741","outbound_label":"DOI","outbound_source":"dc:identifier.doi"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Gilbert, Fiona"]},{"key":"dc:creator","label":"Author","values":["Selby, Ian Andrew"]},{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["0000000342448893"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2024-12-13"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Cambridge"]},{"key":"dc:relation.isreferencedby.uri","label":"Dc Relation Isreferencedby URI","values":["https://www.repository.cam.ac.uk/handle/1810/391708"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["Doctoral"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Chest Radiography","Artificial Intelligence","Convolutional Neural Networks","COVID-19","Shortcut Learning"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["https://www.repository.cam.ac.uk/bitstreams/5ccaedbf-e4f5-472d-ba90-e32a6549e286/download","http://purl.org/NET/rdflicense/allrightsreserved"]},{"key":"dc:rights.embargodate","label":"Dc Rights Embargodate","values":["2026-10-31"]},{"key":"dc:rights.embargotype","label":"Dc Rights Embargotype","values":["embargo"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.17863/CAM.122741"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://www.repository.cam.ac.uk/bitstreams/da688e22-f305-47d2-bdbd-b596eb0652e2/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["The Coronavirus Disease 2019 (COVID-19) pandemic has emerged as one of the most significant global health crises in history. By December 2024, COVID-19 had resulted in more than 776 million cases and 7 million deaths worldwide. Early detection, accurate diagnosis, and timely management of COVID-19 were critical in the early stages of the pandemic, especially given the initial testing shortages. Chest radiography played a key role in diagnosis and clinical management, and radiological findings were found to predict patient outcomes. The calls for artificial intelligence (AI) solutions to help clinicians resulted in unprecedented access to medical datasets. Large, novel public datasets were created, such as the National COVID-19 Chest Imaging Database (NCCID) in the UK and the Medical Imaging and Data Resource Center (MIDRC) in the US. Thousands of AI models have been developed to improve diagnostic accuracy and predict outcomes. However, these efforts highlighted challenges related to data quality, generalisability, and bias, with few models being adopted in clinical practice due to these issues. One concern is shortcut learning, where AI models rely on spuriously correlated features with the outcome rather than the underlying. Technical factors such as projection and positioning of the patient on chest radiography can introduce such biases. This thesis takes advantage of unprecedented data access and the unique circumstances of the pandemic to deepen our understanding of the importance of data in radiological AI. It explores how adopting a more data-centric approach can enhance reliability and effectiveness. Chapter 2 begins with a comprehensive background on COVID-19, radiography, neural networks, and data curation methods. Chapter 3 then delves into the importance of data standards and the critical role of expert knowledge in preparing clinical datasets for use in imaging. Using the NCCID as a case study, we explore why rigorous data curation is vital to ensuring the accuracy and generalisability of AI models. Chapter 4 applies the curated data to explore how AI can be used to study uncommon conditions, focusing specifically on COVID-19-related pneumothorax. The focus then shifts in Chapter 5 to developing automated quality control tools to help developers and radiologists efficiently manage and interpret their data. Using the extensive COVID-19 datasets, these tools aim to streamline the data review process and improve the overall quality of AI input. Chapter 6 evaluates the impact of these tools on the performance of an AI model, highlighting the crucial role of quality control in mitigating shortcut learning and improving model robustness and generalisability. Finally, Chapter 7 summarises the main findings of this thesis and outlines directions for future research."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["4cd1486cd8b4771e61529195a67bcee2","87eda9de84448d1f82354d60eee3eb5f"]},{"key":"dc:title","label":"Title","values":["Investigating the Use of Artificial Intelligence in Chest Radiography: Efficacy, Bias, and Lessons from the COVID-19 Pandemic"]}]}],"canonical_facts":{"dc:contributor.advisor":["Gilbert, Fiona"],"dc:creator":["Selby, Ian Andrew"],"dc:creator.authoridentifier":["0000000342448893"],"dc:date.issued":["2024-12-13"],"dc:description.abstract":["The Coronavirus Disease 2019 (COVID-19) pandemic has emerged as one of the most significant global health crises in history. By December 2024, COVID-19 had resulted in more than 776 million cases and 7 million deaths worldwide. Early detection, accurate diagnosis, and timely management of COVID-19 were critical in the early stages of the pandemic, especially given the initial testing shortages. Chest radiography played a key role in diagnosis and clinical management, and radiological findings were found to predict patient outcomes. The calls for artificial intelligence (AI) solutions to help clinicians resulted in unprecedented access to medical datasets. Large, novel public datasets were created, such as the National COVID-19 Chest Imaging Database (NCCID) in the UK and the Medical Imaging and Data Resource Center (MIDRC) in the US. Thousands of AI models have been developed to improve diagnostic accuracy and predict outcomes. However, these efforts highlighted challenges related to data quality, generalisability, and bias, with few models being adopted in clinical practice due to these issues. One concern is shortcut learning, where AI models rely on spuriously correlated features with the outcome rather than the underlying. Technical factors such as projection and positioning of the patient on chest radiography can introduce such biases. This thesis takes advantage of unprecedented data access and the unique circumstances of the pandemic to deepen our understanding of the importance of data in radiological AI. It explores how adopting a more data-centric approach can enhance reliability and effectiveness. Chapter 2 begins with a comprehensive background on COVID-19, radiography, neural networks, and data curation methods. Chapter 3 then delves into the importance of data standards and the critical role of expert knowledge in preparing clinical datasets for use in imaging. Using the NCCID as a case study, we explore why rigorous data curation is vital to ensuring the accuracy and generalisability of AI models. Chapter 4 applies the curated data to explore how AI can be used to study uncommon conditions, focusing specifically on COVID-19-related pneumothorax. The focus then shifts in Chapter 5 to developing automated quality control tools to help developers and radiologists efficiently manage and interpret their data. Using the extensive COVID-19 datasets, these tools aim to streamline the data review process and improve the overall quality of AI input. Chapter 6 evaluates the impact of these tools on the performance of an AI model, highlighting the crucial role of quality control in mitigating shortcut learning and improving model robustness and generalisability. Finally, Chapter 7 summarises the main findings of this thesis and outlines directions for future research."],"dc:format.checksum.md5":["4cd1486cd8b4771e61529195a67bcee2","87eda9de84448d1f82354d60eee3eb5f"],"dc:identifier.doi":["https://doi.org/10.17863/CAM.122741"],"dc:identifier.uri":["https://www.repository.cam.ac.uk/bitstreams/da688e22-f305-47d2-bdbd-b596eb0652e2/download"],"dc:language":["eng"],"dc:publisher.institution":["University of Cambridge"],"dc:relation.isreferencedby.uri":["https://www.repository.cam.ac.uk/handle/1810/391708"],"dc:rights":["https://www.repository.cam.ac.uk/bitstreams/5ccaedbf-e4f5-472d-ba90-e32a6549e286/download","http://purl.org/NET/rdflicense/allrightsreserved"],"dc:rights.embargodate":["2026-10-31"],"dc:rights.embargotype":["embargo"],"dc:subject":["Chest Radiography","Artificial Intelligence","Convolutional Neural Networks","COVID-19","Shortcut Learning"],"dc:title":["Investigating the Use of Artificial Intelligence in Chest Radiography: Efficacy, Bias, and Lessons from the COVID-19 Pandemic"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["Doctoral"],"dc:type.qualificationname":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-07-22T22:24:21Z"}