You are here

  • Home
  • Archive
  • Volume 90, Issue 11
  • A multimodal MRI-based classification signature emerges just prior to symptom onset in frontotemporal dementia mutation carriers

Article Text

Article menu

Download PDFPDF

Neurodegeneration
Research paper
A multimodal MRI-based classification signature emerges just prior to symptom onset in frontotemporal dementia mutation carriers
  1. Rogier A Feis1,2,3,
  2. Mark J R J Bouts1,2,3,
  3. Frank de Vos1,2,3,
  4. Tijn M Schouten1,2,3,
  5. Jessica L Panman1,4,
  6. Lize C Jiskoot1,4,
  7. Elise G P Dopper4,
  8. Jeroen van der Grond1,
  9. John C van Swieten4,5,
  10. Serge A R B Rombouts1,2,3
  1. 1 Department of Radiology, Leiden University Medical Centre, Leiden, The Netherlands
  2. 2 Leiden Institute for Brain and Cognition, Leiden University, Leiden, The Netherlands
  3. 3 Institute of Psychology, Leiden University, Leiden, The Netherlands
  4. 4 Department of Neurology, Erasmus Medical Centre, Rotterdam, The Netherlands
  5. 5 Department of Clinical Genetics, VU University Medical Centre, Amsterdam, The Netherlands
  1. Correspondence to Mr Rogier A Feis, Department of Radiology, Leiden University Medical Centre, Leiden 2333 ZA, The Netherlands; r.a.feis{at}lumc.nl

Abstract

Background Multimodal MRI-based classification may aid early frontotemporal dementia (FTD) diagnosis. Recently, presymptomatic FTD mutation carriers, who have a high risk of developing FTD, were separated beyond chance level from controls using MRI-based classification. However, it is currently unknown how these scores from classification models progress as mutation carriers approach symptom onset. In this longitudinal study, we investigated multimodal MRI-based classification scores between presymptomatic FTD mutation carriers and controls. Furthermore, we contrasted carriers that converted during follow-up (‘converters’) and non-converting carriers (‘non-converters’).

Methods We acquired anatomical MRI, diffusion tensor imaging and resting-state functional MRI in 55 presymptomatic FTD mutation carriers and 48 healthy controls at baseline, and at 2, 4, and 6 years of follow-up as available. At each time point, FTD classification scores were calculated using a behavioural variant FTD classification model. Classification scores were tested in a mixed-effects model for mean differences and differences over time.

Results Presymptomatic mutation carriers did not have higher classification score increase over time than controls (p=0.15), although carriers had higher FTD classification scores than controls on average (p=0.032). However, converters (n=6) showed a stronger classification score increase over time than non-converters (p<0.001).

Conclusions Our findings imply that presymptomatic FTD mutation carriers may remain similar to controls in terms of MRI-based classification scores until they are close to symptom onset. This proof-of-concept study shows the promise of longitudinal MRI data acquisition in combination with machine learning to contribute to early FTD diagnosis.

  • frontotemporal dementia
  • mapt protein, human
  • grn protein, human
  • c9orf72, human
  • diffusion tensor imaging
  • resting-state functional mri
  • multimodal mri
  • classification
  • machine learning

http://dx.doi.org/10.1136/jnnp-2019-320774

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Introduction

    MRI-based classification is a novel tool that may aid early frontotemporal dementia (FTD) diagnosis. Classification algorithms using structural T1-weighted scans1 2 or multimodal imaging3–7 accurately distinguish patients with FTD from controls. However, detection of patients with FTD in earlier disease stages is necessary to further advance future treatment through increasing recruitment accuracy and clinical trial efficiency.8

    To detect FTD-related neuropathology in an earlier stage, we applied MRI-based classification on a sample of presymptomatic FTD mutation carriers, who have a nearly full risk of developing FTD, and controls.9 We found that it was possible to separate presymptomatic FTD mutation carriers from healthy controls beyond chance level, showing the promise of MRI-based machine learning to detect early-stage FTD-related changes before symptom onset in a single-subject setting. However, classification performance was modest, partly due to the cross-sectional design of the study. Specifically, the time to symptom onset in presymptomatic mutation carriers was unknown and carriers who were still far from conversion may have had less FTD-related pathological changes than those who were close to symptom onset. To understand how classification scores develop over time, it is therefore necessary to apply classification models on longitudinal MRI data in presymptomatic FTD populations.

    Here, we use longitudinal data of presymptomatic FTD mutation carriers and healthy controls to relate FTD disease progression with classification scores from an MRI-based classification model. First, we explore the progression of FTD classification scores over time between presymptomatic FTD mutation carriers and controls. Second, we contrast classification score trajectories of mutation carriers that developed symptoms of FTD during the course of the study (‘converters’) and mutation carriers who did not (‘non-converters’). Our aim is to test whether the MRI-based classification model is sensitive to early FTD-related changes in genetic FTD that occur before and around symptom onset.

    Materials and methods

    Participants

    The FTD-Risk Cohort (FTD-RisC9–14) is a longitudinal study that follows healthy, 50% at-risk family members of patients with genetic FTD on a 2-year basis. Subjects were assigned to mutation (microtubule-associated protein tau (MAPT), progranulin (GRN) or chromosome 9 open reading frame 72 (C9orf72)) carrier or control groups by DNA genotyping, as described in previous work.10 11 For the current study, we included the baseline data of 55 presymptomatic FTD mutation carriers (8 MAPT, 35 GRN, 12 C9orf72) and 48 healthy familial controls (6 MAPT family, 31 GRN family and 11 C9orf72 family), who entered FTD-RisC between May 2010 and March 2016 and were between 40 and 70 years old at inclusion.9 At study entry, all subjects were asymptomatic in accordance with established diagnostic criteria for behavioural variant FTD (bvFTD),15 primary progressive aphasia (PPA)16 and amyotrophic lateral sclerosis.17 We included follow-up data as available at 2, 4 and 6 years (figure 1). Exclusion criteria were current or past neurological or primary psychiatric disorders, history of drug abuse, large image artefacts and gross brain pathology other than atrophy. As such, we included 103 subjects at baseline, 98 subjects after 2 years’ follow-up, 56 subjects after 4 years’ follow-up and 43 subjects after 6 years’ follow-up (figure 1). However, 20 subjects had incomplete data (n=10) or artefacts (n=10) at a certain time point, leaving 55 mutation carriers and 48 controls at baseline, 45 mutation carriers and 42 controls after 2 years’ follow-up, 27 mutation carriers and 23 controls at 4 years’ follow-up, and 20 mutation carriers and 20 controls at 6 years’ follow-up.

    Figure 1
    Figure 1

    Study flow diagram blue fields show number of subjects included at each time point. Red fields show excluded subjects, reasons for exclusion and conversion information. C9orf72, chromosome nine open reading frame 72; (bv)FTD, (behavioural variant) frontotemporal dementia; GRN, progranulin; MAPT, microtubule-associated protein tau; nfvPPA, non-fluent variant primary progressive aphasia.

    For details on the sample on which the FTD classification model was trained, please refer to Bouts et al.7 In short, 23 patients with sporadic bvFTD and 35 controls between 40 and 80 years old were included to undergo a clinical assessment and MRI between November 2009 and November 2012. The MRI acquisition protocol was similar to the protocol applied in the current sample of carriers and controls. Imaging processing steps were identical to processing steps in the current analysis.

    Standard protocol approvals, registrations, and patient consents

    Participants and clinical investigators were blinded to the participants’ genetic status, except for those who underwent predictive testing at their own request. For converters, genetic counselling was offered to the patient and family members, and genetic status was unblinded to confirm the presence of the pathogenic mutation. The study was conducted in accordance with regional regulations and the Declaration of Helsinki. The Erasmus Medical Centre and Leiden University Medical Centre local medical ethics committees approved the study, and every participant provided written informed consent.

    Conversion

    Conversion was determined in a multidisciplinary consensus meeting of the Erasmus Medical Centre FTD Expertise Centre, involving neurologists (JCvS), neuropsychologists (JLP, LCJ), neuroradiologists, geriatricians, a clinical geneticist, and a care consultant. In the consensus meetings, anamnestic and heteroanamnestic information, neuropsychological assessment and MR imaging of the brain were reviewed, and genetic status was unblinded. For detailed information on conversion criteria, see Jiskoot et al.12 14

    MRI data acquisition

    All subjects were scanned at the Leiden University Medical Centre using a 3 T MRI scanner (Achieva; Philips Medical Systems, Best, The Netherlands) with an eight-channel SENSE head coil. The imaging protocol included a whole-brain near-isotropic T1-weighted three-dimensional MRI sequence for cortical and subcortical tissue-type segmentation, a diffusion-weighted imaging sequence for assessments of white matter diffusivity and resting-state functional MRI (rs-fMRI) for the calculation of functional connectivity measures. Participants were instructed to lie still with their eyes closed and not to fall asleep during rs-fMRI. For scan parameters, see table 1. Although MRI sequence parameters were fixed over time, a routine MRI software upgrade was performed at our MRI site by the manufacturer in September 2015. We added a covariate to all our imaging analyses to correct for this.

    View this table:
    Table 1

    MRI sequence parameter settings

    Image pre-processing

    Image pre-processing was performed identically to earlier work.9 For a detailed description, please see online supplemental material. All registration and segmentation steps were critically reviewed and errors were corrected accordingly.

    Supplemental material

    Feature selection

    We selected MRI features for classification based on earlier work, in which patients with bvFTD were separated from controls with high accuracy.7 We selected the best performing combination of MRI features, including grey matter density (GMD), functional correlations (FCor), fractional anisotropy (FA), and mean diffusivity (MD). Features were calculated using the same procedures as in previous work.7 9 18 For a detailed description, please see online supplemental material. The GMD feature included 96 cortical and 14 subcortical values, the FCor feature consisted of 2415 functional correlation values between 70 independent components, and the FA and MD features each contained 20 values.

    Classification model application

    We trained the bvFTD model on the features of patients with bvFTD and controls (ie, GMD, FCor, FA, and MD features, as well as gender) using an elastic net logistic regression classifier.7 Elastic net hyperparameters were optimised in a 10-fold cross-validation. Next, we applied the resulting bvFTD classification model on our longitudinal data of FTD mutation carriers and controls. As such, we obtained a predicted probability at each time point for each subject, which we call bvFTD classification scores. Normally, these classification scores are transformed using a logit link function to range between 0 (representing control) and 1 (representing patient with bvFTD). However, such a transformation introduces non-linearity to the scores, which makes them unsuited for repeated measures analyses. Therefore, we used the raw scores, which in this case ranged between –8.9 and 8.4. Here, negative values represent controls and positive values represent patients with bvFTD.

    Statistical analysis

    To investigate the bvFTD classification scores’ progression over time, we used a repeated measures linear mixed-effects model. Mixed-effects models estimate both fixed and random effects of predictor variables in order to determine global effects and account for between-subject variance, respectively.19 20 We used a random intercept to account for between-subject variation in classification score. Our main covariate of interest was the interaction between label and time point (ie, the progression over time contrasted between mutation carriers and controls). Other covariates included time point, label, age at baseline, sex and a covariate to correct for the MRI scanner software upgrade. Additionally, we added an interaction between label and age at baseline to see whether the effect of age at baseline was different for carriers and controls. Covariates were tested for significance using likelihood ratio χ2 tests. In the presymptomatic carrier–control analysis, we excluded data points from subjects who had at that time developed symptoms.

    In a subgroup analysis within the carrier group, we contrasted converters and non-converters. This subgroup analysis was performed using the same covariates as the carrier–control analysis and included presymptomatic as well as symptomatic data points.

    Statistical analyses of baseline demographic variables included unpaired t-tests for age and education, a χ2 test for gender distribution and a Mann-Whitney U test for mini-mental state examination (MMSE) scores (0–30). All statistical analyses were performed using R (R Core 2016, Vienna, Austria). For the mixed-effects model analyses, we used the nlme package (V.3.1.12821).

    Data availability

    Raw data were generated at the Leiden University Medical Centre. The derived data, as well as scripts, that support the findings of this study are available from the corresponding author on reasonable request.

    Results

    Demographics

    Presymptomatic mutation carriers and controls

    At baseline, there were no differences between mutation carriers and controls in terms of age, gender, education or MMSE (table 2), nor were there differences at any follow-up time point. The proportion of scans acquired after the MRI software upgrade was installed varied per time point but did not differ between mutation carriers and controls at each time point. The total follow-up time of all subjects was 376 years, and the average follow-up time was 3.7 years.

    View this table:
    Table 2

    Baseline demographics

    Converters and non-converters

    During the study, six mutation carriers converted to FTD. One MAPT mutation carrier and one mutation GRN carrier converted to bvFTD after 2 years and two MAPT mutation carriers converted to bvFTD after 4 years. These subjects presented with progressive behavioural deterioration, functional decline, and frontal and/or temporal lobe atrophy on structural MRI, fulfilling the international diagnostic consensus criteria for bvFTD with definite frontotemporal lobar degeneration pathology.15 Two GRN mutation carriers converted after 4 years and presented with isolated language difficulties and no impairment in daily living activities. Both showed a non-fluent, halting speech with sound errors and agrammatism, fulfilling the diagnostic criteria for non-fluent variant primary progressive aphasia (nfvPPA16). For a full description of these subjects’ clinical profile including neuropsychological assessment, see Jiskoot et al.12 14 Baseline demographics for the converters and non-converters are shown in table 3, and a timeline for each converter, including bvFTD classification scores and time of conversion, is given in figure 2. Other than higher education for converters than non-converters (p=0.003), there were no significant demographic differences at baseline. However, at 4 years’ follow-up, converters had significantly lower MMSE than non-converters (p=0.001). At that time, all converters had converted to the symptomatic phase.

    View this table:
    Table 3

    Converter baseline demographics

    Figure 2
    Figure 2

    Converter timelines including baseline demographic information, bvFTD classification scores, clinical follow-up information, and data exclusion information. For reference, classification scores in our population ranged from −8.9 to 8.4, where higher scores represent greater resemblance to patients with bvFTD according to the classification model. Blue panels indicate presymptomatic status; red panels indicate subjects have converted. (bv)FTD, (behavioural variant) frontotemporal dementia; DTI, diffusion tensor imaging; GRN, progranulin; MAPT, microtubule-associated protein tau; nfvPPA, non-fluent variant primary progressive aphasia.

    Classification scores over time

    Presymptomatic mutation carriers and controls

    Application of the bvFTD model resulted in a classification score for each included time point. Figure 3A shows the results for all carriers and controls; online supplementary figure 1A–C shows the results split per gene. Data points at which mutation carriers had developed symptoms (triangles) were not included in this presymptomatic analysis. Classification scores of mutation carriers did not show a stronger increase over time than controls’ scores (p=0.15). However, on average, mutation carriers’ classification scores were higher than scores of controls (p=0.032). There was a strong effect of age at baseline on the classification score (p<0.001), but this effect was not different between mutation carriers and controls (p=0.72). Male participants had higher classification scores than female participants (p<0.001).

    Figure 3
    Figure 3

    Carrier and control bvFTD classification score values indicate bvFTD classification scores as determined by the bvFTD classification model.7 Classification scores for all FTD mutation carriers (yellow) and controls (blue) are shown in (A). Classification scores for converters (pink) and non-converters (green) are shown in (B). Triangles indicate converted status at appropriate time points. Regression lines were based on predicted values from the repeated measures linear mixed-effects model. (bv)FTD, (behavioural variant frontotemporal dementia.

    Converters and non-converters

    The bvFTD classification scores for all converters and non-converters are shown in figure 3B. In online supplementary figure 1D–F, the results are shown split per gene. Classification scores of converters increased more strongly over time than non-converters' scores (p<0.001), though they were similar on average (p=0.33). There was a strong effect of age at baseline on classification score (p=0.006), which was more pronounced in converters than in non-converters (p=0.035). Again, male participants had higher classification scores than female participants (p=0.001).

    Discussion

    In this study, we aimed to assess the sensitivity of MRI-based classification to FTD-related changes in the brain before and around FTD symptom onset. We used an MRI-based bvFTD classification model7 on longitudinal MRI data of presymptomatic FTD mutation carriers and healthy controls with a follow-up of up to 6 years. Overall, mutation carriers had higher bvFTD classification scores on average than controls, but did not differ from controls in longitudinal score development. However, a subgroup analysis within mutation carriers showed that mutation carriers who converted during follow-up had a stronger score increase over time than non-converting mutation carriers. Converters showed a stronger increase of classification scores between time points, as well as a stronger effect of age at baseline. Our longitudinal findings highlight that MRI-based bvFTD classification is sensitive to FTD-related changes in the brain that arise around symptom onset in FTD mutation carriers.

    Carrier–control analysis revealed that over the course of 2 to 6 years of follow-up, mutation carriers did not have significantly stronger increases in bvFTD classification scores over time than controls. These results are in line with our previous work, where we showed that the bvFTD model could not separate presymptomatic mutation carriers from controls beyond chance level at baseline.9 That work was limited by its cross-sectional design, which greatly reduced power to find differences in a population with uncertain time to symptom onset. However, even in our current longitudinal design, the mutation carrier group as a whole did not differ from controls in bvFTD classification score increase over time. On the other hand, mutation carriers that converted to FTD during the follow-up of this study did show stronger classification score increases over time. These findings may imply that presymptomatic mutation carriers remain quite similar to controls in terms of individually detectable changes in MRI features until they are within a few years of symptom onset. A direct correlation between MRI-based classification scores and neuropathology has yet to be established. Nonetheless, our results seem in line with the hypothesis that FTD-related neuropathological processes considerably accelerate in the final years before conversion, which was previously postulated in studies focusing on cognitive, neuroimaging, and fluid biomarkers. Specifically, FTD mutation carriers that convert to FTD typically start to show signs of cognitive decline in the language or executive domains 2 years before diagnosis.12 22 Furthermore, in longitudinal neuroimaging studies, grey matter atrophy14 22 and white matter FA14 were also first picked up 2 years before diagnosis, after which they rapidly expanded. In a large multicentre study23 of the Genetic Frontotemporal Dementia Initiative (GENFI),24 neurofilament light chain levels in cerebrospinal fluid and serum were much higher in symptomatic carriers than in presymptomatic carriers, who had similar levels to non-carriers. Additionally, longitudinal samples showed a threefold to fourfold increase in neurofilament light chain levels in the cerebrospinal fluid in carriers that converted during the study.23 All of these results suggest that the presymptomatic stage remains stable until a couple of years before symptom onset, when neurodegeneration rapidly develops.

    Strengths of this study include its unique longitudinal design, the inclusion of multimodal MRI data of presymptomatic FTD mutation carriers, and the inclusion of mutation carriers that converted during follow-up. Even so, the length of follow-up may have been insufficient to capture the differences between mutation carriers and controls over time using classification models. A limitation to our study was the heterogeneity of our sample, as we had to pool MAPT, GRN and C9orf72 mutation carriers in order to obtain sufficient sample size for robust analyses. Also, the classification model was trained on patients with bvFTD and controls, and may therefore not be sensitive to MRI changes associated with non-behavioural variants.25 The greater part (ie, 35 out of 55) of FTD mutation carriers were GRN carriers, who usually develop bvFTD, but also frequently convert to nfvPPA, or may develop atypical parkinsonian syndromes.26 If MRI changes associated with these other diseases are not picked up by the bvFTD model, it might explain why we found no differences between mutation carriers and controls. For example, patients with bvFTD show diffuse atrophy and white matter diffusivity changes in bilateral frontotemporal regions, while in patients with nfvPPA, changes are predominantly left hemispheric and are more specifically located.27 Lastly, since the FTD-RisC is an ongoing study, the variable follow-up time is a limitation. Although the repeated measures model is robust for differences in follow-up time, the shorter follow-up time in some carriers could explain the limited number of converters. Specifically, there were no converters in the C9orf72 carrier group, which was included in the FTD-RisC study at a later time than the MAPT and GRN groups. Therefore, the results of our converters' analysis should not be generalised to C9orf72 carriers. Lastly, it should be noted that we chose to not interpret the bvFTD classification model's beta values, even though it might seem interesting to investigate which features drove the classification. Contrary to explanatory regression models, the beta values of our classification model do not designate direct relationships between the features and classification score, nor do they reflect mean differences between the groups.28 For example, each feature’s effect is conditional on the effects of all other features in the model, and multicollinearity between features may result in suppression of the effect of some features. Furthermore, the model was forced to be sparse, which may have resulted in the suppression or exclusion of true effects from the model. Given the large feature space (ie, 110 GMD values, 2415 FCor values, 20 FA values and 20 MD values) on which the bvFTD classification model was trained, these effects are likely to have occurred, obscuring the true meaning of the model’s beta values.

    Despite these limitations, we were able to show that FTD-related MRI changes can be recognised by an MRI-based classification model around symptom onset. This proof-of-concept study further emphasises the importance of longitudinal data acquisition such as the FTD-RisC and GENFI. With longer follow-up times and more converters, it may eventually be possible to find exclusively presymptomatic changes on an individual level, which would be an important step towards early and accurate FTD diagnosis. Larger sample sizes might also facilitate stratification according to the different gene mutations (ie, MAPT, GRN and C9orf72) or different clinical variants, which was currently infeasible. In order to increase the clinical relevance of MRI-based machine learning for dementia, future studies should focus on replicating classification models in larger samples, validating them in separately established samples, and finally applying them in undiagnosed memory clinic populations to establish—with the use of follow-up information or postmortem examination—whether they are useful in the earliest phases of the disease.

    To conclude, we showed that FTD-related changes are measurable by MRI-based classification in years around symptom onset in FTD mutation carriers. This indicates that MRI-based classification is a reasonable candidate to aid early diagnosis of FTD and could contribute to improved development of disease-modifying treatments that can slow down or possibly reverse the underlying neuropathological processes.

    References

    1. ↵
      2. Raamana PR ,
      3. Rosen H ,
      4. Miller B , et al
      . Three-cClass differential diagnosis among Alzheimer disease, frontotemporal dementia, and controls. Front. Neurol. 2014;5:71.doi:10.3389/fneur.2014.00071
    2. ↵
      2. Meyer S ,
      3. Mueller K ,
      4. Stuke K , et al
      . Predicting behavioral variant frontotemporal dementia with pattern classification in multi-center structural MRI data. Neuroimage Clin 2017;14:656–62.doi:10.1016/j.nicl.2017.02.001
      OpenUrl
    3. ↵
      2. Klöppel S ,
      3. Peter J ,
      4. Ludl A , et al
      . Applying automated MR-based diagnostic methods to the memory clinic: a prospective study. JAD 2015;47:939–54.doi:10.3233/JAD-150334
      OpenUrl
    4. ↵
      2. Koikkalainen J ,
      3. Rhodius-Meester H ,
      4. Tolonen A , et al
      . Differential diagnosis of neurodegenerative diseases using structural MRI data. Neuroimage Clin 2016;11:435–49.doi:10.1016/j.nicl.2016.02.019
      OpenUrl
    5. ↵
      2. Bron EE ,
      3. Smits M ,
      4. Papma JM , et al
      . Multiparametric computer-aided differential diagnosis of Alzheimer’s disease and frontotemporal dementia using structural and advanced MRI. Eur Radiol 2017;27:3372–82.doi:10.1007/s00330-016-4691-x
      OpenUrl
    6. ↵
      2. Canu E ,
      3. Agosta F ,
      4. Mandic-Stojmenovic G , et al
      . Multiparametric MRI to distinguish early onset Alzheimer's disease and behavioural variant of frontotemporal dementia. Neuroimage Clin 2017;15:428–38.doi:10.1016/j.nicl.2017.05.018
      OpenUrl
    7. ↵
      2. Bouts MJRJ ,
      3. Möller C ,
      4. Hafkemeijer A , et al
      . Single subject classification of Alzheimer’s disease and behavioral variant frontotemporal dementia using anatomical, diffusion tensor, and resting-state functional magnetic resonance imaging. JAD 2018;62:1827–39.doi:10.3233/JAD-170893
      OpenUrl
    8. ↵
      2. Tsai RM ,
      3. Boxer AL
      . Therapy and clinical trials in frontotemporal dementia: past, present, and future. J Neurochem 2016;138:211–21.doi:10.1111/jnc.13640
      OpenUrl
    9. ↵
      2. Feis RA ,
      3. Bouts MJRJ ,
      4. Panman JL , et al
      . Single-subject classification of presymptomatic frontotemporal dementia mutation carriers using multimodal MRI. Neuroimage Clin 2019:101718.doi:10.1016/j.nicl.2019.101718
    10. ↵
      2. Dopper EGP ,
      3. Rombouts SARB ,
      4. Jiskoot LC , et al
      . Structural and functional brain connectivity in presymptomatic familial frontotemporal dementia. Neurology 2014;83:e19–26.doi:10.1212/WNL.0000000000000583
      OpenUrlCrossRef
    11. ↵
      2. Jiskoot LC ,
      3. Dopper EGP ,
      4. Heijer Tden , et al
      . Presymptomatic cognitive decline in familial frontotemporal dementia: a longitudinal study. Neurology 2016;87:384–91.doi:10.1212/WNL.0000000000002895
      OpenUrl
    12. ↵
      2. Jiskoot LC ,
      3. Panman JL ,
      4. van Asseldonk L , et al
      . Longitudinal cognitive biomarkers predicting symptom onset in presymptomatic frontotemporal dementia. J Neurol 2018;265:1381–92.doi:10.1007/s00415-018-8850-7
      OpenUrl
    13. ↵
      2. Papma JM ,
      3. Jiskoot LC ,
      4. Panman JL , et al
      . Cognition and gray and white matter characteristics of presymptomatic C9orf72 repeat expansion. Neurology 2017;89:1256–64.doi:10.1212/WNL.0000000000004393
      OpenUrl
    14. ↵
      2. Jiskoot LC ,
      3. Panman JL ,
      4. Meeter LH , et al
      . Longitudinal multimodal MRI as prognostic and diagnostic biomarker in presymptomatic familial frontotemporal dementia. Brain 2019;142:193–208.doi:10.1093/brain/awy288
      OpenUrl
    15. ↵
      2. Rascovsky K ,
      3. Hodges JR ,
      4. Knopman D , et al
      . Sensitivity of revised diagnostic criteria for the behavioural variant of frontotemporal dementia. Brain 2011;134:2456–77.doi:10.1093/brain/awr179
      OpenUrlCrossRefPubMedWeb of Science
    16. ↵
      2. Gorno-Tempini ML ,
      3. Hillis AE ,
      4. Weintraub S , et al
      . Classification of primary progressive aphasia and its variants. Neurology 2011;76:1006–14.doi:10.1212/WNL.0b013e31821103e6
      OpenUrlCrossRefPubMed
    17. ↵
      2. Ludolph A ,
      3. Drory V ,
      4. Hardiman O , et al
      . A revision of the El Escorial criteria - 2015. Amyotroph Lateral Scler Frontotemporal Degener 2015;16:291–2.doi:10.3109/21678421.2015.1049183
      OpenUrlPubMed
    18. ↵
      2. Schouten TM ,
      3. Koini M ,
      4. de Vos F , et al
      . Combining anatomical, diffusion, and resting state functional magnetic resonance imaging for individual classification of mild and moderate Alzheimer's disease. Neuroimage Clin 2016;11:46–51.doi:10.1016/j.nicl.2016.01.002
      OpenUrl
    19. ↵
      2. Laird NM ,
      3. Ware JH
      . Random-effects models for longitudinal data. Biometrics 1982;38:963–74.doi:10.2307/2529876
      OpenUrlCrossRefPubMedWeb of Science
    20. ↵
      2. Sullivan LM ,
      3. Dukes KA ,
      4. Losina E
      . An introduction to hierarchical linear modelling. Stat Med 1999;18:855–88.doi:10.1002/(SICI)1097-0258(19990415)18:7<855::AID-SIM117>3.0.CO;2-7
      OpenUrlCrossRefPubMedWeb of Science
    21. ↵
      2. Pinheiro J ,
      3. Bates D ,
      4. DebRoy S , et al
      . nlme: linear and nonlinear mixed effects models [R package nlme version 3.1-137], 2018. Available: https://cran.r-project.org/web/packages/nlme/index.html [Accessed 16 Aug 2018].
    22. ↵
      2. Rohrer JD ,
      3. Warren JD ,
      4. Barnes J , et al
      . Mapping the progression of progranulin-associated frontotemporal lobar degeneration. Nat Rev Neurol 2008;4:455–60.doi:10.1038/ncpneuro0869
      OpenUrl
    23. ↵
      2. Meeter LH ,
      3. Dopper EG ,
      4. Jiskoot LC , et al
      . Neurofilament light chain: a biomarker for genetic frontotemporal dementia. Ann Clin Transl Neurol 2016;3:623–36.doi:10.1002/acn3.325
      OpenUrl
    24. ↵
      2. Rohrer JD ,
      3. Nicholas JM ,
      4. Cash DM , et al
      . Presymptomatic cognitive and neuroanatomical changes in genetic frontotemporal dementia in the genetic frontotemporal dementia Initiative (GENFI) study: a cross-sectional analysis. Lancet Neurol 2015;14:253–62.doi:10.1016/S1474-4422(14)70324-2
      OpenUrlCrossRefPubMed
    25. ↵
      2. Seelaar H ,
      3. Rohrer JD ,
      4. Pijnenburg YAL , et al
      . Clinical, genetic and pathological heterogeneity of frontotemporal dementia: a review. J Neurol Neurosurg Psychiatry 2011;82:476–86.doi:10.1136/jnnp.2010.212225
      OpenUrlAbstract/FREE Full Text
    26. ↵
      2. Benussi A ,
      3. Padovani A ,
      4. Borroni B
      . Phenotypic heterogeneity of monogenic frontotemporal dementia. Front Aging Neurosci 2015;7:171.doi:10.3389/fnagi.2015.00171
    27. ↵
      2. Zhang Y ,
      3. Tartaglia MC ,
      4. Schuff N , et al
      . MRI signatures of brain macrostructural atrophy and microstructural degradation in frontotemporal lobar degeneration subtypes. JAD 2013;33:431–44.doi:10.3233/JAD-2012-121156
      OpenUrl
    28. ↵
      2. Shmueli G
      . To explain or to predict? Statist Sci 2010;25:289–310.doi:10.1214/10-STS330
      OpenUrl

    Footnotes

    • Correction notice This article has been corrected since it was published Online First. Serge Rombouts has been linked to affiliation 3.

    • Contributors RAF was involved in the study design, analyses, interpretation of the data, and the drafting and revision of the manuscript. MJRJB was involved in the analysis and interpretation of the data, and the revision of the manuscript. FdV was involved in the interpretation of the data and the revision of the manuscript. TMS was involved in the interpretation of the data and the revision of the manuscript. JLP was involved in the data acquisition and the revision of the manuscript. LCJ was involved in the data acquisition and the revision of the manuscript. EGPD was involved in the data acquisition and the revision of the manuscript. JvdG was involved in the interpretation of the data and the revision of the manuscript. JCvS was involved in the study design, interpretation of the data and the revision of the manuscript. SARBR was involved in the study design, interpretation of the data and the revision of the manuscript.

    • Funding The authors of this work were supported by the Leiden University Medical Centre MD/PhD Scholarship (to RAF), ZonMw programme Memorabel project 733050103, JPND PreFrontAls consortium project 733051042 (to JCvS) and NWO VICI grant 016-130-667 (to SARBR).

    • Disclaimer The views expressed are those of the authors and not necessarily those of the funding sources. The funding sources were not involved in the design of the study; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.

    • Competing interests None declared.

    • Patient consent for publication Obtained.

    • Ethics approval Erasmus Medical Centre and Leiden University Medical Centre local medical ethics committees.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Data availability statement Data are available on reasonable request.

    Linked Articles