Differential diagnosis of dementias using in vivo MRI and data-driven disease progression modelling: a case study in Alzheimer’s disease and dementia with Lewy bodies

Differential diagnosis of dementias using in vivo MRI and data-driven disease progression modelling: a case study in Alzheimer’s disease and dementia with Lewy bodies | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Differential diagnosis of dementias using in vivo MRI and data-driven disease progression modelling: a case study in Alzheimer’s disease and dementia with Lewy bodies View ORCID Profile Gonzalo Castro Leal , Ajay Konuri , Alexandra L. Young , Niloufar Zebarjadi , Annegret Habich , Nicolás Castellanos-Perilla , María Camila Gonzalez , John-Paul Taylor , Michael Firbank , Simon J.G. Lewis , Daniel Alcolea , Alexandre Bejanin , Kurt Segers , Ahmet Turan Isik , Bedia Samanci , Consuelo Cháfer-Pericás , Christian Lambert , Ramón Landin-Romero , Rohan Bhome , Ivelina Dobreva , Rimona S. Weil , Zuzana Walker , Dag Aarsland , Eric Westman , Daniel Ferreira , Elie Matar , Neil P. Oxtoby doi: https://doi.org/10.1101/2025.10.03.25337171 Gonzalo Castro Leal 1 UCL Hawkes Institute, University College of London , London, UK 2 UCL Department of Computer Science, University College of London , London, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Gonzalo Castro Leal For correspondence: gonzalo.leal.23{at}ucl.ac.uk Ajay Konuri 3 School of Medical Sciences, Faculty of Medicine and Health, University of Sydney , Australia 4 Parkinson’s Disease Research Clinic, Macquarie University , Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Alexandra L. Young 1 UCL Hawkes Institute, University College of London , London, UK 2 UCL Department of Computer Science, University College of London , London, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Niloufar Zebarjadi 6 Division of Clinical Geriatrics, Department of Neurobiology, Care Sciences and Society (NVS), Karolinska Institutet , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Annegret Habich 6 Division of Clinical Geriatrics, Department of Neurobiology, Care Sciences and Society (NVS), Karolinska Institutet , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Nicolás Castellanos-Perilla 17 Division of Clinical Geriatrics, Center for Alzheimer Research, Department of Neurobiology, Care Sciences and Society (NVS), Karolinska Institutet , Stockholm, Sweden 18 Department of Clinical Medicine, University of Bergen , Bergen, Norway 19 Centre for Age-Related Medicine (SESAM), Stavanger University Hospital , Stavanger, Norway Find this author on Google Scholar Find this author on PubMed Search for this author on this site María Camila Gonzalez 19 Centre for Age-Related Medicine (SESAM), Stavanger University Hospital , Stavanger, Norway Find this author on Google Scholar Find this author on PubMed Search for this author on this site John-Paul Taylor 23 Translational and Clinical Research Institute , Campus for Ageing and Vitality, Newcastle Upon Tyne, United Kingdom Find this author on Google Scholar Find this author on PubMed Search for this author on this site Michael Firbank 23 Translational and Clinical Research Institute , Campus for Ageing and Vitality, Newcastle Upon Tyne, United Kingdom Find this author on Google Scholar Find this author on PubMed Search for this author on this site Simon J.G. Lewis 21 Faculty of Medicine, Health and Human Sciences, Macquarie Medical School, Macquarie University , Australia 4 Parkinson’s Disease Research Clinic, Macquarie University , Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Daniel Alcolea 7 Sant Pau Memory Unit, IR SANT PAU, Hospital de la Santa Creu i Sant Pau , Barcelona, Spain 8 Centro de Investigación Biomédica en Red en Enfermedades Neurodegenerativas (CIBERNED) , Madrid, Spain Find this author on Google Scholar Find this author on PubMed Search for this author on this site Alexandre Bejanin 7 Sant Pau Memory Unit, IR SANT PAU, Hospital de la Santa Creu i Sant Pau , Barcelona, Spain 8 Centro de Investigación Biomédica en Red en Enfermedades Neurodegenerativas (CIBERNED) , Madrid, Spain Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kurt Segers 9 Neurology and Geriatrics Department, Brugmann University Hospital, Université Libre De Bruxelles , Brussels, Belgium Find this author on Google Scholar Find this author on PubMed Search for this author on this site Ahmet Turan Isik 10 Unit for Brain Aging and Dementia, Department of Geriatric Medicine, Dokuz Eylul University, School of Medicine , Izmir, Turkey Find this author on Google Scholar Find this author on PubMed Search for this author on this site Bedia Samanci 11 Behavioral Neurology and Movement Disorders Unit, Department of Neurology, Istanbul Faculty of Medicine, Istanbul University , Istanbul, Turkey Find this author on Google Scholar Find this author on PubMed Search for this author on this site Consuelo Cháfer-Pericás 12 Instituto de Investigación Sanitaria La Fe , Valencia, Spain Find this author on Google Scholar Find this author on PubMed Search for this author on this site Christian Lambert 13 Functional Imaging Laboratory, Department of Imaging Neuroscience, Institute of Neurology, University College London , London, United Kingdom Find this author on Google Scholar Find this author on PubMed Search for this author on this site Ramón Landin-Romero 5 Central Clinical School and Brain and Mind Centre, Faculty of Medicine and Health, University of Sydney , Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Rohan Bhome 1 UCL Hawkes Institute, University College of London , London, UK 14 Dementia Research Centre, University College London , London, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Ivelina Dobreva 14 Dementia Research Centre, University College London , London, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Rimona S. Weil 14 Dementia Research Centre, University College London , London, UK 15 National Hospital for Neurology and Neurosurgery, University London Hospitals NHS Foundation Trust , United Kingdom Find this author on Google Scholar Find this author on PubMed Search for this author on this site Zuzana Walker 24 Division of Psychiatry, University College London , London, UK 25 Essex Partnership University NHS Foundation Trust , Essex, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Dag Aarsland 19 Centre for Age-Related Medicine (SESAM), Stavanger University Hospital , Stavanger, Norway 22 Centre for Healthy Brain Ageing, Institute of Psychiatry, Psychology, and Neuroscience, King’s College London , London, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Eric Westman 6 Division of Clinical Geriatrics, Department of Neurobiology, Care Sciences and Society (NVS), Karolinska Institutet , Stockholm, Sweden 16 The Ageing Epidemiology Research Unit, School of Public Health, Imperial College London , London, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Daniel Ferreira 17 Division of Clinical Geriatrics, Center for Alzheimer Research, Department of Neurobiology, Care Sciences and Society (NVS), Karolinska Institutet , Stockholm, Sweden 20 Facultad de Ciencias de La Salud, Universidad Fernando Pessoa Canarias , Las Palmas, Spain Find this author on Google Scholar Find this author on PubMed Search for this author on this site Elie Matar 5 Central Clinical School and Brain and Mind Centre, Faculty of Medicine and Health, University of Sydney , Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Neil P. Oxtoby 1 UCL Hawkes Institute, University College of London , London, UK 2 UCL Department of Computer Science, University College of London , London, UK Find this author on Google Scholar Find this author on PubMed Search for this author on this site Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Alzheimer’s disease (AD) and Dementia with Lewy bodies (DLB) are neurodegenerative diseases often sharing clinical symptoms, causing frequent misdiagnoses. Using data from multiple cohorts we improve and apply the SuStaIn algorithm on regional brain volumes from magnetic resonance imaging (MRI), then assess biomarker/phenotype/histopathology associations among the discovered atrophy subtypes. The three data-driven subtypes of brain atrophy show divergent clinical/biomarker/histopathological profiles that could support differential diagnosis. Both clinical syndrome and post-mortem diagnosis aligned imperfectly but plausibly with subtype: Limbic (more AD), Cortical (more DLB). The Limbic subtype showed lower cerebrospinal fluid (CSF) amyloid- β , higher CSF phosphorylated tau, worse memory, and fewer hallucinations than the Cortical subtype. Our novel data-driven transdiagnostic approach shows promise for supporting in vivo differential diagnosis using only MRI. Introduction The two most common neurodegenerative types of dementia in the elderly are Alzheimer’s disease (AD) dementia and Lewy body dementia (LBD). AD is defined by the presence of extracellular amyloid-beta (A β ) plaques ( Masters et al., 1985 ; Thal et al., 2002 ) and intracellular neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau (p-tau) protein ( Braak et al., 2006 ; Braak and Braak, 1991 ). LBD, which includes Parkinson’s disease with dementia (PDD) and dementia with Lewy bodies (DLB), is pathologically characterized by inclusions of α -synuclein ( α Syn) called Lewy bodies and Lewy neurites ( Spillantini et al., 1997 ). Despite the evidence of DLB being a unique phenotype, there is significant overlap in the presentation of DLB and AD, especially at early stages, often leading to misdiagnosis ( McKeith et al., 2017 ; Walker et al., 2015 ). The overlapping clinical presentation of AD and DLB presents challenges for differential diagnosis in the clinic. This is reflected in the high specificity ( Rizzo et al., 2018 ) but poor sensitivity of the DLB diagnostic criteria ( Nelson et al., 2010 ). This is compounded by the high rates of AD co-pathology in DLB ( Barker et al., 2002 ; Wei et al., 2023 ; Plastini et al., 2024 ). For example, within the National Alzheimer’s Coordinating Center (NACC) database, over a third of patients with autopsy data available that were diagnosed clinically with AD had concurrent LB pathology, and almost half of those diagnosed at autopsy with LBD had a clinical diagnosis of AD ( Wei et al., 2023 ). Evidence from pathologically confirmed cohorts suggests that higher severity of AD co-pathology obscures the expression of clinical features of DLB ( Ferman et al., 2020 ). In both conditions, the presence of significant co-pathology is also associated with a more aggressive disease course ( Malek-Ahmadi et al., 2019 ; Tan et al., 2025 ). This significant clinical and pathological overlap highlights the need for methods capable of harnessing biomarker data to identify and characterize DLB and AD co-pathology in vivo – including, but not limited to, cerebrospinal fluid (CSF) or plasma protein levels ( Janelidze et al., 2020 ; Lantero Rodriguez et al., 2020 ; Gonzalez et al., 2022 ; Irwin et al., 2018 ; Chouliaras et al., 2022 ; Coughlin et al., 2019 ; Franzmeier et al., 2025 ), clinical presentation ( Ferman et al., 2020 ; Coughlin et al., 2019 ; Brenowitz et al., 2017 ; Kraybill et al., 2005 ; Ryman et al., 2021 ), genetic risk factors ( Corder et al., 1994 ; Verghese et al., 2011 ; Tsuang et al., 2013 ), and image derived biomarkers. MRI currently plays only a supportive role in the clinical diagnosis of DLB ( McKeith et al., 2017 ). However, an increasing number of studies have explored the utility of structural MRI for differential diagnosis ( Blanc et al., 2022 , 2016 ; Kantarci et al., 2016 ; Khadhraoui et al., 2022 ; Roquet et al., 2017 ), particularly region-specific differences in brain volume between DLB and AD. Brain regions that might be sensitive or specific to DLB include fronto-insular, temporal, parietal, occipital, and olfactory cortices ( Blanc et al., 2022 ; Roquet et al., 2017 ), as well as the insula, cingulate, amygdala, entorhinal cortex, parahippocampal gyrus, and hippocampus ( Blanc et al., 2016 ; Khadhraoui et al., 2022 ; Roquet et al., 2017 ). Coupled with the wealth of literature on AD-specific brain atrophy, MRI-based biomarkers such as regional brain volumes seem promising for the differential diagnosis of DLB and AD. In addition to the pathological and clinical overlap between AD and DLB, there is increasing evidence of biological heterogeneity within each dementia, that is, biomarker-based disease subtypes ( Ferreira et al., 2020 ). This compounds the challenge of differential diagnosis because such biomarker-based subtypes of neurodegenerative disease may have different clinical phenotypes. The good news is that this limitation is becoming addressable through the increasing availability of large datasets that allow clustering of individuals independently of diagnostic group, allowing exploration of heterogeneity and subtypes within these conditions ( Verdi et al., 2021 ). Several studies have attempted to subtype AD and more recently DLB using MRI. In AD, studies typically converge on similar subtypes, namely: Typical AD, Limbic-predominant, Hippocampal-sparing and Minimal atrophy ( Dong et al., 2017 ; Hwang et al., 2016 ; Noh et al., 2014 ; Park et al., 2017 ; Poulakis et al., 2018 ; Varol et al., 2017 ) (see also Ferreira et al. review paper ( Ferreira et al., 2020 )). In an example in DLB, Inguanzo et al. ( Inguanzo et al., 2023 ) used gray matter atrophy measured in vivo to identify distinct spatial clusters within DLB driven by the pallidum, caudate, cingulate, and olfactory cortex. However, this type of clustering can be confounded by disease severity, which limits the interpretability of the clusters, such as in terms of disease mechanisms or co-pathology ( Young et al., 2024 ; Alexander et al., 2021 ; Poulakis et al., 2022 ). The Subtype and Stage Inference (SuStaIn) algorithm ( Young et al., 2018 ) leverages data-driven disease progression modeling ( Young et al., 2024 ), unsupervised machine learning and cross-sectional biomarker information to identify subject clusters that capture both spatial and temporal progression simultaneously. This algorithm has been applied in multiple non-DLB dementias in a top-down approach – seeking to understand clinical heterogeneity in terms of biomarker-based clusters, typically from medical imaging data ( Young et al., 2018 ; Mastenbroek et al., 2024 ; Vogel et al., 2021 ; Young et al., 2023 , 2021 ). In the current study, we leverage SuStaIn to study two clinically enriched populations of syndromic AD and DLB patients, which allows us to explore similarities and differences in atrophy patterns. We produce, for the first time, a combined transdiagnostic subtyping model and assess its performance for differential diagnosis, validating on the subset of available postmortem data. Materials and Methods Subtyping was performed using SuStaIn ( Young et al., 2018 ) based on in vivo MRI volumetric measurements (biomarkers) from cohorts of patients diagnosed with DLB or AD. This algorithm clusters individuals into subtypes based on disease progression modeling, i.e., each subtype has a unique disease progression profile. Here we used Z-score SuStaIn, which uses a piecewise linear Z-score model to describe disease biomarker progression, i.e., as accumulation of atrophy in sequential stages of increased severity. SuStaIn simultaneously optimizes the subtype clusters and their disease progression profiles, building on the methods developed for the event-based model ( Fonteijn et al., 2012 ). The likelihood equation to optimize is described in: Where X ={ x ij | i = 1, …, I ; j = 1, …, J } is a set of biomarker measurements i for subject j, M is the overall SuStaIn model, C is the number of clusters; f c is the proportion of subjects assigned to a particular subtype, N is the number of Z-score events. The model requires a series of biomarkers (spatial dimension) and a set of events (temporal dimension) to be predefined. Finally, individuals are assigned to either their most probable subtype or are deemed unable to be subtyped if their likelihood of being at stage 0 (reflecting no detectable atrophy) is higher than that of being at any subtype. Individuals with assigned subtypes are also assigned the stage with the highest likelihood for their assigned subtype. SuStaIn configuration is described further below. Data All data were collected with appropriate informed consent under ethically approved clinical studies. Our secondary analysis of these data was approved by the UCL Research Ethics Committee under application 8019/005 and the UCL Department of Computer Science Research Ethics Committee under application UCL-CSREC-209-B. The data come from multiple cohorts: Alzheimer’s Disease Neuroimaging Initiative (ADNI – http://adni.loni.usc.edu ), National Alzheimer’s Coordinating Center (NACC – https://naccdata.org ), European DLB Consortium (E-DLB – https://www.e-dlb.com ), Parkinson’s Disease Biomarkers Program (PDBP – https://pdbp.ninds.nih.gov/ about) and other inhouse cohorts – Vision in Parkinson’s Disease (ViPD) and University of Sydney (‘Predicting Parkinson’s and Dementia’, PreD cohort). Inclusion criteria: We included data from participants who were clinically diagnosed at the last available visit with AD, DLB, MCI or iRBD, and who had an available 3D T1w MRI with spatial resolution no greater than 1.5 mm in each direction. Cognitively Normal (CN) participant data was used as a normative group (described later). The number of included individuals (and cognitive status) from each cohort is as follows: ADNI (790 CN, 997 MCI, 360 AD Dementia at baseline), NACC (878 CN, 560 MCI, 557 Dementia at baseline – 509 AD and 48 DLB), E-DLB (158 CN, 121 MCI, 183 Dementia at baseline – 137 DLB and 46 AD), PDBP (16 DLB Dementia), and in-house cohorts (32 CN, 37 iRBD, 63 DLB Dementia). Basic demographic data from all subjects are presented in Table 1 . View this table: View inline View popup Download powerpoint Table 1. Basic demographic information by cognitive status. There is also data that is cohort specific and is used in post-hoc analyses. Domain specific cognitive scores – memory, language, executive and visuospatial function – are available for some subjects in ADNI and NACC datasets. This data is provided by the Alzheimer’s Disease Sequencing Project Phenotype Harmonization Consortium (ADSP-PHC; Mukherjee et al. 2023 ). ADNI provides CSF measurements of A β 1-42 and p-tau 181 analysed by the electrochemiluminescence immunoassays (ECLIA) Elecsys. The measuring range of the assay is 200 to 1700 pg/mL for A β and 8 to 120 pg/mL for p-tau and thus we exclude values beyond these cutoffs. Additionally, ADNI also provides Apolipoprotein E ϵ 4 (APOE4) allele number of copies. E-DLB and other in-house cohorts also report the presence of four core clinical features of DLB — hallucinations, RBD, cognitive fluctuations and parkinsonism. Each centre recorded the presence/absence of clinical features based on the 2005 International Consensus Criteria for probable DLB ( McKeith et al., 2005 ) to allow for harmonized diagnosis across all centres because many of the patients were assessed before the 2017 International Consensus Criteria ( McKeith et al., 2017 ). Finally, ADNI and NACC also provide autopsy information for some subjects and following the criteria in Wei et al. (2023) we formulate the AD, LBD and AD+LBD neuropathology diagnosis. This information is summarised in Table S1. Imaging Biomarkers Structural T1-weighted MPRAGE-like MRI scans were processed by FreeSurfer ( https://surfer.nmr.mgh . harvard.edu) and parcellated using the Desikan-Killiany atlas parcellation to produce regional volumetric data ( Fischl et al., 2004 ; Desikan et al., 2006 ). Available scans were processed using FreeSurfer version 7.1.1, except for E-DLB’s which were processed with FreeSurfer version 7.3.2 (compatible with 7.1.1). Volumetric measurements from both hemispheres were combined into bilateral regions of interest (ROIs) described in Table 2 —-these are the imaging biomarkers used in this study. Potentially confounding trends due to age, sex, and intracranial volume (ICV) were removed from all the data via the residuals method ( Gur et al., 1991 ; Fotenos et al., 2008 ; Voevodskaya et al., 2014 ) with ICV outliers discarded pre-emptively (beyond ± 1.96 standard deviations of the mean), and then converted into z-scores robust to outliers using the median and median absolute deviation. For this preprocessing, we performed robust linear model (RLM) regression on data from cognitively normal controls – with cognitive follow-up assessments of minimum of 3 years after the scan, thus ensuring that these subjects did not have any underlying pathology that was not detected at the MRI assessment. Additional considerations on multiple cohort and inter-scanner variability are described in the Appendix. View this table: View inline View popup Download powerpoint Table 2. Imaging biomarkers used as inputs to our modelling. Regions of interest and their component sub-regions are shown, from the Desikan-Killiany atlas. ROI = Region of Interest. Subtyping Model Configuration Table 2 shows the imaging biomarkers used as input features in the subtyping model, which were chosen from the available Desikan-Killiany ROIs based on previous work from the literature where a statistically significant signal was measured in AD, DLB or both. The events are defined as the 0.67, 1 and 2 z-scores (i.e. 75%, 86% and 98% percentiles of atrophy severity) to capture mid-to-late atrophy progression. We train a total of two models – one each for the AD and DLB clinical cohorts – and combine them into a single supermodel. A limitation of the off-the-shelf SuStaIn subtype and stage assignment (described above) is the assumption that all individuals with some atrophy should be classifiable. To address this limitation, we formulate an auxiliary (null) subtype with a uniform probability distribution of events throughout the sequence. This prevents participant data having a non-zero atrophy pattern that does not match the prevailing clusters from being forced into one of the clusters. Subjects assigned to this null subtype, along with those unable to be subtyped or beyond stage 25 – i.e. subtypes become indistinguishable at late stages due to convergence of the atrophy signatures – were excluded from further analyses. All subtypes obtained are combined into a final inference model used for transdiagnostic subtyping and staging (differential diagnosis) of individuals from all cohorts. Statistical Analysis The optimal number of subtypes supported by the data was determined through 10-fold cross-validation information criteria (CVIC). Similarity between solutions was measured using the Hellinger distance as it has in previous work ( Young et al., 2018 ; Oxtoby et al., 2021 ). The optimal and sufficiently different subtypes – along with the null subtype – were combined into our transdiagnostic model. The transdiagnostic SuStaIn model (TranSuStaIn) was evaluated using several clinical and biological markers. Demographic (age, sex, diagnosis) and cognitive (status and scores) variables were available for all individuals, while additional subsets of subjects had complementary data, including cognitive domain-specific scores, DLB core feature assessments, CSF protein measurements (A β and p-tau), genetic (APOE4) information, and neuropathological data. The model provides both spatial (subtype) and pseudo-temporal (stage) variables. For continuous-valued outcome measures, we use RLMs to evaluate associations with model stage at the whole population level, before evaluating interactions between stage and subtype. In order to ensure a Gaussian distribution of residuals, some biomarkers need prior adjustment: A β and p-tau are log transformed and MMSE scores are transformed as described by Jacqmin-Gadda et al. (1997) into the square root of the number of errors. Before exploring group differences, we regress out the influence of model stage and other relevant covariates – and reversing any transformations applied. After evaluating the normality of the residuals using the Shapiro-Wilk test, subtype-specific associations were assessed using one-way ANOVA or the Kruskal-Wallis test. Where group differences were supported, we conducted pairwise comparisons using a T-Test or Mann-Whitney U test and reported effect sizes using Cohen’s d or Cliff’s delta. For categorical-valued outcomes, we explored subtype associations using a Chi-squared test and Cramer’s V for effect sizes (or the residuals). Given the number of comparisons, we adopted a conservative significance threshold at p = .01 and reported confidence intervals (CI) at the 99% level. Results SuStaIn subtypes across DLB and AD cohorts The SuStaIn algorithm was initially applied on each clinical cohort separately ( Figure 1 ). The Positional Variance Diagram visualization in Figure 1A depicts color-coded spatiotemporal atrophy (left-right): red (z=0.67), magenta (z=1) and blue (z=2). Based on CVIC values ( Figure 1B ), we identified two distinct subtypes within each population, characterized by their earliest-affected brain regions: Cortical (DLB), Limbic (in both AD and DLB), and Cortico-Limbic (AD). Download figure Open in new tab Figure 1. SuStaIn model results for DLB (top half) and AD (bottom half) cohorts, showing subtype progression patterns and model validation. The figure is organized in four rows representing distinct subtypes: Cortical (i), Limbic DLB Variant (ii), Limbic AD Variant (iii), and Cortico-Limbic (iv). (A) Each row contains 2 panels: Positional Variance Diagram (left) displaying the uncertainty in the ordering of events, with intensity indicating probability of event(red: z-score=0.67, magenta: z-score=1, blue: z-score=2); and Maximum Likelihood Sequence (right) showing the temporal evolution of regional atrophy (x-axis: stages 1-33; y-axis: brain regions), where colors indicate severity (same thresholds as before). (B) CVIC values across 10-Folds, where the minimum for both models is at 2 subtypes, thus indicating that the 2-cluster solution is best in both scenarios. The Cortical subtype progression shown in Figure 1A(i) is characterized by widespread cortical involvement (z-score ≥ 1, magenta), with delayed hippocampal, amygdala, entorhinal, and parahippocampal atrophy. The Limbic subtypes in clinical DLB ( Figure 1A(ii) ) and clinical AD ( Figure 1A(iii) ) demonstrated similar progression patterns to each other, showing early moderate atrophy (z-score ≥ 1, magenta) in three key regions: hippocampus, amygdala, and entorhinal cortex. These regions progressed to severe atrophy (z-score ≥ 2) by the middle stages of disease progression. Notably, the DLB-enriched variant in Figure 1A(ii) showed earlier involvement of cortical regions (middle temporal and parietal). The Cortico-limbic subtype ( Figure 1A(iv) ) begins with atrophy in the hippocampus, amygdala, medial temporal lobe, and parietal regions. Comparison of AD and DLB subtypes Quantitative comparison using Hellinger distance demonstrated that the Cortical subtype and both Limbic subtypes were the most divergent from each other (0.718 and 0.61 for the comparisons of the Cortical subtype with the AD and DLB limbic variants, respectively). We observed varying degrees of overlap between the other subtypes, with moderate similarity between the Cortico-Limbic (AD) and both DLB subtypes (0.518 and 0.525 for the comparisons of the Cortico-Limbic with the Cortical and Limbic subtypes, respectively). The greatest similarity was observed between the two Limbic subtypes (0.254). Given the high degree of similarity with its homologue from the AD cohort, we discarded the Limbic subtype from the DLB variant, thus avoiding redundancy in the final transdiagnostic model. The other subtypes were combined – including the null subtype – into a final TranSuStaIn model (see Subtyping Model Configuration) that we use for transdiagnostic subtyping and staging (differential diagnosis) of individuals from all cohorts. TranSuStaIn identifies disease-specific signatures After subtyping and staging individuals from all datasets using TranSuStaIn, the final clinical assignments per model subtype were as follows: 447 (containing 97 and 93 subjects with final clinical diagnosis of AD and DLB respectively) to the Cortical subtype, 568 (299 AD and 47 DLB) to the Cortico-Limbic subtype, 971 (533 AD and 48 DLB) to the Limbic subtype, 1191 to the Null subtype (248 AD and 138 DLB) and 1501 to Stage 0 (81 AD and 29 DLB). We found evidence that clinical diagnosis is significantly and moderately associated with subtype ( χ 2 = 173.089, p < 1 × 10 8 , V = 0.28). AD diagnosis is strongly associated with the Limbic subtype and weakly associated with the Cortical subtype (residuals of 2.26 and –4.85, respectively) and the opposite for DLB diagnosis (residuals –5.03 and 10.79, respectively). Age-related changes across subtypes and stages Figure 2A presents age trends across stages for each subgroup. At the population level, there is no evidence of an association between model stage and age (mean trend = 0.05 years per stage, CI = [-0.02, 0.12]). However, subtype-specific interactions were observed, indicating that the age trajectory across stages varied by subtype – most notably in the Cortical subtype (0.232 years older per model stage), which was statistically significantly different from both the group-level trend and all other subtype trends (p < 10 −3 ), which themselves were not significantly different from zero (Limbic and Cortico-Limbic subtypes). At the subtype level (adjusted for stage), there was also evidence of the Cortical (mean age 70.8, CI = [69.7, 71.9]) and Cortico-Limbic (mean age 72.1, CI = [71.1, 73.0]) subtypes being younger (p < 10 −6 , Cliff’s delta = 0.28 and 0.19) than the Limbic subtype (mean age = 74.8, CI = [74.1, 75.5]). Download figure Open in new tab Figure 2. (A) Scatter plot with regression lines showing age trends as a function of model stage (atrophy severity) per subtype, with density distributions displayed on top and right margins. (B) Boxplots showing disease stage by cognitive status (CN=Cognitively Normal, MCI=Mild Cognitive Impairment, Dementia) across four AD subtypes. Statistical significance indicated by *p < 10 −2 and ***p < 10 −6 , with line thickness denoting effect size (dashed=small, thin=medium, thick=large). Cognitive function across subtypes and stages Figure 2B shows model stage distributions by cognitive status within each subtype, i.e., subtype-specific cumulative spatiotemporal atrophy severity in CN, MCI and Dementia groups. This reveals that cognitive decline is associated with accumulating atrophy within subtype (p < 10 −6 in all subtypes). Mann-Whitney U tests indicate statistically significant differences (p < 10 −6 for all comparisons except CN vs MCI and MCI vs Dementia in the Cortical subtype, with p = 6.4 × 10 −4 and 0.00497 respectively) with moderate to large effect sizes (Cliff’s delta = 0.18 to 0.71), implying the potential for diagnostic value of stage within subtype. Figure 3A shows an increase of the transformed MMSE (i.e. cognitive decline) as a function of model stage, per subtype. The overall population showed a positive association with stage (0.051 points per stage, CI = [0.042, 0.06]) after covarying for years of education and age. There was not enough evidence to suggest subtype-specific trends. Statistical analysis revealed significant differences in MMSE score among subtypes (p = 7 × 10 −6 ), adjusted for model stage. The Cortical subtype demonstrated slightly higher MMSE score (mean 26, CI = [25.7, 26.3]) than the Limbic subtype (mean 25.2, CI = [25, 25.4], p = 10 −5 , Cliff’s delta = 0.28). The Cortico-Limbic subtype showed a trend towards relatively preserved cognition on the MMSE (mean 25.5, CI = [25.2, 25.7], p = 9 × 10 −6 and .0088, Cliff’s delta = 0.21 and 0.1 vs Cortical and Limbic, respectively). Download figure Open in new tab Figure 3. (A) Square root of the error in MMSE (MMSE = Mini Mental State Examination) by Subtype and Stage. (B) Domain specific cognitive performance by Subtype and Stage: Memory (i), Language (ii), Executive (iii) and Visuospatial (iv). General rends are shown in a dashed black line. Figure 3B highlights the differences in each domain-specific score across subtypes and stages. Memory ( Figure 3B (iii) ) scores demonstrated significant correlation with model stage (mean trend = -0.044, CI = [-0.051, -0.037]) after controlling for confounders, with significant but marginally slower decline rate in the Cortico-Limbic subtype (mean trend = - 0.045, CI = [-0.064, -0.27]; mean trend = -0.031, CI = [-0.044, -0.019]; mean trend = -0.042, CI = [-0.05, -0.034] for the Cortical, Cortico-Limbic and Limbic subtypes, respectively). Subtypes exhibited significant differences in memory performance (p < 10 −6 ). The Cortical subtype demonstrated superior memory performance (mean = 0.351, CI = [0.226, 0.476]) compared to both Cortico-Limbic (mean = -0.084, CI = [-0.179, 0.011]) and Limbic subtypes (mean = -0.300, CI = [-0.3691, -0.239]). All pairwise comparisons reached statistical significance, with moderate effect sizes between Cortical and Cortico-Limbic (p < 10 −6 , Cliff’s delta = 0.35), Cortical and Limbic (p < 10 −6 , Cliff’s delta = 0.51), and a small effect size between Limbic and Cortico-Limbic (p = 2 × 10 −5 , Cliff’s delta = 0.15). Executive function scores ( Figure 3B (ii) ) are significantly associated with model stage (mean trend = -0.026, CI = [-0.032, -0.02]) after confounding for other covariates, however there is not enough evidence to suggest subtype-specific trends. Nevertheless, some differences in executive function performance were observed across subtypes (p = .0016). The Cortical (mean = 0.091, CI = [-0.025, 0.207]) and Limbic subtypes (mean = 0.043, CI = [-0.018, 0.104]) exhibited superior performance compared to the Cortico-Limbic (mean = -0.078, CI = [-0.170, 0.015], p = .0021 and .0025, Cliff’s delta = 0.15 and 0.11 for the comparisons with Cortical and Limbic, respectively). Language function scores ( Figure 3B (ii) ) exhibited significant correlation with disease stage (mean trend = -0.03, CI = [-0.035, -0.024]), with no evidence of differential decline rates across subtypes. Similarly, no differences were found among subtypes (p = 0.086). Finally, no significant associations were observed between visuospatial performance ( Figure 3B (iv) ) and disease progression model stage, neither in the overall sample, nor within/between subtypes. Prevalence of DLB features across subtypes Figure 4 displays the prevalence of DLB core features in each subtype for the clinical DLB population. Parkinsonism, cognitive fluctuations, and RBD had similar prevalence across subtypes. Hallucinations showed a trend towards higher prevalence in the Cortical subtype: 56.45%, CI = [40.23, 72.66] versus Cortico-Limbic (40.54%, CI = [19.75, 61.32]) and Limbic (41.02%, CI = [20.74, 61.30]). Download figure Open in new tab Figure 4. Prevalence of DLB core features – visual hallucinations (i), REM disorder (ii), Cognitive Fluctuations (iii) and Parkinsonism (iv) – across subtypes (Cortical, Limbic, Cortico-Limbic). Sample size for each reported feature in each subtype is displayed on top. CSF biomarker profiles across stage and subtypes Figure 5A shows CSF profiles across subtypes, based on the subset of participant data available in clinical AD, MCI, and CN diagnostic populations. Both A β and p-tau were associated with model stage across all subtypes: A β decreased by –1.17% per stage on average (CI = [–1.97, –0.37]) and p-tau increased by 1.17% per stage on average (CI = [0.29, 2.043]). Evidence for subtype-specific differences in protein levels was observed (adjusted for model stage): p = 0.007 and p = 1.1 × 10 −4 for A β and p-tau residuals, respectively. Pairwise comparisons suggested that the Cortical subtype has a slightly non-AD profile: both higher levels of A β (mean = 880.2 pg/mL, CI = [756.4, 1003.9]) and lower levels of p-tau (mean = 25.7 pg/mL, CI = [21.00, 0.45]) than the other two subtypes (p = .0025, 0.0045 and 4.6 × 10 −5 , 3.1 × 10 −4 with Cliff’s delta = 0.28, 0.25 and 0.36, 0.29 for the comparisons in A β and p-tau levels of Cortical vs Limbic, Cortical vs Cortico-Limbic). However, we found no statistical evidence for a subtype difference of either protein level between the Limbic (A β mean = 745.7 pg/mL, CI = [687.5, 804.0]; p-tau mean = 32.14 pg/mL, CI = [29.48, 34.80]) and Cortico-Limbic subtypes (A β mean = 735.7 pg/mL, CI = [662.17, 809.3]; p-tau mean = 34.22 pg/mL, CI = [30.61, 37.82]): p = 0.639 and 0.186 for the A β and p-tau comparisons, respectively. Download figure Open in new tab Figure 5. Characterization of fluid and tissue-based disease markers. (A) Longitudinal trajectories of CSF amyloid beta (i) and phosphorylated tau (ii) lotted against disease stage, stratified by subtype (Cortical, Cortico-Limbic, Limbic subtypes). Individual data points are shown with linear regression lines (robust to outliers) and 99% confidence intervals for each subtype. The dashed black line represents the trend across all samples. (B) Distribution of ApoE ϵ 4 status (0, 1, 2 copies) across subtypes, showing the frequency of each status among cases with available genetic data. (C) Distribution of neuropathological diagnoses (Mixed AD/DLB, Pure AD, and Pure DLB as described in Wei et al. 2023 ) across subtypes, showing the frequency of each diagnosis among cases with available post-mortem examination data. APOE4 status across subtypes Figure 5B shows the prevalence of APOE4 across subtypes. We found evidence of an association between subtype and APOE4 count ( χ 2 = 34.39, p < 10 −6 , V = 0.15). Prevalence of APOE4 status (carriage regardless of allele count) was lower in the Cortical subtype (34.6%, CI = [24.1, 45.1]) compared to both Cortico-Limbic (61.5%, CI = [53.2, 69.7]) and Limbic (60%, CI = [53.7, 66.3]) subtypes. Neuropathology diagnosis across subtypes Figure 5C shows subtype assignment for the subset of participant data for which neuropathology was available. While we could not properly assess the classification for those with pure LB (only 10 subjects available), we found evidence of an association between the presence of AD pathology (pure AD or AD+LB) and subtype. Specifically, a preference for the Cortico-Limbic (32.5%, CI = [15.9, 49.1]) and Limbic (57.05%, CI = [43.8, 70.3]) subtypes. Discussion This work presents a novel approach to data-driven differential diagnosis, leveraging large multicenter cohorts with MRI-derived measurements and state-of-the-art clustering. Using the SuStaIn algorithm, we simultaneously accounted for both disease severity (cumulative atrophy) and subtype assignment (spatial atrophy signature) while compensating for a key methodological limitation of SuStaIn by adding a null subtype to capture discordant individuals that would otherwise add noise to the clusters. Focusing on a range of brain regions previously identified as discriminatory for AD and DLB, we found that the SuStaIn model identified distinct atrophy patterns in each dementia when modeled separately and additionally revealed clinically and biologically meaningful patterns. SuStaIn identified three predominant atrophy patterns across clinical AD and DLB populations. While the Limbic and Cortical subtypes have been previously described in AD ( Young et al., 2018 ; Chen et al., 2023 ), the Cortico-Limbic subtype is a novel finding of this work, probably facilitated by our selection of more DLB-relevant brain regions in the model. The association between clinical diagnosis and subtype assignment is noteworthy: AD with the Limbic subtype and DLB with the Cortical subtype. The Cortical subtype was also weakly associated with hallucinations – a core feature of DLB – but this did not reach statistical significance, possibly due to the heterogeneity of the clinical data which was collected across multiple sites in multiple countries and a modest sample size. Nevertheless, these spatiotemporal atrophy signatures are consistent with previous studies and may relate to the underlying pathology – our CSF results are consistent with this for AD pathology (DLB pathology not available). The Cortico-Limbic subtype seems to be a mix of features from the other two subtypes, which provides an interesting avenue for future prospective studies looking at co-pathology. Distinct cognitive profiles in each subtype further validated their clinical relevance. Memory function was relatively preserved in our DLB-like Cortical subtype (where CSF AD pathology was also lower), which aligns with autopsy-confirmed studies showing that memory is worse in the presence of AD pathology, or AD and concurrent Lewy body (LB) pathology ( Ferman et al., 2020 ; Coughlin et al., 2019 ; Brenowitz et al., 2017 ; Kraybill et al., 2005 ; Ryman et al., 2021 ). Executive function was superior in the Cortical (DLB-like) and Limbic (AD-like) subtypes compared to the Cortico-Limbic subtype, which aligns with work from Brenowitz et al. (2017) who showed that AD+LB groups fared worse on executive function than AD and LB groups. We found no statistical evidence for differences in visuospatial scores, which diverges from prospective studies in the literature ( Ferman et al., 2020 ; Ryman et al., 2021 ) but our secondary analysis was not powered to investigate this. Similarly, language function was not associated with subtype in our study, whereas Ryman et al. (2021) found LB pathology-related language dysfunction. For global cognition (i.e. MMSE), previous work has reported slower rates of decline due to LB pathology ( Ferman et al., 2020 ; Ryman et al., 2021 ; Kraybill et al., 2005 ) when compared with AD+LB (or in the case of Ryman et al. (2021) when compared with AD alone as well). We lack the longitudinal follow-up data required to assess this properly, but we did find slightly preserved MMSE score (less than one point, but statistically significant) in the DLB-like Cortical subtype compared to the AD-like Limbic subtype. Previous work ( Ferman et al., 2020 ; Coughlin et al., 2019 ) has reported no such AD vs DLB differences in mean MMSE scores. Biological validation of our MRI-based subtypes was found in CSF biomarkers and genetic data, although only AD pathology measures in CSF were available. ADNI subjects assigned to the Limbic (AD-like) subtype exhibited the characteristic AD profile of elevated p-tau and reduced A β levels, while those assigned to the Cortical (DLB-like) subtype showed the opposite. This is consistent with previous research in autopsy-confirmed studies in clinically diagnosed AD subjects later confirmed to have low levels of AD pathology ( Janelidze et al., 2020 ; Lantero Rodriguez et al., 2020 ; Irwin et al., 2018 ). APOE4 carrier status, a well-known risk factor for AD ( Verghese et al., 2011 ; Corder et al., 1994 ), was higher in the Limbic (AD-like) and Cortico-Limbic subtypes, and lower in the Cortical (DLB-like) subtype. This aligns with an autopsy confirmed study ( Tsuang et al., 2013 ) that reported higher prevalence of APOE4 in AD and AD+LB confirmed groups, and lower prevalence (albeit still 1 in 3) in the LB group. Finally, our subtypes are consistent with postmortem neuropathology. We found that AD and AD+LB neuropathology was associated with the Limbic and Cortico-Limbic subtypes. This is perhaps particularly remarkable considering the average 5-year gap between in vivo MRI scan (used for the modeling) and death. Further analysis should be carried out in other postmortem studies to confirm this finding. The large cohort, and use of multiple international datasets is a significant strength of the present study. This has allowed us to isolate disease-specific atrophy patterns and to validate them clinically, and biologically. However, our work would have been strengthened by the availability of in vivo α -synuclein biomarkers (CSF seed amplification assay, or skin biopsy). Prospective studies using these more recent α -synuclein markers, as well as frequent longitudinal sampling across patients and ideally also neuropathological confirmation closer to the time of the MRI will be necessary. This is a fundamental limitation of secondary analyses such as ours – we have combined data from (clinical) AD-enriched studies that lack DLB-specific measures, with DLB-enriched studies that lack AD-specific measures. Future prospective studies on differential diagnosis of dementias should, ideally, collect data relevant to all domains, but this is a great challenge. This work represents a novel approach to differential diagnosis of dementias based on state-of-the-art data-driven subtypes of neurodegeneration in the spectrum of Alzheimer’s disease and dementia with Lewy bodies. Leveraging the SuStaIn algorithm allowed us to avoid the limitation of classical clustering algorithms that confound the subtype (spatial atrophy) with severity (overall atrophy). Moreover, SuStaIn’s unsupervised clustering allowed us to overcome the limitation of misdiagnosis, and the TransSuStain framework made clustering possible in the presence of clearly imbalanced populations. This was facilitated by our methodological innovation of a null subgroup for catching discordant individuals (outliers) that would otherwise pollute the clusters. Subgroup assignments showed plausible clinical, biomarker, and histopathological profiles which, once validated further in prospective studies, provide encouragement for future clinical deployment in transdiagnostic applications such as memory clinics. Data Availability All data were collected with appropriate informed consent under ethically approved clinical studies. Our secondary analysis of these data was approved by the UCL Research Ethics Committee under application 8019/005 and the UCL Department of Computer Science Research Ethics Committee under application UCL-CSREC-209-B. http://adni.loni.usc.edu https://naccdata.org https://www.e-dlb.com https://pdbp.ninds.nih.gov/about Disclosures NPO is a paid consultant for Queen Square Analytics Limited (UK) on unrelated projects. EM has received speaking and writing honoraria for the International Parkinson and Movement Disorders Society, Somnomed, and CSL Seqirus. RSW has received speaking and writing honoraria from GE Healthcare, Bial, Omnix Pharma, and Britannia; and consultancy fees from Therakind and Accenture. DA participated in advisory boards from Fujirebio-Europe, Roche Diagnostics, Grifols S.A. and Lilly, and received speaker honoraria from Fujirebio-Europe, Roche Diagnostics, Nutricia, Krka Farmacéutica S.L., Zambon S.A.U., Neuraxpharm, Alter Medica, Lilly and Esteve Pharmaceuticals S.A. DA declares a filed patent application (WO2019175379 Markers of synaptopathy in neurodegenerative disease). DF consults for BioArctic and has received honoraria from Esteve Pharmaceuticals S.A. JPT has received speaking honoraria from GE Healthcare and acted as a consultant for CervoMed and Eisai. ZW acted as a consultant for GE Healthcare. Acknowledgements This work was made possible by an Ignition grant from the University of Sydney and University College London (Global Engagement Fund), and a travel grant from Alzheimer’s Research UK. GCL and NPO acknowledge funding from the UKRI Medical Research Council (MR/S03546X/1, MR/X024288/1, MR/T046422/1). EM was supported by a National Health and Medical Research Council (NHMRC) grant (2008565). JPT is supported by the NIHR Newcastle Biomedical Research Centre. RSW is supported by a Wellcome Career Development Award (#225263/Z/22/Z), Parkinson’s UK, the Lewy Body Society, Michael J Fox Foundation, and by the National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre CL was supported by the Medical Research Council (MR/R006504/1), Parkinson’s UK, Hilary-Galen Weston Foundation and Michael J Fox Foundation. SJGL was supported by National Health and Medical Research Council fellowship grant (#1195830) and the NHMRC/EU Joint Programme on Neurodegenerative Disease Research (#2014513). DA was supported by research grants from Institute of Health Carlos III (ISCIII), Spain, PI18/00435, PI22/00611, INT19/00016, INT23/00048 to DA, and by the Department of Health Generalitat de Catalunya PERIS program SLT006/17/125. AB acknowledges support from Instituto de Salud Carlos III and co-funded by the European Union through the Miguel Servet grant (CP20/00038) and Fondo de Investigaciones Sanitario (PI22/00307), the Alzheimer’s Association (AARG-22-923680), and the Ajuntament de Barcelona, in collaboration with Fundació La Caixa (23S06157-001). EW was supported by the Swedish Research Council (VR) No. 2016-02282, 2021-01861; the Center for Innovative Medicine (CIMED) No. FoUI-954459, FoUI-975174; the regional agreement on medical training and clinical research (ALF) between Stockholm County Council and Karolinska Institutet No. FoUI-952838, FoUI-954893; The Swedish Brain Foundation (Hjärnfonden) No. FO2022-0084, FO2024-0239; The Swedish Alzheimer’s Foundation (Alzheimerfonden) No. AF-967495, AF-980387; The Swedish Parkinson’s foundation (Parkinsonfonden) No. 1557/24, 1521/23; EU Innovative Health Initiative Joint Undertaking (IHI JU) AD-RIDDLE; King Gustaf V:s and Queen Victorias Foundation; Olle Engkvists Foundation (Olle Engkvists Stiftelse) No. 186-0660, 224-0069. AH was supported by funding from StratNeuro, Demensfonden, Gun and Bertil Stohnes Stiftelsen (2024-029), Stiftelsen för Gamla Tjänarinnor (2023-016), and various grants from Karolinska Institutet (2024-02083). DF receives funding from the Swedish Research Council (Vetenskapsrådet, grant 2022-00916), the Center for Innovative Medicine (CIMED, grants 20200505 and FoUI-988826), the regional agreement on medical training and clinical research of Stockholm Region (ALF Medicine, grants FoUI-962240 and FoUI-987534), the Swedish Brain Foundation (Hjärnfonden FO2023-0261, FO2022-0175, FO2021-0131), the Swedish Alzheimer Foundation (Alzheimerfonden AF-968032, AF-980580, AF-994058, AF-1010553), the Swedish Dementia Foundation (Demensfonden), the Gamla Tjänarinnor Foundation, the Gun och Bertil Stohnes Foundation, the Åke Wiberg Foundation, StratNeuro, Parkinsonfonden, Lindhes, Funding for Research from Karolinska Institutet, Neurofonden, and the Foundation for Geriatric Diseases at Karolinska Institutet, contributions from private bequests and academic agreements with industry. ZW received funding from ARUK and Lewy Body Society. This research was funded in whole, or in part by the Wellcome Trust [227341/Z/23/Z]. For the purpose of open access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. Appendix Previous studies have shown that there are substantial ‘batch effects’ present in volumetric measurements derived from structural MRI scans, possibly confounding the underlying disease signal ( Bayer et al., 2022 ). We explore these effects on a subset of subjects where original scans were available, and use the MRIQC ( Esteban et al., 2017 ) and NeuroHarmony ( Garcia-Dias et al., 2020 ) tools to harmonize the data. We define the ‘batch effects’ as scanner manufacturer, model and field strength – all previously described to affect image quality and thus any measurement derived from the scan ( Jovicich et al., 2009 ). We investigate the effect of scan manufacturer, model and field strength on the biomarkers that the SuStaIn models are trained on. This combination of ‘batch effects’ results in a total of 34 groups (detailed in Table S2). In this experiment, data were z-scored and then the NeuroHarmony model was trained on the groups that contain at least 20 samples and later applied to everyone. To understand the ‘batch effects’, we use Linear Discriminant Analysis (LDA) for visualization and the LazyClassifier python package to explore a wide range of classifiers – the task being the correct assignment of the batch based on the set of biomarkers — and we selected the top performers at any point and compared them. The LDA results show that there is in fact some ‘batch effect’ on the signal in some of the FreeSurfer outputs. Harmonization reduces the ‘batch effects’ (while preserving meaningful biological signal) as seen by the reduction in the clustering separation. Nevertheless, the biomarkers selected as model inputs show no cluster separation, highlighting that LDA uses other metrics to discriminate between groups ( Figure S1 ). Additionally, the classification performance dropped after harmonization when compared to the non-harmonized data, but using the model-selected biomarkers also caused a drop in performance in the non-harmonized scenario, supporting the idea that these biomarkers retain little ‘batch effects’ ( Figure S2 ). The results using SuStaIn show little difference. In both scenarios – harmonized and not harmonized data – the model converges to the 2-subtype solution, with very similar subtypes found, as shown by both the maximum likelihood sequences and the positional variance diagrams ( Figure S3 ). Hellinger distance analysis shows strong similarity between harmonized and not harmonized pairs (0.137 and 0.078 for the Cortico-Limbic and Limbic subtypes). These results are in line with previous studies where the use of harmonization techniques did not significantly change the model outputs ( Chen et al., 2023 ). Download figure Open in new tab Figure S1. Linear discriminant analysis of neuroimaging data by manufacturer and batch (only top 6 with most subjects). Harmonized (left) and non-harmonized (right) data are shown with either all biomarkers or only model-selected biomarkers. Points are colored by scanner manufacturer (top panels: GE; Philips; Siemens) or specific ‘batch’ (bottom panels). View this table: View inline View popup Download powerpoint Table S1. Demographics of subjects by Dataset. CU = Cognitively Unimpaired; MCI = Mild Cognitive Impairment; Dem = Dementia; STD = Standard Deviation; MMSE = Mini Mental State Examination; MoCA = Montreal Cognitive Assessment; AD = Alzheimer’s disease; LBD = Lewy body disease; DLB = Dementia with Lewy bodies. View this table: View inline View popup Download powerpoint Table S2. Batch effects (as manufacturer, model and field strength) groups and counts of samples. Download figure Open in new tab Figure S2. Classification accuracy using neuroimaging data of manufacturer and batch (only top 6 with most subjects). Harmonized (left) and non-harmonized (right) results are shown with either all biomarkers or only model-selected biomarkers. Boxplots are colored by classification method (Gaussian Naive Bayes, blue; LDA, orange; Logistic Regression, green). Download figure Open in new tab Figure S3. SuStaIn model results for AD using Harmonized and Not Harmonized data, showing subtype progression patterns and model validation. The figure is organized in four rows representing distinct subtypes: (i) Cortico-Limbic Harmonized, (ii) Limbic Harmonized, and (iii)Cortico-Limbic Not Harmonized, (iv) Limbic Not Harmonized. Each row contains three panels, from left to right: (A) Maximum Likelihood Sequence showing the temporal evolution of regional atrophy (x-axis: stages 1-33; y-axis: brain regions), where colors indicate severity thresholds (red: z-score=0.67, magenta: z-score=1, blue: z-score=2); (B) Positional Variance Diagram displaying the uncertainty in the ordering of events, with intensity indicating probability of event (same thresholds as before). Bibliography ↵ Alexander , N. , Alexander , D. C. , Barkhof , F. , and Denaxas , S. ( 2021 ). Identifying and evaluating clinical subtypes of alzheimer’s disease in care electronic health records using unsupervised machine learning . BMC Medical Informatics and Decision Making , 21 ( 1 ): 343 . OpenUrl ↵ Barker , W. W. , Luis , C. A. , Kashuba , A. , Luis , M. , Harwood , D. G. , Loewenstein , D. , Waters , C. , Jimison , P. , Shepherd , E. , Sevush , S. , et al. ( 2002 ). Relative frequencies of alzheimer disease, lewy body, vascular and frontotemporal dementia, and hippocampal sclerosis in the state of florida brain bank . Alzheimer Disease & Associated Disorders , 16 ( 4 ): 203 – 212 . OpenUrl CrossRef PubMed Web of Science ↵ Bayer , J. M. , Thompson , P. M. , Ching , C. R. , Liu , M. , Chen , A. , Panzenhagen , A. C. , Jahanshad , N. , Marquand , A. , Schmaal , L. , and Sämann , P. G. ( 2022 ). Site effects how-to and when: An overview of retrospective techniques to accommodate site effects in multi-site neuroimaging analyses . Frontiers in neurology , 13 : 923988 . OpenUrl PubMed ↵ Blanc , F. , Colloby , S. J. , Cretin , B. , de Sousa , P. L. , Demuynck , C. , O’Brien , J. T. , Martin-Hunyadi , C. , McKeith , I. , Philippi , N. , and Taylor , J.-P. ( 2016 ). Grey matter atrophy in prodromal stage of dementia with lewy bodies and alzheimer’s disease . Alzheimer’s research & therapy , 8 ( 1 ): 31 . OpenUrl ↵ Blanc , F. , Bouteloup , V. , Paquet , C. , Chupin , M. , Pasquier , F. , Gabelle , A. , Ceccaldi , M. , de Sousa , P. L. , Krolak-Salmon , P. , David , R. , et al. ( 2022 ). Prodromal characteristics of dementia with lewy bodies: baseline results of the memento memory clinics nationwide cohort . Alzheimer’s research & therapy , 14 ( 1 ): 96 . OpenUrl ↵ Braak , H. and Braak , E. ( 1991 ). Neuropathological stageing of alzheimer-related changes . Acta Neuropathologica , 82 ( 4 ): 239 – 259 . doi: 10.1007/BF00308809 . OpenUrl CrossRef PubMed Web of Science ↵ Braak , H. , Alafuzoff , I. , Arzberger , T. , Kretzschmar , H. , and Del Tredici , K. ( 2006 ). Staging of alzheimer disease-associated neurofibrillary pathology using paraffin sections and immunocytochemistry . Acta neuropathologica , 112 ( 4 ): 389 – 404 . doi: 10.1007/s00401-006-0127-z . OpenUrl CrossRef PubMed Web of Science ↵ Brenowitz , W. D. , Hubbard , R. A. , Keene , C. D. , Hawes , S. E. , Longstreth Jr , W. , Woltjer , R. L. , and Kukull , W. A. ( 2017 ). Mixed neuropathologies and associations with domain-specific cognitive decline . Neurology , 89 ( 17 ): 1773 – 1781 . OpenUrl CrossRef PubMed ↵ Chen , H. , Young , A. , Oxtoby , N. P. , Barkhof , F. , Alexander , D. C. , Altmann , A. , et al. ( 2023 ). Transferability of alzheimer’s disease progression subtypes to an independent population cohort . NeuroImage , 271 : 120005 . OpenUrl PubMed ↵ Chouliaras , L. , Thomas , A. , Malpetti , M. , Donaghy , P. , Kane , J. , Mak , E. , Savulich , G. , Prats-Sedano , M. A. , Heslegrave , A. J. , Zetterberg , H. , et al. ( 2022 ). Differential levels of plasma biomarkers of neurodegeneration in lewy body dementia, alzheimer’s disease, frontotemporal dementia and progressive supranuclear palsy . Journal of Neurology, Neurosurgery & Psychiatry , 93 ( 6 ): 651 – 658 . OpenUrl Abstract / FREE Full Text ↵ Corder , E. , Saunders , A. M. , Risch , N. , Strittmatter , W. , Schmechel , D. , Gaskell Jr , P. , Rimmler , J. , Locke , P. , Conneally , P. , Schmader , K. , et al. ( 1994 ). Protective effect of apolipoprotein e type 2 allele for late onset alzheimer disease . Nature genetics , 7 ( 2 ): 180 – 184 . OpenUrl CrossRef PubMed Web of Science ↵ Coughlin , D. , Xie , S. X. , Liang , M. , Williams , A. , Peterson , C. , Weintraub , D. , McMillan , C. T. , Wolk , D. A. , Akhtar , R. S. , Hurtig , H. I. , et al. ( 2019 ). Cognitive and pathological influences of tau pathology in lewy body disorders . Annals of neurology , 85 ( 2 ): 259 – 271 . OpenUrl CrossRef PubMed ↵ Desikan , R. S. , Ségonne , F. , Fischl , B. , Quinn , B. T. , Dickerson , B. C. , Blacker , D. , Buckner , R. L. , Dale , A. M. , Maguire , R. P. , Hyman , B. T. , et al. ( 2006 ). An automated labeling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest . Neuroimage , 31 ( 3 ): 968 – 980 . OpenUrl CrossRef PubMed Web of Science ↵ Dong , A. , Toledo , J. B. , Honnorat , N. , Doshi , J. , Varol , E. , Sotiras , A. , Wolk , D. , Trojanowski , J. Q. , Davatzikos , C. , and Initiative , A. D. N. ( 2017 ). Heterogeneity of neuroanatomical patterns in prodromal alzheimer’s disease: links to cognition, progression and biomarkers . Brain , 140 ( 3 ): 735 – 747 . OpenUrl CrossRef PubMed ↵ Esteban , O. , Birman , D. , Schaer , M. , Koyejo , O. O. , Poldrack , R. A. , and Gorgolewski , K. J. ( 2017 ). Mriqc: Advancing the automatic prediction of image quality in mri from unseen sites . PloS one , 12 ( 9 ): e0184661 . OpenUrl CrossRef PubMed ↵ Ferman , T. J. , Aoki , N. , Boeve , B. F. , Aakre , J. A. , Kantarci , K. , Graff-Radford , J. , Parisi , J. E. , Van Gerpen , J. A. , Graff-Radford , N. R. , Uitti , R. J. , et al. ( 2020 ). Subtypes of dementia with lewy bodies are associated with α-synuclein and tau distribution . Neurology , 95 ( 2 ): e155 – e165 . OpenUrl CrossRef PubMed ↵ Ferreira , D. , Nordberg , A. , and Westman , E. ( 2020 ). Biological subtypes of alzheimer disease: a systematic review and meta-analysis . Neurology , 94 ( 10 ): 436 – 448 . OpenUrl CrossRef PubMed ↵ Fischl , B. , Van Der Kouwe , A. , Destrieux , C. , Halgren , E. , Ségonne , F. , Salat , D. H. , Busa , E. , Seidman , L. J. , Goldstein , J. , Kennedy , D. , et al. ( 2004 ). Automatically parcellating the human cerebral cortex . Cerebral cortex , 14 ( 1 ): 11 – 22 . OpenUrl CrossRef PubMed Web of Science ↵ Fonteijn , H. M. , Modat , M. , Clarkson , M. J. , Barnes , J. , Lehmann , M. , Hobbs , N. Z. , Scahill , R. I. , Tabrizi , S. J. , Ourselin , S. , Fox , N. C. , et al. ( 2012 ). An event-based model for disease progression and its application in familial alzheimer’s disease and huntington’s disease . NeuroImage , 60 ( 3 ): 1880 – 1889 . OpenUrl CrossRef PubMed Web of Science ↵ Fotenos , A. F. , Mintun , M. A. , Snyder , A. Z. , Morris , J. C. , and Buckner , R. L. ( 2008 ). Brain volume decline in aging: evidence for a relation between socioeconomic status, preclinical alzheimer disease, and reserve . Archives of neurology , 65 ( 1 ): 113 – 120 . OpenUrl CrossRef PubMed Web of Science ↵ Franzmeier , N. , Roemer-Cassiano , S. N. , Bernhardt , A. M. , Dehsarvi , A. , Dewenter , A. , Steward , A. , Biel , D. , Frontzkowski , L. , Zhu , Z. , Gnörich , J. , et al. ( 2025 ). Alpha synuclein co-pathology is associated with accelerated amyloid-driven tau accumulation in alzheimer’s disease . Molecular neurodegeneration , 20 ( 1 ): 31 . OpenUrl PubMed ↵ Garcia-Dias , R. , Scarpazza , C. , Baecker , L. , Vieira , S. , Pinaya , W. H. , Corvin , A. , Redolfi , A. , Nelson , B. , Crespo-Facorro , B. , McDonald , C. , et al. ( 2020 ). Neuroharmony: A new tool for harmonizing volumetric mri data from unseen scanners . Neuroimage , 220 : 117127 . OpenUrl CrossRef PubMed ↵ Gonzalez , M. C. , Ashton , N. J. , Gomes , B. F. , Tovar-Rios , D. A. , Blanc , F. , Karikari , T. K. , Mollenhauer , B. , Pilotto , A. , Lemstra , A. , Paquet , C. , et al. ( 2022 ). Association of plasma p-tau181 and p-tau231 concentrations with cognitive decline in patients with probable dementia with lewy bodies . JAMA neurology , 79 ( 1 ): 32 – 37 . OpenUrl PubMed ↵ Gur , R. C. , Mozley , P. D. , Resnick , S. M. , Gottlieb , G. L. , Kohn , M. , Zimmerman , R. , Herman , G. , Atlas , S. , Grossman , R. , and Berretta , D. ( 1991 ). Gender differences in age effect on brain atrophy measured by magnetic resonance imaging . Proceedings of the National Academy of Sciences , 88 ( 7 ): 2845 – 2849 . OpenUrl Abstract / FREE Full Text ↵ Hwang , J. , Kim , C. M. , Jeon , S. , Lee , J. M. , Hong , Y. J. , Roh , J. H. , Lee , J.-H. , Koh , J.-Y. , Na , D. L. , Initiative , A. D. N. , et al. ( 2016 ). Prediction of alzheimer’s disease pathophysiology based on cortical thickness patterns . Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring , 2 : 58 – 67 . OpenUrl ↵ Inguanzo , A. , Poulakis , K. , Mohanty , R. , Schwarz , C. G. , Przybelski , S. A. , Diaz-Galvan , P. , Lowe , V. J. , Boeve , B. F. , Lemstra , A. W. , van De Beek , M. , et al. ( 2023 ). Mri data-driven clustering reveals different subtypes of dementia with lewy bodies . npj Parkinson’s Disease , 9 ( 1 ): 5 . OpenUrl ↵ Irwin , D. J. , Xie , S. X. , Coughlin , D. , Nevler , N. , Akhtar , R. S. , McMillan , C. T. , Lee , E. B. , Wolk , D. A. , Weintraub , D. , Chen-Plotkin , A. , et al. ( 2018 ). Csf tau and β-amyloid predict cerebral synucleinopathy in autopsied lewy body disorders . Neurology , 90 ( 12 ): e1038 – e1046 . OpenUrl CrossRef PubMed ↵ Jacqmin-Gadda , H. , Fabrigoule , C. , Commenges , D. , and Dartigues , J.-F. ( 1997 ). A 5-year longitudinal study of the mini-mental state examination in normal aging . American journal of epidemiology , 145 ( 6 ): 498 – 506 . OpenUrl CrossRef PubMed Web of Science ↵ Janelidze , S. , Mattsson , N. , Palmqvist , S. , Smith , R. , Beach , T. G. , Serrano , G. E. , Chai , X. , Proctor , N. K. , Eichenlaub , U. , Zetterberg , H. , et al. ( 2020 ). Plasma p-tau181 in alzheimer’s disease: relationship to other biomarkers, differential diagnosis, neuropathology and longitudinal progression to alzheimer’s dementia . Nature medicine , 26 ( 3 ): 379 – 386 . OpenUrl CrossRef PubMed ↵ Jovicich , J. , Czanner , S. , Han , X. , Salat , D. , van der Kouwe , A. , Quinn , B. , Pacheco , J. , Albert , M. , Killiany , R. , Blacker , D. , et al. ( 2009 ). Mri-derived measurements of human subcortical, ventricular and intracranial brain volumes: reliability effects of scan sessions, acquisition sequences, data analyses, scanner upgrade, scanner vendors and field strengths . Neuroimage , 46 ( 1 ): 177 – 192 . OpenUrl CrossRef PubMed Web of Science ↵ Kantarci , K. , Lesnick , T. , Ferman , T. J. , Przybelski , S. A. , Boeve , B. F. , Smith , G. E. , Kremers , W. K. , Knopman , D. S. , Jack Jr , C. R. , and Petersen , R. C. ( 2016 ). Hippocampal volumes predict risk of dementia with lewy bodies in mild cognitive impairment . Neurology , 87 ( 22 ): 2317 – 2323 . OpenUrl PubMed ↵ Khadhraoui , E. , Müller , S. J. , Hansen , N. , Riedel , C. H. , Langer , P. , Timäeus , C. , Wiltfang , J. , Bouter , C. , Lange , C. , and Ernst , M. ( 2022 ). Manual and automated analysis of atrophy patterns in dementia with lewy bodies on mri . BMC neurology , 22 ( 1 ): 114 . OpenUrl PubMed ↵ Kraybill , M. L. , Larson , E. B. , Tsuang , D. , Teri , L. , McCormick , W. , Bowen , J. , Kukull , W. , Leverenz , J. , and Cherrier , M. ( 2005 ). Cognitive differences in dementia patients with autopsy-verified ad, lewy body pathology, or both . Neurology , 64 ( 12 ): 2069 – 2073 . OpenUrl CrossRef PubMed ↵ Lantero Rodriguez , J. , Karikari , T. K. , Suárez-Calvet , M. , Troakes , C. , King , A. , Emersic , A. , Aarsland , D. , Hye , A. , Zetterberg , H. , Blennow , K. , et al. ( 2020 ). Plasma p-tau181 accurately predicts alzheimer’s disease pathology at least 8 years prior to post-mortem and improves the clinical characterisation of cognitive decline . Acta neuropathologica , 140 ( 3 ): 267 – 278 . OpenUrl CrossRef PubMed ↵ Malek-Ahmadi , M. , Beach , T. G. , Zamrini , E. , Adler , C. H. , Sabbagh , M. N. , Shill , H. A. , Jacobson , S. A. , Belden , C. M. , Caselli , R. J. , Woodruff , B. K. , et al. ( 2019 ). Faster cognitive decline in dementia due to alzheimer disease with clinically undiagnosed lewy body disease . PloS one , 14 ( 6 ): e0217566 . OpenUrl CrossRef PubMed ↵ Mastenbroek , S. E. , Vogel , J. W. , Collij , L. E. , Serrano , G. E. , Tremblay , C. , Young , A. L. , Arce , R. A. , Shill , H. A. , Driver-Dunckley , E. D. , Mehta , S. H. , et al. ( 2024 ). Disease progression modelling reveals heterogeneity in trajectories of lewy-type α-synuclein pathology . Nature communications , 15 ( 1 ): 5133 . OpenUrl PubMed ↵ Masters , C. L. , Simms , G. , Weinman , N. A. , Multhaup , G. , McDonald , B. L. , and Beyreuther , K. ( 1985 ). Amyloid plaque core protein in alzheimer disease and down syndrome . Proceedings of the National Academy of Sciences , 82 ( 12 ): 4245 – 4249 . doi: 10.1073/pnas.82.12.4245 . OpenUrl Abstract / FREE Full Text ↵ McKeith , I. G. , Dickson , D. W. , Lowe , J. , Emre , M. , O’brien , J. , Feldman , H. , Cummings , J. , Duda , J. , Lippa , C. , Perry , E. , et al. ( 2005 ). Diagnosis and management of dementia with lewy bodies: third report of the dlb consortium . Neurology , 65 ( 12 ): 1863 – 1872 . OpenUrl CrossRef PubMed ↵ McKeith , I. G. , Boeve , B. F. , Dickson , D. W. , Halliday , G. , Taylor , J.-P. , Weintraub , D. , Aarsland , D. , Galvin , J. , Attems , J. , Ballard , C. G. , et al. ( 2017 ). Diagnosis and management of dementia with lewy bodies: Fourth consensus report of the dlb consortium . Neurology , 89 ( 1 ): 88 – 100 . OpenUrl CrossRef PubMed ↵ Mukherjee , S. , Choi , S.-E. , Lee , M. L. , Scollard , P. , Trittschuh , E. H. , Mez , J. , Saykin , A. J. , Gibbons , L. E. , Sanders , R. E. , Zaman , A. F. , et al. ( 2023 ). Cognitive domain harmonization and cocalibration in studies of older adults . Neuropsychology , 37 ( 4 ): 409 . OpenUrl CrossRef PubMed ↵ Nelson , P. T. , Jicha , G. A. , Kryscio , R. J. , Abner , E. L. , Schmitt , F. A. , Cooper , G. , Xu , L. O. , Smith , C. D. , and Markesbery , W. R. ( 2010 ). Low sensitivity in clinical diagnoses of dementia with lewy bodies . Journal of neurology , 257 ( 3 ): 359 – 366 . OpenUrl CrossRef PubMed Web of Science ↵ Noh , Y. , Jeon , S. , Lee , J. M. , Seo , S. W. , Kim , G. H. , Cho , H. , Ye , B. S. , Yoon , C. W. , Kim , H. J. , Chin , J. , et al. ( 2014 ). Anatomical heterogeneity of alzheimer disease: based on cortical thickness on mris . Neurology , 83 ( 21 ): 1936 – 1944 . OpenUrl CrossRef PubMed ↵ Oxtoby , N. P. , Leyland , L.-A. , Aksman , L. M. , Thomas , G. E. , Bunting , E. L. , Wijeratne , P. A. , Young , A. L. , Zarkali , A. , Tan , M. M. , Bremner , F. D. , et al. ( 2021 ). Sequence of clinical and neurodegeneration events in parkinson’s disease progression . Brain , 144 ( 3 ): 975 – 988 . OpenUrl CrossRef PubMed ↵ Park , J.-Y. , Na , H. K. , Kim , S. , Kim , H. , Kim , H. J. , Seo , S. W. , Na , D. L. , Han , C. E. , and Seong , J.-K. ( 2017 ). Robust identification of alzheimer’s disease subtypes based on cortical atrophy patterns . Scientific reports , 7 ( 1 ): 43270 . OpenUrl PubMed ↵ Plastini , M. J. , Abdelnour , C. , Young , C. B. , Wilson , E. N. , Shahid-Besanti , M. , Lamoureux , J. , Andreasson , K. I. , Kerchner , G. A. , Montine , T. J. , Henderson , V. W. , et al. ( 2024 ). Multiple biomarkers improve diagnostic accuracy across lewy body and alzheimer’s disease spectra . Annals of Clinical and Translational Neurology , 11 ( 5 ): 1197 – 1210 . OpenUrl ↵ Poulakis , K. , Pereira , J. B. , Mecocci , P. , Vellas , B. , Tsolaki , M. , Kłoszewska , I. , Soininen , H. , Lovestone , S. , Simmons , A. , Wahlund , L.-O. , et al. ( 2018 ). Heterogeneous patterns of brain atrophy in alzheimer’s disease . Neurobiology of aging , 65 : 98 – 108 . OpenUrl CrossRef PubMed ↵ Poulakis , K. , Pereira , J. B. , Muehlboeck , J.-S. , Wahlund , L.-O. , Smedby , Ö. , Volpe , G. , Masters , C. L. , Ames , D. , Niimi , Y. , Iwatsubo , T. , et al. ( 2022 ). Multi-cohort and longitudinal bayesian clustering study of stage and subtype in alzheimer’s disease . Nature communications , 13 ( 1 ): 4566 . OpenUrl PubMed ↵ Rizzo , G. , Arcuti , S. , Copetti , M. , Alessandria , M. , Savica , R. , Fontana , A. , Liguori , R. , and Logroscino , G. ( 2018 ). Accuracy of clinical diagnosis of dementia with lewy bodies: a systematic review and meta-analysis . Journal of Neurology, Neurosurgery & Psychiatry , 89 ( 4 ): 358 – 366 . OpenUrl Abstract / FREE Full Text ↵ Roquet , D. , Noblet , V. , Anthony , P. , Philippi , N. , Demuynck , C. , Cretin , B. , Martin-Hunyadi , C. , Loureiro de Sousa , P. , and Blanc , F. ( 2017 ). Insular atrophy at the prodromal stage of dementia with lewy bodies: a vbm dartel study . Scientific reports , 7 ( 1 ): 9437 . OpenUrl PubMed ↵ Ryman , S. G. , Yutsis , M. , Tian , L. , Henderson , V. W. , Montine , T. J. , Salmon , D. P. , Galasko , D. , and Poston , K. L. ( 2021 ). Cognition at each stage of lewy body disease with co-occurring alzheimer’s disease pathology . Journal of Alzheimer’s Disease , 80 ( 3 ): 1243 – 1256 . OpenUrl ↵ Spillantini , M. G. , Schmidt , M. L. , Lee , V. M.-Y. , Trojanowski , J. Q. , Jakes , R. , and Goedert , M. ( 1997 ). α-synuclein in lewy bodies . Nature , 388 ( 6645 ): 839 – 840 . OpenUrl CrossRef PubMed Web of Science ↵ Tan , J. H. , Laurell , A. A. , Sidhom , E. , Rowe , J. B. , and O’Brien , J. T. ( 2025 ). The effect of amyloid and tau co-pathology on disease progression in lewy body dementia: A systematic review . Parkinsonism & Related Disorders , 131 : 107255 . OpenUrl PubMed ↵ Thal , D. R. , Rüb , U. , Orantes , M. , and Braak , H. ( 2002 ). Phases of aβ-deposition in the human brain and its relevance for the development of ad . Neurology , 58 ( 12 ): 1791 – 1800 . doi: 10.1212/WNL.58.12.1791 . OpenUrl CrossRef PubMed ↵ Tsuang , D. , Leverenz , J. B. , Lopez , O. L. , Hamilton , R. L. , Bennett , D. A. , Schneider , J. A. , Buchman , A. S. , Larson , E. B. , Crane , P. K. , Kaye , J. A. , et al. ( 2013 ). Apoe 4 increases risk for dementia in pure synucleinopathies . JAMA neurology , 70 ( 2 ): 223 – 228 . OpenUrl PubMed ↵ Varol , E. , Sotiras , A. , Davatzikos , C. , Initiative , A. D. N. , et al. ( 2017 ). Hydra: Revealing heterogeneity of imaging and genetic patterns through a multiple max-margin discriminative analysis framework . Neuroimage , 145 : 346 – 364 . OpenUrl CrossRef PubMed ↵ Verdi , S. , Marquand , A. F. , Schott , J. M. , and Cole , J. H. ( 2021 ). Beyond the average patient: how neuroimaging models can address heterogeneity in dementia . Brain , 144 ( 10 ): 2946 – 2953 . OpenUrl CrossRef PubMed ↵ Verghese , P. B. , Castellano , J. M. , and Holtzman , D. M. ( 2011 ). Apolipoprotein e in alzheimer’s disease and other neurological disorders . The Lancet Neurology , 10 ( 3 ): 241 – 252 . OpenUrl PubMed ↵ Voevodskaya , O. , Simmons , A. , Nordenskjöld , R. , Kullberg , J. , Ahlström , H. , Lind , L. , Wahlund , L.-O. , Larsson , E.-M. , Westman , E. , and Initiative , A. D. N. ( 2014 ). The effects of intracranial volume adjustment approaches on multiple regional mri volumes in healthy aging and alzheimer’s disease . Frontiers in aging neuroscience , 6 : 264 . OpenUrl PubMed ↵ Vogel , J. W. , Young , A. L. , Oxtoby , N. P. , Smith , R. , Ossenkoppele , R. , Strandberg , O. T. , La Joie , R. , Aksman , L. M. , Grothe , M. J. , Iturria-Medina , Y. , et al. ( 2021 ). Four distinct trajectories of tau deposition identified in alzheimer’s disease . Nature medicine , 27 ( 5 ): 871 – 881 . OpenUrl CrossRef PubMed ↵ Walker , Z. , Possin , K. L. , Boeve , B. F. , and Aarsland , D. ( 2015 ). Lewy body dementias . The Lancet , 386 ( 10004 ): 1683 – 1697 . OpenUrl ↵ Wei , H. , Masurkar , A. V. , and Razavian , N. ( 2023 ). On gaps of clinical diagnosis of dementia subtypes: A study of alzheimer’s disease and lewy body disease . Frontiers in Aging Neuroscience , 15 : 1149036 . OpenUrl PubMed ↵ Young , A. , Oxtoby , N. , Garbarino , S. , Fox , N. , Barkhof , F. , Schott , J. , and Alexander , D. ( 2024 ). Data-driven modelling of neurodegenerative disease progression: thinking outside the black box . Nature Reviews Neuroscience , 25 ( 2 ): 111 – 130 . doi: 10.1038/s41583-023-00779-6 . OpenUrl CrossRef PubMed ↵ Young , A. L. , Marinescu , R. V. , Oxtoby , N. P. , Bocchetta , M. , Yong , K. , Firth , N. C. , Cash , D. M. , Thomas , D. L. , Dick , K. M. , Cardoso , J. , et al. ( 2018 ). Uncovering the heterogeneity and temporal complexity of neurodegenerative diseases with subtype and stage inference . Nature communications , 9 ( 1 ): 4273 . OpenUrl PubMed Young , A. L. , Vogel , J. W. , Aksman , L. M. , Wijeratne , P. A. , Eshaghi , A. , Oxtoby , N. P. , Williams , S. C. , Alexander , D. C. , and Initiative , A. D. N. ( 2021 ). Ordinal sustain: subtype and stage inference for clinical scores, visual ratings, and other ordinal data . Frontiers in artificial intelligence , 4 : 613261 . OpenUrl PubMed ↵ Young , A. L. , Vogel , J. W. , Robinson , J. L. , McMillan , C. T. , Ossenkoppele , R. , Wolk , D. A. , Irwin , D. J. , Elman , L. , Grossman , M. , Lee , V. M. , et al. ( 2023 ). Data-driven neuropathological staging and subtyping of tdp-43 proteinopathies . Brain , 146 ( 7 ): 2975 – 2988 . OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted October 08, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Differential diagnosis of dementias using in vivo MRI and data-driven disease progression modelling: a case study in Alzheimer’s disease and dementia with Lewy bodies Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Differential diagnosis of dementias using in vivo MRI and data-driven disease progression modelling: a case study in Alzheimer’s disease and dementia with Lewy bodies Gonzalo Castro Leal , Ajay Konuri , Alexandra L. Young , Niloufar Zebarjadi , Annegret Habich , Nicolás Castellanos-Perilla , María Camila Gonzalez , John-Paul Taylor , Michael Firbank , Simon J.G. Lewis , Daniel Alcolea , Alexandre Bejanin , Kurt Segers , Ahmet Turan Isik , Bedia Samanci , Consuelo Cháfer-Pericás , Christian Lambert , Ramón Landin-Romero , Rohan Bhome , Ivelina Dobreva , Rimona S. Weil , Zuzana Walker , Dag Aarsland , Eric Westman , Daniel Ferreira , Elie Matar , Neil P. Oxtoby medRxiv 2025.10.03.25337171; doi: https://doi.org/10.1101/2025.10.03.25337171 Share This Article: Copy Citation Tools Differential diagnosis of dementias using in vivo MRI and data-driven disease progression modelling: a case study in Alzheimer’s disease and dementia with Lewy bodies Gonzalo Castro Leal , Ajay Konuri , Alexandra L. Young , Niloufar Zebarjadi , Annegret Habich , Nicolás Castellanos-Perilla , María Camila Gonzalez , John-Paul Taylor , Michael Firbank , Simon J.G. Lewis , Daniel Alcolea , Alexandre Bejanin , Kurt Segers , Ahmet Turan Isik , Bedia Samanci , Consuelo Cháfer-Pericás , Christian Lambert , Ramón Landin-Romero , Rohan Bhome , Ivelina Dobreva , Rimona S. Weil , Zuzana Walker , Dag Aarsland , Eric Westman , Daniel Ferreira , Elie Matar , Neil P. Oxtoby medRxiv 2025.10.03.25337171; doi: https://doi.org/10.1101/2025.10.03.25337171 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Neurology Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4436) Dentistry and Oral Medicine (444) Dermatology (382) Emergency Medicine (608) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1509) Epidemiology (15229) Forensic Medicine (30) Gastroenterology (1124) Genetic and Genomic Medicine (6600) Geriatric Medicine (668) Health Economics (997) Health Informatics (4538) Health Policy (1368) Health Systems and Quality Improvement (1613) Hematology (542) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15916) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (146) Nephrology (667) Neurology (6599) Nursing (346) Nutrition (998) Obstetrics and Gynecology (1144) Occupational and Environmental Health (957) Oncology (3333) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (663) Pediatrics (1693) Pharmacology and Therapeutics (691) Primary Care Research (711) Psychiatry and Clinical Psychology (5447) Public and Global Health (9232) Radiology and Imaging (2198) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (593) Sexual and Reproductive Health (712) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a00f4144a940300f',t:'MTc3OTY1NjY4OQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();