{"id":5076,"date":"2023-11-06T18:05:16","date_gmt":"2023-11-06T23:05:16","guid":{"rendered":"https:\/\/ccna-ccnv.ca\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/"},"modified":"2024-12-03T14:58:30","modified_gmt":"2024-12-03T19:58:30","slug":"image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2","status":"publish","type":"ccna_publication","link":"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/","title":{"rendered":"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data"},"content":{"rendered":"<h2 class=\"title\">Abstract<\/h2>\n<div id=\"eng-abstract\" class=\"abstract-content selected\">\n<p><strong class=\"sub-title\">Objective:\u00a0<\/strong>This work investigates if deep learning (DL) models can classify originating site locations directly from magnetic resonance imaging (MRI) scans with and without correction for intensity differences.<\/p>\n<p><strong class=\"sub-title\">Material and methods:\u00a0<\/strong>A large database of 1880 T1-weighted MRI scans collected across 41 sites originally for Parkinson&rsquo;s disease (PD) classification was used to classify sites in this study. Forty-six percent of the datasets are from PD patients, while 54% are from healthy participants. After preprocessing the T1-weighted scans, 2 additional data types were generated: intensity-harmonized T1-weighted scans and log-Jacobian deformation maps resulting from nonlinear atlas registration. Corresponding DL models were trained to classify sites for each data type. Additionally, logistic regression models were used to investigate the contribution of biological (age, sex, disease status) and non-biological (scanner type) variables to the models&rsquo; decision.<\/p>\n<p><strong class=\"sub-title\">Results:\u00a0<\/strong>A comparison of the 3 different types of data revealed that DL models trained using T1-weighted and intensity-harmonized T1-weighted scans can classify sites with an accuracy of 85%, while the model using log-Jacobian deformation maps achieved a site classification accuracy of 54%. Disease status and scanner type were found to be significant confounders.<\/p>\n<p><strong class=\"sub-title\">Discussion:\u00a0<\/strong>Our results demonstrate that MRI scans encode relevant site-specific information that models could use as shortcuts that cannot be removed using simple intensity harmonization methods.<\/p>\n<p><strong class=\"sub-title\">Conclusion:\u00a0<\/strong>The ability of DL models to exploit site-specific biases as shortcuts raises concerns about their reliability, generalization, and deployability in clinical settings.<\/p>\n<\/div>\n<p><strong class=\"sub-title\">Keywords:\u00a0<\/strong>MRI; data harmonization; deep learning; multisite classification; shortcut learning.<\/p>\n","protected":false},"author":19,"featured_media":0,"template":"","meta":{"_acf_changed":false},"studies-relation":[],"class_list":["post-5076","ccna_publication","type-ccna_publication","status-publish","hentry"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v25.8 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data - CCNA - CCNV<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/\" \/>\n<meta property=\"og:locale\" content=\"fr_CA\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data - CCNA - CCNV\" \/>\n<meta property=\"og:description\" content=\"Abstract Objective:\u00a0This work investigates if deep learning (DL) models can ...\" \/>\n<meta property=\"og:url\" content=\"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/\" \/>\n<meta property=\"og:site_name\" content=\"CCNA - CCNV\" \/>\n<meta property=\"article:modified_time\" content=\"2024-12-03T19:58:30+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/ccna-ccnv.ca\/wp-content\/uploads\/2025\/01\/CCNA.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1200\" \/>\n\t<meta property=\"og:image:height\" content=\"675\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Estimation du temps de lecture\" \/>\n\t<meta name=\"twitter:data1\" content=\"1 minute\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/\",\"url\":\"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/\",\"name\":\"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data - CCNA - CCNV\",\"isPartOf\":{\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/#website\"},\"datePublished\":\"2023-11-06T23:05:16+00:00\",\"dateModified\":\"2024-12-03T19:58:30+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/#breadcrumb\"},\"inLanguage\":\"fr-CA\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/ccna-ccnv.ca\/fr\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/#website\",\"url\":\"https:\/\/ccna-ccnv.ca\/fr\/\",\"name\":\"Canadian Consortium on Neurodegeneration in Aging\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/#organization\"},\"alternateName\":\"CCNA - CCNV\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/ccna-ccnv.ca\/fr\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"fr-CA\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/#organization\",\"name\":\"Canadian Consortium on Neurodegeneration in Aging\",\"alternateName\":\"CCNA - CCNV\",\"url\":\"https:\/\/ccna-ccnv.ca\/fr\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"fr-CA\",\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/ccna-ccnv.ca\/wp-content\/uploads\/2025\/01\/ccna-logo-on-white-background.jpg\",\"contentUrl\":\"https:\/\/ccna-ccnv.ca\/wp-content\/uploads\/2025\/01\/ccna-logo-on-white-background.jpg\",\"width\":696,\"height\":696,\"caption\":\"Canadian Consortium on Neurodegeneration in Aging\"},\"image\":{\"@id\":\"https:\/\/ccna-ccnv.ca\/fr\/#\/schema\/logo\/image\/\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data - CCNA - CCNV","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/","og_locale":"fr_CA","og_type":"article","og_title":"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data - CCNA - CCNV","og_description":"Abstract Objective:\u00a0This work investigates if deep learning (DL) models can ...","og_url":"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/","og_site_name":"CCNA - CCNV","article_modified_time":"2024-12-03T19:58:30+00:00","og_image":[{"width":1200,"height":675,"url":"https:\/\/ccna-ccnv.ca\/wp-content\/uploads\/2025\/01\/CCNA.jpg","type":"image\/jpeg"}],"twitter_card":"summary_large_image","twitter_misc":{"Estimation du temps de lecture":"1 minute"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/","url":"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/","name":"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data - CCNA - CCNV","isPartOf":{"@id":"https:\/\/ccna-ccnv.ca\/fr\/#website"},"datePublished":"2023-11-06T23:05:16+00:00","dateModified":"2024-12-03T19:58:30+00:00","breadcrumb":{"@id":"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/#breadcrumb"},"inLanguage":"fr-CA","potentialAction":[{"@type":"ReadAction","target":["https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/ccna-ccnv.ca\/fr\/ccna_publication\/image-encoded-biological-and-non-biological-variables-may-be-used-as-shortcuts-in-deep-learning-models-trained-on-multisite-neuroimaging-data-2\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/ccna-ccnv.ca\/fr\/"},{"@type":"ListItem","position":2,"name":"Image-encoded biological and non-biological variables may be used as shortcuts in deep learning models trained on multisite neuroimaging data"}]},{"@type":"WebSite","@id":"https:\/\/ccna-ccnv.ca\/fr\/#website","url":"https:\/\/ccna-ccnv.ca\/fr\/","name":"Canadian Consortium on Neurodegeneration in Aging","description":"","publisher":{"@id":"https:\/\/ccna-ccnv.ca\/fr\/#organization"},"alternateName":"CCNA - CCNV","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/ccna-ccnv.ca\/fr\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"fr-CA"},{"@type":"Organization","@id":"https:\/\/ccna-ccnv.ca\/fr\/#organization","name":"Canadian Consortium on Neurodegeneration in Aging","alternateName":"CCNA - CCNV","url":"https:\/\/ccna-ccnv.ca\/fr\/","logo":{"@type":"ImageObject","inLanguage":"fr-CA","@id":"https:\/\/ccna-ccnv.ca\/fr\/#\/schema\/logo\/image\/","url":"https:\/\/ccna-ccnv.ca\/wp-content\/uploads\/2025\/01\/ccna-logo-on-white-background.jpg","contentUrl":"https:\/\/ccna-ccnv.ca\/wp-content\/uploads\/2025\/01\/ccna-logo-on-white-background.jpg","width":696,"height":696,"caption":"Canadian Consortium on Neurodegeneration in Aging"},"image":{"@id":"https:\/\/ccna-ccnv.ca\/fr\/#\/schema\/logo\/image\/"}}]}},"_links":{"self":[{"href":"https:\/\/ccna-ccnv.ca\/fr\/wp-json\/wp\/v2\/ccna_publication\/5076","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ccna-ccnv.ca\/fr\/wp-json\/wp\/v2\/ccna_publication"}],"about":[{"href":"https:\/\/ccna-ccnv.ca\/fr\/wp-json\/wp\/v2\/types\/ccna_publication"}],"author":[{"embeddable":true,"href":"https:\/\/ccna-ccnv.ca\/fr\/wp-json\/wp\/v2\/users\/19"}],"wp:attachment":[{"href":"https:\/\/ccna-ccnv.ca\/fr\/wp-json\/wp\/v2\/media?parent=5076"}],"wp:term":[{"taxonomy":"studies-relation","embeddable":true,"href":"https:\/\/ccna-ccnv.ca\/fr\/wp-json\/wp\/v2\/studies-relation?post=5076"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}