Classifying real-world macroscopic images in the primary-secondary care interface using transfer learning: implications for development of artificial intelligence solutions using nondermoscopic images

The application of deep learning (DL) to diagnostic dermatology has been the subject of numerous studies, with some reporting skin lesion classification performance on curated datasets comparable to that of experienced dermatologists. Most skin disease images encountered in clinical settings are mac...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Clinical and experimental dermatology 2024-06, Vol.49 (7), p.699-706
Hauptverfasser: Carse, Jacob, Süveges, Tamás, Chin, Gillian, Muthiah, Shareen, Morton, Colin, Proby, Charlotte, Trucco, Emanuele, Fleming, Colin, McKenna, Stephen
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page 706
container_issue 7
container_start_page 699
container_title Clinical and experimental dermatology
container_volume 49
creator Carse, Jacob
Süveges, Tamás
Chin, Gillian
Muthiah, Shareen
Morton, Colin
Proby, Charlotte
Trucco, Emanuele
Fleming, Colin
McKenna, Stephen
description The application of deep learning (DL) to diagnostic dermatology has been the subject of numerous studies, with some reporting skin lesion classification performance on curated datasets comparable to that of experienced dermatologists. Most skin disease images encountered in clinical settings are macroscopic, without dermoscopic information, and exhibit considerable variability. Further research is necessary to determine the generalizability of DL algorithms across populations and acquisition settings. To assess the extent to which DL can generalize to nondermoscopic datasets acquired at the primary-secondary care interface in the National Health Service (NHS); to explore how to obtain a clinically satisfactory performance on nonstandardized, real-world local data without the availability of large diagnostically labelled local datasets; and to measure the impact of pretraining DL algorithms on external, public datasets. Diagnostic macroscopic image datasets were created from previous referrals from primary to secondary care. These included 2213 images referred from primary care practitioners in NHS Tayside and 1510 images from NHS Forth Valley acquired by medical photographers. Two further datasets with identical diagnostic labels were obtained from sources in the public domain, namely the International Skin Imaging Collaboration (ISIC) dermoscopic dataset and the SD-260 nondermoscopic dataset. DL algorithms, specifically EfficientNets and Self-attention with Window-wise Inner-product based Network (SWIN) transformers, were trained using data from each of these datasets. Algorithms were also fine-tuned on images from the NHS datasets after pretraining on different data combinations, including the larger public-domain datasets. Receiver operating characteristic curves and the area under such curves (AUC) were used to assess performance. SWIN transformers tested on Forth Valley data had AUCs of 0.85 and 0.89 when trained on SD-260 and Forth Valley data, respectively. Training on SD-260 followed by fine tuning of Forth Valley data gave an AUC of 0.91. Similar effects of pretraining and tuning on local data were observed using Tayside data and EfficientNets. Pretraining on the larger dermoscopic image dataset (ISIC 2019) provided no additional benefit. Pretraining on public macroscopic images followed by tuning to local data gave promising results. Further improvements are needed to afford deployment in real clinical pathways. Larger datasets local to the targe
doi_str_mv 10.1093/ced/llad400
format Article
fullrecord <record><control><sourceid>proquest_cross</sourceid><recordid>TN_cdi_proquest_miscellaneous_2892659199</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2892659199</sourcerecordid><originalsourceid>FETCH-LOGICAL-c256t-cbf6eebc6b0a644687797f5858bc527c19cbcbc6892058454bc92a536b57bf863</originalsourceid><addsrcrecordid>eNpVkcFu3CAQhlHVqNmmPfUecaxUucHGYJNbtWqaSJF6Sc4WjIctFQYH7ER5wzxW2ezmUHEYmPnnYzQ_IV9q9r1mil8Ajhfe67Fl7B3Z1FyKqmk4e082jLOukor3p-Rjzn8Zq3ndiQ_klHdKMdXyDXnZep2zs88u7GhC7aunmPxIJw0pZoizA-omvcNMXaDLH6RzKu_0XGWEGMZyo6ATluqCyWpAuuY9a0k6ZIuJetQplMxl4czegV5cDJnamOiIj-jjPGFYaLRUp8VZB077V5r3boehAHP066HpgA7lX0zT_-N9IidW-4yfj_GM3F_9vNteV7e_f91sf9xW0Ai5VGCsRDQgDdOybWXfdaqzohe9AdF0UCsw5cheNUz0rWgNqEYLLo3ojO0lPyNfD9w5xYcV8zJMLkMZVgeMax6a0imFqpUq0m8H6X6VOaEdjrsbajbsrRuKdcPRuqI-P4JXM5X8m_bNK_4PrACdSg</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2892659199</pqid></control><display><type>article</type><title>Classifying real-world macroscopic images in the primary-secondary care interface using transfer learning: implications for development of artificial intelligence solutions using nondermoscopic images</title><source>Oxford University Press Journals</source><source>MEDLINE</source><creator>Carse, Jacob ; Süveges, Tamás ; Chin, Gillian ; Muthiah, Shareen ; Morton, Colin ; Proby, Charlotte ; Trucco, Emanuele ; Fleming, Colin ; McKenna, Stephen</creator><creatorcontrib>Carse, Jacob ; Süveges, Tamás ; Chin, Gillian ; Muthiah, Shareen ; Morton, Colin ; Proby, Charlotte ; Trucco, Emanuele ; Fleming, Colin ; McKenna, Stephen</creatorcontrib><description>The application of deep learning (DL) to diagnostic dermatology has been the subject of numerous studies, with some reporting skin lesion classification performance on curated datasets comparable to that of experienced dermatologists. Most skin disease images encountered in clinical settings are macroscopic, without dermoscopic information, and exhibit considerable variability. Further research is necessary to determine the generalizability of DL algorithms across populations and acquisition settings. To assess the extent to which DL can generalize to nondermoscopic datasets acquired at the primary-secondary care interface in the National Health Service (NHS); to explore how to obtain a clinically satisfactory performance on nonstandardized, real-world local data without the availability of large diagnostically labelled local datasets; and to measure the impact of pretraining DL algorithms on external, public datasets. Diagnostic macroscopic image datasets were created from previous referrals from primary to secondary care. These included 2213 images referred from primary care practitioners in NHS Tayside and 1510 images from NHS Forth Valley acquired by medical photographers. Two further datasets with identical diagnostic labels were obtained from sources in the public domain, namely the International Skin Imaging Collaboration (ISIC) dermoscopic dataset and the SD-260 nondermoscopic dataset. DL algorithms, specifically EfficientNets and Self-attention with Window-wise Inner-product based Network (SWIN) transformers, were trained using data from each of these datasets. Algorithms were also fine-tuned on images from the NHS datasets after pretraining on different data combinations, including the larger public-domain datasets. Receiver operating characteristic curves and the area under such curves (AUC) were used to assess performance. SWIN transformers tested on Forth Valley data had AUCs of 0.85 and 0.89 when trained on SD-260 and Forth Valley data, respectively. Training on SD-260 followed by fine tuning of Forth Valley data gave an AUC of 0.91. Similar effects of pretraining and tuning on local data were observed using Tayside data and EfficientNets. Pretraining on the larger dermoscopic image dataset (ISIC 2019) provided no additional benefit. Pretraining on public macroscopic images followed by tuning to local data gave promising results. Further improvements are needed to afford deployment in real clinical pathways. Larger datasets local to the target domain might be expected to yield further improved performance.</description><identifier>ISSN: 0307-6938</identifier><identifier>ISSN: 1365-2230</identifier><identifier>EISSN: 1365-2230</identifier><identifier>DOI: 10.1093/ced/llad400</identifier><identifier>PMID: 37990943</identifier><language>eng</language><publisher>England</publisher><subject>Algorithms ; Artificial Intelligence ; Deep Learning ; Dermatology ; Dermoscopy - methods ; Humans ; Primary Health Care ; Secondary Care ; Skin Diseases - diagnostic imaging ; Skin Diseases - pathology ; State Medicine</subject><ispartof>Clinical and experimental dermatology, 2024-06, Vol.49 (7), p.699-706</ispartof><rights>The Author(s) 2023. Published by Oxford University Press on behalf of British Association of Dermatologists.</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c256t-cbf6eebc6b0a644687797f5858bc527c19cbcbc6892058454bc92a536b57bf863</citedby><cites>FETCH-LOGICAL-c256t-cbf6eebc6b0a644687797f5858bc527c19cbcbc6892058454bc92a536b57bf863</cites><orcidid>0000-0002-3292-4836 ; 0000-0003-0530-2035</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>314,780,784,27924,27925</link.rule.ids><backlink>$$Uhttps://www.ncbi.nlm.nih.gov/pubmed/37990943$$D View this record in MEDLINE/PubMed$$Hfree_for_read</backlink></links><search><creatorcontrib>Carse, Jacob</creatorcontrib><creatorcontrib>Süveges, Tamás</creatorcontrib><creatorcontrib>Chin, Gillian</creatorcontrib><creatorcontrib>Muthiah, Shareen</creatorcontrib><creatorcontrib>Morton, Colin</creatorcontrib><creatorcontrib>Proby, Charlotte</creatorcontrib><creatorcontrib>Trucco, Emanuele</creatorcontrib><creatorcontrib>Fleming, Colin</creatorcontrib><creatorcontrib>McKenna, Stephen</creatorcontrib><title>Classifying real-world macroscopic images in the primary-secondary care interface using transfer learning: implications for development of artificial intelligence solutions using nondermoscopic images</title><title>Clinical and experimental dermatology</title><addtitle>Clin Exp Dermatol</addtitle><description>The application of deep learning (DL) to diagnostic dermatology has been the subject of numerous studies, with some reporting skin lesion classification performance on curated datasets comparable to that of experienced dermatologists. Most skin disease images encountered in clinical settings are macroscopic, without dermoscopic information, and exhibit considerable variability. Further research is necessary to determine the generalizability of DL algorithms across populations and acquisition settings. To assess the extent to which DL can generalize to nondermoscopic datasets acquired at the primary-secondary care interface in the National Health Service (NHS); to explore how to obtain a clinically satisfactory performance on nonstandardized, real-world local data without the availability of large diagnostically labelled local datasets; and to measure the impact of pretraining DL algorithms on external, public datasets. Diagnostic macroscopic image datasets were created from previous referrals from primary to secondary care. These included 2213 images referred from primary care practitioners in NHS Tayside and 1510 images from NHS Forth Valley acquired by medical photographers. Two further datasets with identical diagnostic labels were obtained from sources in the public domain, namely the International Skin Imaging Collaboration (ISIC) dermoscopic dataset and the SD-260 nondermoscopic dataset. DL algorithms, specifically EfficientNets and Self-attention with Window-wise Inner-product based Network (SWIN) transformers, were trained using data from each of these datasets. Algorithms were also fine-tuned on images from the NHS datasets after pretraining on different data combinations, including the larger public-domain datasets. Receiver operating characteristic curves and the area under such curves (AUC) were used to assess performance. SWIN transformers tested on Forth Valley data had AUCs of 0.85 and 0.89 when trained on SD-260 and Forth Valley data, respectively. Training on SD-260 followed by fine tuning of Forth Valley data gave an AUC of 0.91. Similar effects of pretraining and tuning on local data were observed using Tayside data and EfficientNets. Pretraining on the larger dermoscopic image dataset (ISIC 2019) provided no additional benefit. Pretraining on public macroscopic images followed by tuning to local data gave promising results. Further improvements are needed to afford deployment in real clinical pathways. Larger datasets local to the target domain might be expected to yield further improved performance.</description><subject>Algorithms</subject><subject>Artificial Intelligence</subject><subject>Deep Learning</subject><subject>Dermatology</subject><subject>Dermoscopy - methods</subject><subject>Humans</subject><subject>Primary Health Care</subject><subject>Secondary Care</subject><subject>Skin Diseases - diagnostic imaging</subject><subject>Skin Diseases - pathology</subject><subject>State Medicine</subject><issn>0307-6938</issn><issn>1365-2230</issn><issn>1365-2230</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>EIF</sourceid><recordid>eNpVkcFu3CAQhlHVqNmmPfUecaxUucHGYJNbtWqaSJF6Sc4WjIctFQYH7ER5wzxW2ezmUHEYmPnnYzQ_IV9q9r1mil8Ajhfe67Fl7B3Z1FyKqmk4e082jLOukor3p-Rjzn8Zq3ndiQ_klHdKMdXyDXnZep2zs88u7GhC7aunmPxIJw0pZoizA-omvcNMXaDLH6RzKu_0XGWEGMZyo6ATluqCyWpAuuY9a0k6ZIuJetQplMxl4czegV5cDJnamOiIj-jjPGFYaLRUp8VZB077V5r3boehAHP066HpgA7lX0zT_-N9IidW-4yfj_GM3F_9vNteV7e_f91sf9xW0Ai5VGCsRDQgDdOybWXfdaqzohe9AdF0UCsw5cheNUz0rWgNqEYLLo3ojO0lPyNfD9w5xYcV8zJMLkMZVgeMax6a0imFqpUq0m8H6X6VOaEdjrsbajbsrRuKdcPRuqI-P4JXM5X8m_bNK_4PrACdSg</recordid><startdate>20240625</startdate><enddate>20240625</enddate><creator>Carse, Jacob</creator><creator>Süveges, Tamás</creator><creator>Chin, Gillian</creator><creator>Muthiah, Shareen</creator><creator>Morton, Colin</creator><creator>Proby, Charlotte</creator><creator>Trucco, Emanuele</creator><creator>Fleming, Colin</creator><creator>McKenna, Stephen</creator><scope>CGR</scope><scope>CUY</scope><scope>CVF</scope><scope>ECM</scope><scope>EIF</scope><scope>NPM</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>7X8</scope><orcidid>https://orcid.org/0000-0002-3292-4836</orcidid><orcidid>https://orcid.org/0000-0003-0530-2035</orcidid></search><sort><creationdate>20240625</creationdate><title>Classifying real-world macroscopic images in the primary-secondary care interface using transfer learning: implications for development of artificial intelligence solutions using nondermoscopic images</title><author>Carse, Jacob ; Süveges, Tamás ; Chin, Gillian ; Muthiah, Shareen ; Morton, Colin ; Proby, Charlotte ; Trucco, Emanuele ; Fleming, Colin ; McKenna, Stephen</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c256t-cbf6eebc6b0a644687797f5858bc527c19cbcbc6892058454bc92a536b57bf863</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Algorithms</topic><topic>Artificial Intelligence</topic><topic>Deep Learning</topic><topic>Dermatology</topic><topic>Dermoscopy - methods</topic><topic>Humans</topic><topic>Primary Health Care</topic><topic>Secondary Care</topic><topic>Skin Diseases - diagnostic imaging</topic><topic>Skin Diseases - pathology</topic><topic>State Medicine</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Carse, Jacob</creatorcontrib><creatorcontrib>Süveges, Tamás</creatorcontrib><creatorcontrib>Chin, Gillian</creatorcontrib><creatorcontrib>Muthiah, Shareen</creatorcontrib><creatorcontrib>Morton, Colin</creatorcontrib><creatorcontrib>Proby, Charlotte</creatorcontrib><creatorcontrib>Trucco, Emanuele</creatorcontrib><creatorcontrib>Fleming, Colin</creatorcontrib><creatorcontrib>McKenna, Stephen</creatorcontrib><collection>Medline</collection><collection>MEDLINE</collection><collection>MEDLINE (Ovid)</collection><collection>MEDLINE</collection><collection>MEDLINE</collection><collection>PubMed</collection><collection>CrossRef</collection><collection>MEDLINE - Academic</collection><jtitle>Clinical and experimental dermatology</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Carse, Jacob</au><au>Süveges, Tamás</au><au>Chin, Gillian</au><au>Muthiah, Shareen</au><au>Morton, Colin</au><au>Proby, Charlotte</au><au>Trucco, Emanuele</au><au>Fleming, Colin</au><au>McKenna, Stephen</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Classifying real-world macroscopic images in the primary-secondary care interface using transfer learning: implications for development of artificial intelligence solutions using nondermoscopic images</atitle><jtitle>Clinical and experimental dermatology</jtitle><addtitle>Clin Exp Dermatol</addtitle><date>2024-06-25</date><risdate>2024</risdate><volume>49</volume><issue>7</issue><spage>699</spage><epage>706</epage><pages>699-706</pages><issn>0307-6938</issn><issn>1365-2230</issn><eissn>1365-2230</eissn><abstract>The application of deep learning (DL) to diagnostic dermatology has been the subject of numerous studies, with some reporting skin lesion classification performance on curated datasets comparable to that of experienced dermatologists. Most skin disease images encountered in clinical settings are macroscopic, without dermoscopic information, and exhibit considerable variability. Further research is necessary to determine the generalizability of DL algorithms across populations and acquisition settings. To assess the extent to which DL can generalize to nondermoscopic datasets acquired at the primary-secondary care interface in the National Health Service (NHS); to explore how to obtain a clinically satisfactory performance on nonstandardized, real-world local data without the availability of large diagnostically labelled local datasets; and to measure the impact of pretraining DL algorithms on external, public datasets. Diagnostic macroscopic image datasets were created from previous referrals from primary to secondary care. These included 2213 images referred from primary care practitioners in NHS Tayside and 1510 images from NHS Forth Valley acquired by medical photographers. Two further datasets with identical diagnostic labels were obtained from sources in the public domain, namely the International Skin Imaging Collaboration (ISIC) dermoscopic dataset and the SD-260 nondermoscopic dataset. DL algorithms, specifically EfficientNets and Self-attention with Window-wise Inner-product based Network (SWIN) transformers, were trained using data from each of these datasets. Algorithms were also fine-tuned on images from the NHS datasets after pretraining on different data combinations, including the larger public-domain datasets. Receiver operating characteristic curves and the area under such curves (AUC) were used to assess performance. SWIN transformers tested on Forth Valley data had AUCs of 0.85 and 0.89 when trained on SD-260 and Forth Valley data, respectively. Training on SD-260 followed by fine tuning of Forth Valley data gave an AUC of 0.91. Similar effects of pretraining and tuning on local data were observed using Tayside data and EfficientNets. Pretraining on the larger dermoscopic image dataset (ISIC 2019) provided no additional benefit. Pretraining on public macroscopic images followed by tuning to local data gave promising results. Further improvements are needed to afford deployment in real clinical pathways. Larger datasets local to the target domain might be expected to yield further improved performance.</abstract><cop>England</cop><pmid>37990943</pmid><doi>10.1093/ced/llad400</doi><tpages>8</tpages><orcidid>https://orcid.org/0000-0002-3292-4836</orcidid><orcidid>https://orcid.org/0000-0003-0530-2035</orcidid><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier ISSN: 0307-6938
ispartof Clinical and experimental dermatology, 2024-06, Vol.49 (7), p.699-706
issn 0307-6938
1365-2230
1365-2230
language eng
recordid cdi_proquest_miscellaneous_2892659199
source Oxford University Press Journals; MEDLINE
subjects Algorithms
Artificial Intelligence
Deep Learning
Dermatology
Dermoscopy - methods
Humans
Primary Health Care
Secondary Care
Skin Diseases - diagnostic imaging
Skin Diseases - pathology
State Medicine
title Classifying real-world macroscopic images in the primary-secondary care interface using transfer learning: implications for development of artificial intelligence solutions using nondermoscopic images
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T05%3A16%3A31IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_cross&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Classifying%20real-world%20macroscopic%20images%20in%20the%20primary-secondary%20care%20interface%20using%20transfer%20learning:%20implications%20for%20development%20of%20artificial%20intelligence%20solutions%20using%20nondermoscopic%20images&rft.jtitle=Clinical%20and%20experimental%20dermatology&rft.au=Carse,%20Jacob&rft.date=2024-06-25&rft.volume=49&rft.issue=7&rft.spage=699&rft.epage=706&rft.pages=699-706&rft.issn=0307-6938&rft.eissn=1365-2230&rft_id=info:doi/10.1093/ced/llad400&rft_dat=%3Cproquest_cross%3E2892659199%3C/proquest_cross%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2892659199&rft_id=info:pmid/37990943&rfr_iscdi=true