FLFedor Levin

Fedor Levin

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Fedor Levin has a total of 103 co-authors. Here are the top co-authors by number of shared publications:

  • Martin Dyrba

    Deutsches Zentrum für Neurodegenerative Erkrankungen (DZNE), Rostock, Germany

    8 shared publications
  • Stefan J. Teipel

    Deutsches Zentrum für Neurodegenerative Erkrankungen (DZNE), Rostock, Germany

    7 shared publications
  • Christoph Laske

    German Center for Neurodegenerative Diseases (DZNE), Tuebingen, Germany

    5 shared publications
  • Michel J. Grothe

    Unidad de Trastornos del Movimiento, Servicio de Neurología y Neurofisiología Clínica, Instituto de Biomedicina de Sevilla, Hospital Universitario Virgen del Rocío/CSIC/Universidad de Sevilla, Seville, Spain

    5 shared publications
  • Michel J. Grothe

    CIEN Foundation, Reina Sofia Alzheimer Center, ISCIII, Madrid, Madrid, Spain

    5 shared publications
  • Anja Schneider

    German Center for Neurodegenerative Diseases (DZNE)

    4 shared publications

Latest Publications

  • Open Access

    Hypometabolic subtypes of AD are linked to comorbid hippocampal sclerosis and Lewy body pathology

    by Fedor Levin, Martin Dyrba, Stefan J. Teipel, Michel J. Grothe

    Background: Neuroimaging studies have identified distinct ‘typical/neocortical’ and ‘limbic-predominant’ hypometabolic subtypes of AD with different clinical and biomarker characteristics. We investigated associations of these subtypes with postmortem neuropathological measures in an observational study. Methods: Antemortem FDG-PET scans of 74 participants from the ADNI autopsy cohort were classified into previously described typical/neocortical and limbic-predominant subtype patterns. We used Bayesian regression and ANCOVA to test associations between the subtypes and neuropathological features. Results: Results were inconclusive for Thal phases, Braak stages, CERAD neuritic plaque scores, hippocampal tangle density, and TDP-43 pathology (BF10 between 0.447 and 1.146). However, the limbic-predominant subtype was associated with hippocampal sclerosis (BF10 = 3.842, moderate level of evidence), whereas the typical/neocortical subtype was associated with Lewy body pathology (BF10 = 10.093, strong level of evidence). Conclusions: These findings highlight the influence of AD and non-AD-specific pathologies on neurodegeneration patterns and may provide directions for research into hypometabolic pattern analysis as an indirect marker of comorbid pathology.


  • Open Access

    Associations between digital speech features of automated cognitive tasks and trajectories of brain atrophy and cognitive decline in early Alzheimer's disease

    by Qingyue Li, Stefanie Koehler, Alexandra Koenig, Martin Dyrba, Elisa Mallik, Nicklas Linz, Josef Priller, Eike Spruth, Slawek Altenstein, Jens Wiltfang, et al.

    Background: Speech-based features extracted from telephone-based cognitive tasks show promise for detecting cognitive decline in prodromal and manifest dementia. Little is known about the cerebral underpinnings of these speech features. Objective: To examine associations between speech features, brain atrophy, and longitudinal cognitive decline in individuals at risk for Alzheimer's disease (AD). Methods: Healthy volunteers, individuals with subjective cognitive decline, and those with mild cognitive impairment completed phonebot-guided semantic verbal fluency (SVF) and 15-word verbal learning task (VLT). Speech features were automatically extracted, and a global cognitive score (SB-C score) was computed. We analyzed data from 161 participants for cognitive trajectories, 141 for cross-sectional brain atrophy, and 102 for longitudinal brain changes. Analyses were conducted using multiple linear regressions, mixed-effects models, and voxel-based morphometry. Results: The SB-C score was associated with bilateral hippocampal volumes, SVF features were primarily associated with left hemisphere regions, including the inferior frontal, parahippocampal, and superior/middle temporal gyri ( p uncorr < 0.001). SB-C score, SVF correct counts, and VLT delayed recall were associated with atrophy rates in the hippocampal/parahippocampal gyrus and left middle/inferior temporal gyri ( p FDR < 0.05). These features were also associated with cognitive decline assessed via Preclinical Alzheimer's Cognitive Composite 5, SVF, and Wordlist learning delayed recall ( p FDR < 0.01). Word frequency and temporal cluster switches showed varying associations with cognitive trajectories. Other features did not show robust associations. Conclusions: In this study, we highlight the potential of digital speech features for identifying brain atrophy and cognitive decline over time in at-risk AD populations.


  • Open Access

    An Unsupervised XAI Framework for Dementia Detection with Context Enrichment

    by Devesh Singh, Yusuf Brima, Fedor Levin, Martin Becker, Bjarne Hiller, Andreas Hermann, Irene Villar-Munoz, Lukas Beichert, Alexander Bernhardt, Katharina Buerger, et al.

    Introduction Explainable Artificial Intelligence (XAI) methods enhance the diagnostic efficiency of clinical decision support systems by making the predictions of a convolutional neural network’s (CNN) on brain imaging more transparent and trustworthy. However, their clinical adoption is limited due to limited validation of the explanation quality. Our study introduces a framework that evaluates XAI methods by integrating neuroanatomical morphological features with CNN-generated relevance maps for disease classification. Methods We trained a CNN using brain MRI scans from six cohorts: ADNI, AIBL, DELCODE, DESCRIBE, EDSD, and NIFD (N=3253), including participants that were cognitively normal, with amnestic mild cognitive impairment, dementia due to Alzheimer’s disease and frontotemporal dementia. Clustering analysis benchmarked different explanation space configurations by using morphological features as proxy-ground truth. We implemented three post-hoc explanations methods: i) by simplifying model decisions, ii) explanation-by-example, and iii) textual explanations. A qualitative evaluation by clinicians (N=6) was performed to assess their clinical validity. Results Clustering performance improved in morphology enriched explanation spaces, improving both homogeneity and completeness of the clusters. Post hoc explanations by model simplification largely delineated converters and stable participants, while explanation-by-example presented possible cognition trajectories. Textual explanations gave rule-based summarization of pathological findings. Clinicians’ qualitative evaluation highlighted challenges and opportunities of XAI for different clinical applications. Conclusion Our study refines XAI explanation spaces and applies various approaches for generating explanations. Within the context of AI-based decision support system in dementia research we found the explanations methods to be promising towards enhancing diagnostic efficiency, backed up by the clinical assessments.

  • Open Access

    Detecting amyloid positivity from MRI scans using comprehensible convolutional neural networks

    by Fedor Levin, Qingyue Li, Devesh Singh, Katharina Buerger, Klaus Fliessbach, Michael T. Heneka, Luca Kleineidam, Christoph Laske, Robert Perneczky, Oliver Peters, et al.

    Background: Analysis of neuroimaging data based on convolutional neural networks (CNNs) can improve detection of clinically relevant characteristics of patients with Alzheimer’s disease (AD). Previously, our group developed a CNN‐based approach for detecting AD via magnetic resonance imaging (MRI) scans and for identifying features that are relevant to the decision of the network. In the current study, we aimed to evaluate the potential utility of applying this approach to MRI scans to assist in the identification of individuals at high risk for amyloid positivity to aid in the selection of study samples and case finding for treatment.Method: In the current analysis, we have trained a CNN to detect amyloid positivity using MRI scans from 1461 Alzheimer’s Disease Neuroimaging Initiative (ADNI) participants (498 cognitively normal participants, 103 participants with significant memory concern, 640 participants with mild cognitive impairment, and 220 participants with AD dementia). Amyloid positivity was assessed via amyloid PET scans obtained with [18F]florbetapir or [18F]florbetaben and quantified on a Centiloid scale. A threshold of 24.1 CL categorized 46% of participants as amyloid‐positive. The modeling approach was evaluated using 10‐fold cross‐validation, the number of epochs in training was set to 10.Result: For each of 10 cross‐validation folds, we selected a model state corresponding to an epoch showing best performance in the validation partition. Balanced accuracy across these models ranged from 0.62 to 0.72 with an average of 0.68 (SD = 0.03).Conclusion: We used a previously established approach to train CNNs for detecting amyloid positivity using MRI scans. Such models, particularly when tuned to have low rates of false negatives, have a potential to enhance identification of patients who would benefit from more in‐depth assessments, which could then inform antibody treatment. We are conducting ongoing work to improve and characterize the modeling approach, including evaluation of relevance maps which indicate importance of brain regions for detecting amyloid positivity. Future work will evaluate the role of amyloid positivity threshold selection. Planned analyses also include validation in independent data such as the German DZNE ‐ Longitudinal Cognitive Impairment and Dementia Study (DELCODE) dataset.

  • Open Access

    Detecting amyloid positivity from MRI scans using comprehensible convolutional neural networks

    by Fedor Levin, Qingyue Li, Devesh Singh, Katharina Buerger, Klaus Fliessbach, Michael T. Heneka, Luca Kleineidam, Christoph Laske, Robert Perneczky, Oliver Peters, et al.

    Background: Analysis of neuroimaging data based on convolutional neural networks (CNNs) can improve detection of clinically relevant characteristics of patients with Alzheimer’s disease (AD). Previously, our group developed a CNN‐based approach for detecting AD via magnetic resonance imaging (MRI) scans and for identifying features that are relevant to the decision of the network. In the current study, we aimed to evaluate the potential utility of applying this approach to MRI scans to assist in the identification of individuals at high risk for amyloid positivity to aid in the selection of study samples and case finding for treatment.Method: In the current analysis, we have trained a CNN to detect amyloid positivity using MRI scans from 1461 Alzheimer’s Disease Neuroimaging Initiative (ADNI) participants (498 cognitively normal participants, 103 participants with significant memory concern, 640 participants with mild cognitive impairment, and 220 participants with AD dementia). Amyloid positivity was assessed via amyloid PET scans obtained with [18F]florbetapir or [18F]florbetaben and quantified on a Centiloid scale. A threshold of 24.1 CL categorized 46% of participants as amyloid‐positive. The modeling approach was evaluated using 10‐fold cross‐validation, the number of epochs in training was set to 10.Result: For each of 10 cross‐validation folds, we selected a model state corresponding to an epoch showing best performance in the validation partition. Balanced accuracy across these models ranged from 0.62 to 0.72 with an average of 0.68 (SD = 0.03).Conclusion: We used a previously established approach to train CNNs for detecting amyloid positivity using MRI scans. Such models, particularly when tuned to have low rates of false negatives, have a potential to enhance identification of patients who would benefit from more in‐depth assessments, which could then inform antibody treatment. We are conducting ongoing work to improve and characterize the modeling approach, including evaluation of relevance maps which indicate importance of brain regions for detecting amyloid positivity. Future work will evaluate the role of amyloid positivity threshold selection. Planned analyses also include validation in independent data such as the German DZNE ‐ Longitudinal Cognitive Impairment and Dementia Study (DELCODE) dataset.

  • Open Access

    The limbic‐predominant hypometabolic subtype of AD is not enriched for comorbid TDP‐43 pathology or hippocampal sclerosis

    by Fedor Levin, Michel J. Grothe, Martin Dyrba, Stefan Teipel

    Background: Previous research has identified distinct subtypes of AD that differed with respect to the distribution of neurofibrillary tangles (NFT). One of them, a limbic‐predominant subtype, was enriched for comorbid TDP‐43 pathology. Our previous data‐driven analysis used FDG‐PET to identify a limbic‐predominant subtype of AD that may correspond to the limbic‐predominant NFT subtype. In the current study, we tested the expected associations between measures of neuropathology such as TDP‐43 pathology, hippocampal sclerosis, and Braak stages of NFT accumulation, and the limbic‐predominant hypometabolic subtype.Method: We analyzed ante‐mortem FDG‐PET and neuropathological data of 74 individuals from the autopsy cohort of the Alzheimer’s Disease Neuroimaging Initiative (ADNI), including 7 cognitively normal participants, 12 participants with MCI, and 55 with dementia (as assessed at the last available clinical evaluation). Severity of TDP‐43 pathology was operationalized as the number of brain areas with TDP‐43 immunoreactive inclusions (including amygdala, hippocampus, entorhinal/inferior temporal cortex and neocortex). Hippocampal sclerosis was operationalized as a binary variable. We measured similarity of individual FDG‐PET profiles to the limbic‐predominant subtype. For that, we used Euclidean distances from each individual profile to the mean profile of the limbic‐predominant AD subtype identified previously in a larger ADNI sample using a data‐driven hierarchical clustering approach (Fig.1). Lower Euclidean distances indicated higher similarity to the subtype. Bayesian linear regressions in JASP were used to assess associations between neuropathological markers and similarity to the limbic‐predominant FDG‐PET pattern, controlling for Thal phase of amyloid accumulation, age, sex, and time delay between the FDG‐PET scan and death in the null model. We assessed Bayes factors representing evidence for a given model in comparison to the null model (BF10).Result: Our results suggest that similarity of individual FDG‐PET profiles to the limbic‐predominant FDG‐PET pattern was not related to severity of TDP‐43 pathology (BF10 = 0.449), presence of hippocampal sclerosis (BF10 = 0.544), or Braak NFT stages (BF10 = 0.525).Conclusion: Contrary to the expected effects, our current analysis indicates a lack of reliable associations between neuropathological markers such as TDP‐43 pathology, hippocampal sclerosis, or global NFT accumulation and the limbic‐predominant hypometabolic subtype. Future analyses will assess potential alternative contributors, such as region‐specific patterns of NFT distribution.

  • Open Access

    The limbic‐predominant hypometabolic subtype of AD is not enriched for comorbid TDP‐43 pathology or hippocampal sclerosis

    by Fedor Levin, Michel J. Grothe, Martin Dyrba, Stefan Teipel

    Background: Previous research has identified distinct subtypes of AD that differed with respect to the distribution of neurofibrillary tangles (NFT). One of them, a limbic‐predominant subtype, was enriched for comorbid TDP‐43 pathology. Our previous data‐driven analysis used FDG‐PET to identify a limbic‐predominant subtype of AD that may correspond to the limbic‐predominant NFT subtype. In the current study, we tested the expected associations between measures of neuropathology such as TDP‐43 pathology, hippocampal sclerosis, and Braak stages of NFT accumulation, and the limbic‐predominant hypometabolic subtype.Method: We analyzed ante‐mortem FDG‐PET and neuropathological data of 74 individuals from the autopsy cohort of the Alzheimer’s Disease Neuroimaging Initiative (ADNI), including 7 cognitively normal participants, 12 participants with MCI, and 55 with dementia (as assessed at the last available clinical evaluation). Severity of TDP‐43 pathology was operationalized as the number of brain areas with TDP‐43 immunoreactive inclusions (including amygdala, hippocampus, entorhinal/inferior temporal cortex and neocortex). Hippocampal sclerosis was operationalized as a binary variable. We measured similarity of individual FDG‐PET profiles to the limbic‐predominant subtype. For that, we used Euclidean distances from each individual profile to the mean profile of the limbic‐predominant AD subtype identified previously in a larger ADNI sample using a data‐driven hierarchical clustering approach (Fig.1). Lower Euclidean distances indicated higher similarity to the subtype. Bayesian linear regressions in JASP were used to assess associations between neuropathological markers and similarity to the limbic‐predominant FDG‐PET pattern, controlling for Thal phase of amyloid accumulation, age, sex, and time delay between the FDG‐PET scan and death in the null model. We assessed Bayes factors representing evidence for a given model in comparison to the null model (BF10).Result: Our results suggest that similarity of individual FDG‐PET profiles to the limbic‐predominant FDG‐PET pattern was not related to severity of TDP‐43 pathology (BF10 = 0.449), presence of hippocampal sclerosis (BF10 = 0.544), or Braak NFT stages (BF10 = 0.525).Conclusion: Contrary to the expected effects, our current analysis indicates a lack of reliable associations between neuropathological markers such as TDP‐43 pathology, hippocampal sclerosis, or global NFT accumulation and the limbic‐predominant hypometabolic subtype. Future analyses will assess potential alternative contributors, such as region‐specific patterns of NFT distribution.

  • Open Access

    Longitudinal trajectories of cognitive reserve in hypometabolic subtypes of Alzheimer's disease

    by Fedor Levin, Michel J. Grothe, Martin Dyrba, Nicolai Franzmeier, Stefan J. Teipel


  • Open Access

    Different levels of cognitive reserve among typical and limbic‐predominant hypometabolic subtypes of Alzheimer’s Disease

    by Fedor Levin, Michel J. Grothe, Martin Dyrba, Nicolai Franzmeier, Stefan Teipel

    Background: Residual approaches for estimating cognitive reserve (CR) are assumed to reflect relative resilience of individuals to neuropathological change in Alzheimer’s disease (AD). In the present study, we characterized differences in CR in people expressing distinct hypometabolic subtypes of AD‐related neurodegeneration.Method: We analyzed MRI, FDG‐PET and cognitive performance data of Aβ‐positive participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort, including 58 cognitively normal (CN), 217 prodromal AD and 177 AD dementia participants. Participants’ FDG‐PET scans were classified into the two principal hypometabolic subtypes of AD – typical and limbic‐predominant – as determined previously by a data‐driven clustering analysis. W‐scores were calculated as standardized residuals from a regression model with temporoparietal gray matter (GM) volume adjusted to total intracranial volume (TIV) as the dependent variable and with memory and executive function composite scores, as well as covariates age, gender, TIV and dementia status as independent variables (Fig. 1). The scores were multiplied by “‐1” so that higher values indicated lower than expected temporoparietal GM volume at a given level of cognitive performance and therefore represented higher CR. We evaluated the effects of W‐scores on the risk of clinical progression and longitudinal cognitive decline while including temporoparietal GM as a covariate. We then compared the W‐scores between the two hypometabolic subtypes of AD.Result: Higher W‐scores, indicating higher CR, predicted a lower risk of prodromal AD participants for the progression to AD dementia (hazards ratio = 0.32, 95%CI = [0.19;0.52]). W‐scores were also associated with a slower longitudinal cognitive decline for memory (β = 0.11,p<0.01) and language performance (β = 0.12,p = 0.017) in the subsample of CN and prodromal AD (Fig. 2). The typical hypometabolic subtype showed on average higher W‐scores than the limbic‐predominant subtype in prodromal AD and AD dementia groups.Conclusion: Consistent with previous research, CR as assessed by W‐scores was predictive of a lower risk of clinical progression and slower longitudinal cognitive decline. The typical subtype demonstrated higher W‐scores, which suggests a stronger resilience to the impact of neuropathological change on cognition. Future analysis will investigate longitudinal changes in W‐scores and factors contributing to the observed differences, as well as potential explanations for subtype differences.

  • In vivo amyloid staging in individuals with subjective cognitive decline in DELCODE Study

    by Fedor Levin, Michel J. Grothe, Martin Dyrba, Henning Boecker, Marcel Daamen, Catharina Lange, Katharina Buerger, Michael T. Heneka, Christoph Laske, Oliver Peters, et al.

    Background: In previous research, a data‐driven in vivo staging model of amyloid accumulation was established based on regional frequencies of amyloid positivity in 18F‐florbetapir‐PET scans of cognitively normal older adults. The model was subsequently validated in an independent cohort of participants with subjective memory complaints. In the current study we aim to apply this staging approach in an independent sample of individuals with subjective cognitive decline (SCD) in the multicenter DZNE ‐ Longitudinal Cognitive Impairment and Dementia Study (DELCODE).Method: PET scans with 18F‐florbetaben were obtained in a group of 63 SCD participants (mean age = 71.84y, range: 60.83‐83y; 35% female) from the DELCODE cohort. Participants also underwent MRI acquisition, clinical and neuropsychological assessments. CSF biomarkers and APOE genotypes were assessed for subsets of participants. In the future analysis we will classify participants into amyloid accumulation stages based on regional amyloid positivity as assessed in PET images.Result: Among participants with available CSF assessments, 38% had abnormal CSF Aβ42/Aβ40 ratio values indicating amyloid positivity. Frequency of APOE4 allele carriers of 39% was suggestive of APOE4 enrichment.Conclusion: SCD participants studied within the DELCODE represent a group at an increased risk for preclinical AD. Consistent with that, the current SCD subsample with available 18F‐florbetaben‐PET scans demonstrated heightened rates of abnormal CSF amyloid‐β biomarkers and APOE4 prevalence. Investigation of regional amyloid accumulation in this sample would help assess the validity of the previous in vivo amyloid staging model in preclinical AD, as well as further characterize SCD. In the future analysis we will apply the previously established amyloid staging approach. We will additionally evaluate the hypothesized links between amyloid accumulation stages, cognitive performance, CSF biomarkers and prevalence of APOE genotypes.