Systematic comparison of 3D Deep learning and classical machine learning explanations for Alzheimer’s Disease detection

Black-box deep learning (DL) models trained for the early detection of Alzheimer’s Disease (AD) often lack systematic model interpretation. This work computes the activated brain regions during DL and compares those with classical Machine Learning (ML) explanations. The architectures used for DL wer...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Computers in biology and medicine 2024-03, Vol.170, p.108029-108029, Article 108029
Hauptverfasser: Bloch, Louise, Friedrich, Christoph M.
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page 108029
container_issue
container_start_page 108029
container_title Computers in biology and medicine
container_volume 170
creator Bloch, Louise
Friedrich, Christoph M.
description Black-box deep learning (DL) models trained for the early detection of Alzheimer’s Disease (AD) often lack systematic model interpretation. This work computes the activated brain regions during DL and compares those with classical Machine Learning (ML) explanations. The architectures used for DL were 3D DenseNets, EfficientNets, and Squeeze-and-Excitation (SE) networks. The classical models include Random Forests (RFs), Support Vector Machines (SVMs), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting (LightGBM), Decision Trees (DTs), and Logistic Regression (LR). For explanations, SHapley Additive exPlanations (SHAP) values, Local Interpretable Model-agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (GradCAM), GradCAM++ and permutation-based feature importance were implemented. During interpretation, correlated features were consolidated into aspects. All models were trained on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. The validation includes internal and external validation on the Australian Imaging and Lifestyle flagship study of Ageing (AIBL) and the Open Access Series of Imaging Studies (OASIS). DL and ML models reached similar classification performances. Regarding the brain regions, both types focus on different regions. The ML models focus on the inferior and middle temporal gyri, and the hippocampus, and amygdala regions previously associated with AD. The DL models focus on a wider range of regions including the optical chiasm, the entorhinal cortices, the left and right vessels, and the 4th ventricle which were partially associated with AD. One explanation for the differences is the input features (textures vs. volumes). Both types show reasonable similarity to a ground truth Voxel-Based Morphometry (VBM) analysis. Slightly higher similarities were measured for ML models. [Display omitted] •Comparing the interpretability of deep learning and machine learning for Alzheimer’s Disease.•3D deep learning models for Alzheimer’s Disease detection.•Systematic comparison of relevant regions using volumes, deep learning, and VBM.•Model calibration using Platt scaling to avoid biased model interpretation.•Aspect consolidation to control feature correlations in explainability methods.
doi_str_mv 10.1016/j.compbiomed.2024.108029
format Article
fullrecord <record><control><sourceid>proquest_cross</sourceid><recordid>TN_cdi_proquest_miscellaneous_2929131898</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><els_id>S0010482524001136</els_id><sourcerecordid>2929131898</sourcerecordid><originalsourceid>FETCH-LOGICAL-c397t-56354870e5a20daa8a090b1d4269e46d4169706bbb2a5410119bd910c2b818d13</originalsourceid><addsrcrecordid>eNqFkc1u1DAUhS1ERYfCKyBLbNhkuP5JYi9LBwpSpS5a1pZj36EeJXGwM4hhxWvwejxJHU2rSmxYWfL97rlH5xBCGawZsOb9bu3iMHUhDujXHLgs3wq4fkZWTLW6glrI52QFwKCSiten5GXOOwCQIOAFORVKgFItrMjh5pBnHOwcHF00bQo5jjRuqdjQDeJEe7RpDOM3akdPXW9zDs72dLDuLoz4NMafU2_HIhTHTLcx0fP-1x2GAdPf338y3YSMNiP1OKNboFfkZGv7jK8f3jPy9dPH24vP1dX15ZeL86vKCd3OVd2IWharWFsO3lplQUPHvOSNRtl4yRrdQtN1Hbe1LOkw3XnNwPFOMeWZOCPvjrpTit_3mGczhOywL2Yx7rPhmmsmmNKqoG__QXdxn8birlCCS66hbQuljpRLMeeEWzOlMNh0MAzMUo_Zmad6zFKPOdZTVt88HNh3y-xx8bGPAnw4AlgS-REwmewCjg59SCU242P4_5V76Xym4A</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2932429077</pqid></control><display><type>article</type><title>Systematic comparison of 3D Deep learning and classical machine learning explanations for Alzheimer’s Disease detection</title><source>MEDLINE</source><source>Elsevier ScienceDirect Journals Complete</source><creator>Bloch, Louise ; Friedrich, Christoph M.</creator><creatorcontrib>Bloch, Louise ; Friedrich, Christoph M. ; Alzheimer’s Disease Neuroimaging Initiative</creatorcontrib><description>Black-box deep learning (DL) models trained for the early detection of Alzheimer’s Disease (AD) often lack systematic model interpretation. This work computes the activated brain regions during DL and compares those with classical Machine Learning (ML) explanations. The architectures used for DL were 3D DenseNets, EfficientNets, and Squeeze-and-Excitation (SE) networks. The classical models include Random Forests (RFs), Support Vector Machines (SVMs), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting (LightGBM), Decision Trees (DTs), and Logistic Regression (LR). For explanations, SHapley Additive exPlanations (SHAP) values, Local Interpretable Model-agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (GradCAM), GradCAM++ and permutation-based feature importance were implemented. During interpretation, correlated features were consolidated into aspects. All models were trained on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. The validation includes internal and external validation on the Australian Imaging and Lifestyle flagship study of Ageing (AIBL) and the Open Access Series of Imaging Studies (OASIS). DL and ML models reached similar classification performances. Regarding the brain regions, both types focus on different regions. The ML models focus on the inferior and middle temporal gyri, and the hippocampus, and amygdala regions previously associated with AD. The DL models focus on a wider range of regions including the optical chiasm, the entorhinal cortices, the left and right vessels, and the 4th ventricle which were partially associated with AD. One explanation for the differences is the input features (textures vs. volumes). Both types show reasonable similarity to a ground truth Voxel-Based Morphometry (VBM) analysis. Slightly higher similarities were measured for ML models. [Display omitted] •Comparing the interpretability of deep learning and machine learning for Alzheimer’s Disease.•3D deep learning models for Alzheimer’s Disease detection.•Systematic comparison of relevant regions using volumes, deep learning, and VBM.•Model calibration using Platt scaling to avoid biased model interpretation.•Aspect consolidation to control feature correlations in explainability methods.</description><identifier>ISSN: 0010-4825</identifier><identifier>EISSN: 1879-0534</identifier><identifier>DOI: 10.1016/j.compbiomed.2024.108029</identifier><identifier>PMID: 38308870</identifier><language>eng</language><publisher>United States: Elsevier Ltd</publisher><subject>3D CNN ; Alzheimer Disease - diagnostic imaging ; Alzheimer's disease ; Amygdala ; Australia ; Brain ; Brain mapping ; Decision trees ; Deep Learning ; Disease detection ; GradCAM ; Humans ; Interpretable Machine Learning ; Learning algorithms ; LIME ; Machine Learning ; Magnetic Resonance Imaging - methods ; Medical imaging ; Morphometry ; Neurodegenerative diseases ; Neuroimaging ; Optic chiasm ; Permutations ; SHAP ; Support vector machines ; Ventricle ; Ventricles (cerebral)</subject><ispartof>Computers in biology and medicine, 2024-03, Vol.170, p.108029-108029, Article 108029</ispartof><rights>2024 The Author(s)</rights><rights>Copyright © 2024 The Author(s). Published by Elsevier Ltd.. All rights reserved.</rights><rights>2024. The Author(s)</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><cites>FETCH-LOGICAL-c397t-56354870e5a20daa8a090b1d4269e46d4169706bbb2a5410119bd910c2b818d13</cites></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://dx.doi.org/10.1016/j.compbiomed.2024.108029$$EHTML$$P50$$Gelsevier$$Hfree_for_read</linktohtml><link.rule.ids>315,781,785,3551,27926,27927,45997</link.rule.ids><backlink>$$Uhttps://www.ncbi.nlm.nih.gov/pubmed/38308870$$D View this record in MEDLINE/PubMed$$Hfree_for_read</backlink></links><search><creatorcontrib>Bloch, Louise</creatorcontrib><creatorcontrib>Friedrich, Christoph M.</creatorcontrib><creatorcontrib>Alzheimer’s Disease Neuroimaging Initiative</creatorcontrib><title>Systematic comparison of 3D Deep learning and classical machine learning explanations for Alzheimer’s Disease detection</title><title>Computers in biology and medicine</title><addtitle>Comput Biol Med</addtitle><description>Black-box deep learning (DL) models trained for the early detection of Alzheimer’s Disease (AD) often lack systematic model interpretation. This work computes the activated brain regions during DL and compares those with classical Machine Learning (ML) explanations. The architectures used for DL were 3D DenseNets, EfficientNets, and Squeeze-and-Excitation (SE) networks. The classical models include Random Forests (RFs), Support Vector Machines (SVMs), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting (LightGBM), Decision Trees (DTs), and Logistic Regression (LR). For explanations, SHapley Additive exPlanations (SHAP) values, Local Interpretable Model-agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (GradCAM), GradCAM++ and permutation-based feature importance were implemented. During interpretation, correlated features were consolidated into aspects. All models were trained on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. The validation includes internal and external validation on the Australian Imaging and Lifestyle flagship study of Ageing (AIBL) and the Open Access Series of Imaging Studies (OASIS). DL and ML models reached similar classification performances. Regarding the brain regions, both types focus on different regions. The ML models focus on the inferior and middle temporal gyri, and the hippocampus, and amygdala regions previously associated with AD. The DL models focus on a wider range of regions including the optical chiasm, the entorhinal cortices, the left and right vessels, and the 4th ventricle which were partially associated with AD. One explanation for the differences is the input features (textures vs. volumes). Both types show reasonable similarity to a ground truth Voxel-Based Morphometry (VBM) analysis. Slightly higher similarities were measured for ML models. [Display omitted] •Comparing the interpretability of deep learning and machine learning for Alzheimer’s Disease.•3D deep learning models for Alzheimer’s Disease detection.•Systematic comparison of relevant regions using volumes, deep learning, and VBM.•Model calibration using Platt scaling to avoid biased model interpretation.•Aspect consolidation to control feature correlations in explainability methods.</description><subject>3D CNN</subject><subject>Alzheimer Disease - diagnostic imaging</subject><subject>Alzheimer's disease</subject><subject>Amygdala</subject><subject>Australia</subject><subject>Brain</subject><subject>Brain mapping</subject><subject>Decision trees</subject><subject>Deep Learning</subject><subject>Disease detection</subject><subject>GradCAM</subject><subject>Humans</subject><subject>Interpretable Machine Learning</subject><subject>Learning algorithms</subject><subject>LIME</subject><subject>Machine Learning</subject><subject>Magnetic Resonance Imaging - methods</subject><subject>Medical imaging</subject><subject>Morphometry</subject><subject>Neurodegenerative diseases</subject><subject>Neuroimaging</subject><subject>Optic chiasm</subject><subject>Permutations</subject><subject>SHAP</subject><subject>Support vector machines</subject><subject>Ventricle</subject><subject>Ventricles (cerebral)</subject><issn>0010-4825</issn><issn>1879-0534</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>EIF</sourceid><recordid>eNqFkc1u1DAUhS1ERYfCKyBLbNhkuP5JYi9LBwpSpS5a1pZj36EeJXGwM4hhxWvwejxJHU2rSmxYWfL97rlH5xBCGawZsOb9bu3iMHUhDujXHLgs3wq4fkZWTLW6glrI52QFwKCSiten5GXOOwCQIOAFORVKgFItrMjh5pBnHOwcHF00bQo5jjRuqdjQDeJEe7RpDOM3akdPXW9zDs72dLDuLoz4NMafU2_HIhTHTLcx0fP-1x2GAdPf338y3YSMNiP1OKNboFfkZGv7jK8f3jPy9dPH24vP1dX15ZeL86vKCd3OVd2IWharWFsO3lplQUPHvOSNRtl4yRrdQtN1Hbe1LOkw3XnNwPFOMeWZOCPvjrpTit_3mGczhOywL2Yx7rPhmmsmmNKqoG__QXdxn8birlCCS66hbQuljpRLMeeEWzOlMNh0MAzMUo_Zmad6zFKPOdZTVt88HNh3y-xx8bGPAnw4AlgS-REwmewCjg59SCU242P4_5V76Xym4A</recordid><startdate>202403</startdate><enddate>202403</enddate><creator>Bloch, Louise</creator><creator>Friedrich, Christoph M.</creator><general>Elsevier Ltd</general><general>Elsevier Limited</general><scope>6I.</scope><scope>AAFTH</scope><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>8FD</scope><scope>FR3</scope><scope>JQ2</scope><scope>K9.</scope><scope>M7Z</scope><scope>NAPCQ</scope><scope>P64</scope><scope>7X8</scope></search><sort><creationdate>202403</creationdate><title>Systematic comparison of 3D Deep learning and classical machine learning explanations for Alzheimer’s Disease detection</title><author>Bloch, Louise ; Friedrich, Christoph M.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c397t-56354870e5a20daa8a090b1d4269e46d4169706bbb2a5410119bd910c2b818d13</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>3D CNN</topic><topic>Alzheimer Disease - diagnostic imaging</topic><topic>Alzheimer's disease</topic><topic>Amygdala</topic><topic>Australia</topic><topic>Brain</topic><topic>Brain mapping</topic><topic>Decision trees</topic><topic>Deep Learning</topic><topic>Disease detection</topic><topic>GradCAM</topic><topic>Humans</topic><topic>Interpretable Machine Learning</topic><topic>Learning algorithms</topic><topic>LIME</topic><topic>Machine Learning</topic><topic>Magnetic Resonance Imaging - methods</topic><topic>Medical imaging</topic><topic>Morphometry</topic><topic>Neurodegenerative diseases</topic><topic>Neuroimaging</topic><topic>Optic chiasm</topic><topic>Permutations</topic><topic>SHAP</topic><topic>Support vector machines</topic><topic>Ventricle</topic><topic>Ventricles (cerebral)</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Bloch, Louise</creatorcontrib><creatorcontrib>Friedrich, Christoph M.</creatorcontrib><creatorcontrib>Alzheimer’s Disease Neuroimaging Initiative</creatorcontrib><collection>ScienceDirect Open Access Titles</collection><collection>Elsevier:ScienceDirect:Open Access</collection><collection>Medline</collection><collection>MEDLINE</collection><collection>MEDLINE (Ovid)</collection><collection>MEDLINE</collection><collection>MEDLINE</collection><collection>PubMed</collection><collection>CrossRef</collection><collection>Technology Research Database</collection><collection>Engineering Research Database</collection><collection>ProQuest Computer Science Collection</collection><collection>ProQuest Health &amp; Medical Complete (Alumni)</collection><collection>Biochemistry Abstracts 1</collection><collection>Nursing &amp; Allied Health Premium</collection><collection>Biotechnology and BioEngineering Abstracts</collection><collection>MEDLINE - Academic</collection><jtitle>Computers in biology and medicine</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Bloch, Louise</au><au>Friedrich, Christoph M.</au><aucorp>Alzheimer’s Disease Neuroimaging Initiative</aucorp><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Systematic comparison of 3D Deep learning and classical machine learning explanations for Alzheimer’s Disease detection</atitle><jtitle>Computers in biology and medicine</jtitle><addtitle>Comput Biol Med</addtitle><date>2024-03</date><risdate>2024</risdate><volume>170</volume><spage>108029</spage><epage>108029</epage><pages>108029-108029</pages><artnum>108029</artnum><issn>0010-4825</issn><eissn>1879-0534</eissn><abstract>Black-box deep learning (DL) models trained for the early detection of Alzheimer’s Disease (AD) often lack systematic model interpretation. This work computes the activated brain regions during DL and compares those with classical Machine Learning (ML) explanations. The architectures used for DL were 3D DenseNets, EfficientNets, and Squeeze-and-Excitation (SE) networks. The classical models include Random Forests (RFs), Support Vector Machines (SVMs), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting (LightGBM), Decision Trees (DTs), and Logistic Regression (LR). For explanations, SHapley Additive exPlanations (SHAP) values, Local Interpretable Model-agnostic Explanations (LIME), Gradient-weighted Class Activation Mapping (GradCAM), GradCAM++ and permutation-based feature importance were implemented. During interpretation, correlated features were consolidated into aspects. All models were trained on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. The validation includes internal and external validation on the Australian Imaging and Lifestyle flagship study of Ageing (AIBL) and the Open Access Series of Imaging Studies (OASIS). DL and ML models reached similar classification performances. Regarding the brain regions, both types focus on different regions. The ML models focus on the inferior and middle temporal gyri, and the hippocampus, and amygdala regions previously associated with AD. The DL models focus on a wider range of regions including the optical chiasm, the entorhinal cortices, the left and right vessels, and the 4th ventricle which were partially associated with AD. One explanation for the differences is the input features (textures vs. volumes). Both types show reasonable similarity to a ground truth Voxel-Based Morphometry (VBM) analysis. Slightly higher similarities were measured for ML models. [Display omitted] •Comparing the interpretability of deep learning and machine learning for Alzheimer’s Disease.•3D deep learning models for Alzheimer’s Disease detection.•Systematic comparison of relevant regions using volumes, deep learning, and VBM.•Model calibration using Platt scaling to avoid biased model interpretation.•Aspect consolidation to control feature correlations in explainability methods.</abstract><cop>United States</cop><pub>Elsevier Ltd</pub><pmid>38308870</pmid><doi>10.1016/j.compbiomed.2024.108029</doi><tpages>1</tpages><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier ISSN: 0010-4825
ispartof Computers in biology and medicine, 2024-03, Vol.170, p.108029-108029, Article 108029
issn 0010-4825
1879-0534
language eng
recordid cdi_proquest_miscellaneous_2929131898
source MEDLINE; Elsevier ScienceDirect Journals Complete
subjects 3D CNN
Alzheimer Disease - diagnostic imaging
Alzheimer's disease
Amygdala
Australia
Brain
Brain mapping
Decision trees
Deep Learning
Disease detection
GradCAM
Humans
Interpretable Machine Learning
Learning algorithms
LIME
Machine Learning
Magnetic Resonance Imaging - methods
Medical imaging
Morphometry
Neurodegenerative diseases
Neuroimaging
Optic chiasm
Permutations
SHAP
Support vector machines
Ventricle
Ventricles (cerebral)
title Systematic comparison of 3D Deep learning and classical machine learning explanations for Alzheimer’s Disease detection
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-18T07%3A24%3A38IST&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=Systematic%20comparison%20of%203D%20Deep%20learning%20and%20classical%20machine%20learning%20explanations%20for%20Alzheimer%E2%80%99s%20Disease%20detection&rft.jtitle=Computers%20in%20biology%20and%20medicine&rft.au=Bloch,%20Louise&rft.aucorp=Alzheimer%E2%80%99s%20Disease%20Neuroimaging%20Initiative&rft.date=2024-03&rft.volume=170&rft.spage=108029&rft.epage=108029&rft.pages=108029-108029&rft.artnum=108029&rft.issn=0010-4825&rft.eissn=1879-0534&rft_id=info:doi/10.1016/j.compbiomed.2024.108029&rft_dat=%3Cproquest_cross%3E2929131898%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=2932429077&rft_id=info:pmid/38308870&rft_els_id=S0010482524001136&rfr_iscdi=true