Esten Høyland Leonardsen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Esten Høyland Leonardsen has a total of 117 co-authors. Here are the top co-authors by number of shared publications:
- Lars Tjelta Westlye
University of Oslo
14 shared publications - Ole Andreas Andreassen
Institute of Clinical Medicine, University of Oslo, Oslo, Norway
10 shared publications - Yunpeng Wang
University of Oslo
6 shared publications - Ann-Marie Glasø de Lange
Department of Psychology University of Oslo Oslo Norway
5 shared publications - Didac Vidal-Piñeiro
Center for Lifespan Changes in Brain and Cognition, University of Oslo, Oslo, Norway
5 shared publications - Geir Selbæk
Norwegian National Centre for Ageing and Health Vestfold Hospital Trust Tønsberg Norway
5 shared publications
Latest Publications
- Open Access
- 31 October 2026
- Published by Springer Nature in Child Psychiatry & Human Development
- p. 1-13
Longitudinal Patterns of Irritability During the Transition to Adolescence: Examining Caregiver and Youth perspectives and Sex Differences
- 31 October 2026
- Published by Springer Nature in Child Psychiatry & Human Development
- p. 1-13
Open AccessLongitudinal Patterns of Irritability During the Transition to Adolescence: Examining Caregiver and Youth perspectives and Sex Differences
- Open Access
Latent neural network representations of the brain reflect broad-scale adolescent phenotypic variation
The adolescent brain is attuned to social and environmental exploration, allowing behavioral adaptation as experiences shape lasting patterns of morphological organization. Using a convolutional neural network on longitudinal structural MRI data, we assess the early part of this developmental window and derive latent brain representations reflecting patterns of structural variability linked to personal, social, and neighborhood conditions in adolescence. These representations offer a flexible framework for mapping brain-trait associations in adolescence and beyond.
Open AccessLatent neural network representations of the brain reflect broad-scale adolescent phenotypic variation
The adolescent brain is attuned to social and environmental exploration, allowing behavioral adaptation as experiences shape lasting patterns of morphological organization. Using a convolutional neural network on longitudinal structural MRI data, we assess the early part of this developmental window and derive latent brain representations reflecting patterns of structural variability linked to personal, social, and neighborhood conditions in adolescence. These representations offer a flexible framework for mapping brain-trait associations in adolescence and beyond.
- Open Access
Learning diverse and generic representations of the brain with large-scale multi-task pretraining
Large pretrained models developed and shared by actors with privileged access to data and compute have played a central role in the democratisation of deep learning in a range of domains. Here, we contribute to this endeavour in the field of neuroimaging, by compiling a large dataset of structural magnetic resonance imaging scans (n=114,257) and using them to pretrain a multi-task convolutional neural network to predict age, sex, handedness, BMI, fluid intelligence and neuroticism. Subsequent analyses show that our pretraining approach results in a rich and diverse feature space, implying the model is sensitive to a broad spectrum of nuanced neuroanatomical variation. We also demonstrate the usefulness of the pretrained model by employing it to predict brain age and perform clinical classifications via transfer learning, showing that it consistently outperforms models trained from scratch.
Open AccessLearning diverse and generic representations of the brain with large-scale multi-task pretraining
Large pretrained models developed and shared by actors with privileged access to data and compute have played a central role in the democratisation of deep learning in a range of domains. Here, we contribute to this endeavour in the field of neuroimaging, by compiling a large dataset of structural magnetic resonance imaging scans (n=114,257) and using them to pretrain a multi-task convolutional neural network to predict age, sex, handedness, BMI, fluid intelligence and neuroticism. Subsequent analyses show that our pretraining approach results in a rich and diverse feature space, implying the model is sensitive to a broad spectrum of nuanced neuroanatomical variation. We also demonstrate the usefulness of the pretrained model by employing it to predict brain age and perform clinical classifications via transfer learning, showing that it consistently outperforms models trained from scratch.
- Open Access
Cortical thickness changes precede high levels of amyloid by at least seven years
Alzheimer’s disease (AD) is now defined based on its underlying brain pathology 1 , with the presence of amyloid (Aβ) plaques at high enough levels sufficient to warrant a diagnosis in the absence of cognitive symptoms. High levels of PET-detectable Aβ are widely thought to be the first imaging marker, with structural brain changes detectable on MRI scans thought to occur later. We combined 4570 longitudinal MRIs and 1684 Aβ PET scans from three cognitively healthy cohorts to test the difference in cortical thickness and its change between those that subsequently converted to be Aβ-positive or stayed Aβ-negative, using MRIs acquired exclusively in the years before conversion. We found those that subsequently developed elevated Aβ levels show both thicker cortex and less cortical thinning, even when the last MRI used to estimate their thickness trajectories was acquired at least seven years before conversion. Many effects remained when accounting for quantitative Aβ levels, suggesting some cortical thickness effects may be partly independent of Aβ. Differences in cortical thickness and its change between converters and Aβ-negative individuals showed moderate alignment with patterns of Aβ deposition, and the timing of thickness changes tracked the temporal progression of Aβ accumulation. Thus, if amyloid is AD 1 , we show that high levels of PET-detectable amyloid are not the first imaging marker of AD, as cortical thickness changes can be traced years before pathological amyloid. This has implications for understanding the sequence of events leading up to the earliest stages of AD.
Open AccessCortical thickness changes precede high levels of amyloid by at least seven years
Alzheimer’s disease (AD) is now defined based on its underlying brain pathology 1 , with the presence of amyloid (Aβ) plaques at high enough levels sufficient to warrant a diagnosis in the absence of cognitive symptoms. High levels of PET-detectable Aβ are widely thought to be the first imaging marker, with structural brain changes detectable on MRI scans thought to occur later. We combined 4570 longitudinal MRIs and 1684 Aβ PET scans from three cognitively healthy cohorts to test the difference in cortical thickness and its change between those that subsequently converted to be Aβ-positive or stayed Aβ-negative, using MRIs acquired exclusively in the years before conversion. We found those that subsequently developed elevated Aβ levels show both thicker cortex and less cortical thinning, even when the last MRI used to estimate their thickness trajectories was acquired at least seven years before conversion. Many effects remained when accounting for quantitative Aβ levels, suggesting some cortical thickness effects may be partly independent of Aβ. Differences in cortical thickness and its change between converters and Aβ-negative individuals showed moderate alignment with patterns of Aβ deposition, and the timing of thickness changes tracked the temporal progression of Aβ accumulation. Thus, if amyloid is AD 1 , we show that high levels of PET-detectable amyloid are not the first imaging marker of AD, as cortical thickness changes can be traced years before pathological amyloid. This has implications for understanding the sequence of events leading up to the earliest stages of AD.
- Open Access
Mapping Cerebellar Morphology in 15q11.2 CNV Carriers Using Normative Modeling
Copy number variations (CNVs) at the 15q11.2 locus of the human genome have been associated with altered brain structure and increased risk for neurodevelopmental and neuropsychiatric disorders. The cerebellum is increasingly seen as a crucial brain region for neurodevelopmental conditions, yet the effects of 15q11.2 CNVs on cerebellar morphology remain largely unclear. Importantly, 15q11.2 CNVs shows reduced or incomplete penetrance (meaning that not all CNV carriers are affected) and variable expressivity (meaning that symptoms may differ between individuals with the same genetic alteration). Thus, there is a need to not only assess group differences, but also to quantify anatomical variability at the individual level. Here, we address these issues using normative models of brain anatomy trained on large datasets (n > 52k, age range: 3-85) to assess both group and individual-level deviations in cerebellar anatomy in carriers of 15q11.2 deletions (n = 120, mean [SD] age= 64.95 [7.58]) and duplications (n = 149, mean [SD] age=64.31 [7.21]), compared to non-carriers (n = 19,028, mean [SD] age=64.31 [7.58]). Group-level case-control analyses revealed significantly smaller total and regional cerebellar volumes in both deletion and duplication carriers, though with small effect sizes. Individual-level deviation analyses, capturing pronounced alterations in specific individuals, revealed a heterogeneous pattern among carriers. Overall, our findings suggest that CNVs at the 15q11.2 locus exert modest and highly individualized effects on cerebellar morphology.
Open AccessMapping Cerebellar Morphology in 15q11.2 CNV Carriers Using Normative Modeling
Copy number variations (CNVs) at the 15q11.2 locus of the human genome have been associated with altered brain structure and increased risk for neurodevelopmental and neuropsychiatric disorders. The cerebellum is increasingly seen as a crucial brain region for neurodevelopmental conditions, yet the effects of 15q11.2 CNVs on cerebellar morphology remain largely unclear. Importantly, 15q11.2 CNVs shows reduced or incomplete penetrance (meaning that not all CNV carriers are affected) and variable expressivity (meaning that symptoms may differ between individuals with the same genetic alteration). Thus, there is a need to not only assess group differences, but also to quantify anatomical variability at the individual level. Here, we address these issues using normative models of brain anatomy trained on large datasets (n > 52k, age range: 3-85) to assess both group and individual-level deviations in cerebellar anatomy in carriers of 15q11.2 deletions (n = 120, mean [SD] age= 64.95 [7.58]) and duplications (n = 149, mean [SD] age=64.31 [7.21]), compared to non-carriers (n = 19,028, mean [SD] age=64.31 [7.58]). Group-level case-control analyses revealed significantly smaller total and regional cerebellar volumes in both deletion and duplication carriers, though with small effect sizes. Individual-level deviation analyses, capturing pronounced alterations in specific individuals, revealed a heterogeneous pattern among carriers. Overall, our findings suggest that CNVs at the 15q11.2 locus exert modest and highly individualized effects on cerebellar morphology.
- Open Access
Predicting Mental and Neurological Illnesses Based on Cerebellar Normative Features
Mental and neurological conditions have been linked to structural brain variations. However, aside from dementia, the value of brain structural characteristics derived from brain scans for prediction is relatively low. One reason for this limitation is the clinical and biological heterogeneity inherent to such conditions. Recent studies have implicated aberrations in the cerebellum – a relatively understudied brain region – in these clinical conditions. Here, we used machine learning to test the value of individual deviations from normative cerebellar development across the lifespan (based on trained data from >27k participants) for prediction of autism spectrum disorder (ASD) (n=317), bipolar disorder (BD) (n=238), schizophrenia (SZ) (n=195), mild cognitive impairment (MCI) (n=122), and Alzheimer’s disease (AD) (n=116). We applied several atlases and derived median, variance, and percentages of extreme deviations within each region of interest. Our results show that lobular and voxel-wise cerebellar data can be used to discriminate healthy controls from ASD and SZ with moderate accuracy (the area under the receiver operating characteristic curves ranged from 0.56 to 0.64), The strongest contributions to these predictive models were from posterior regions of the cerebellum, which are more strongly linked to higher cognitive functions than to motor control.
Open AccessPredicting Mental and Neurological Illnesses Based on Cerebellar Normative Features
Mental and neurological conditions have been linked to structural brain variations. However, aside from dementia, the value of brain structural characteristics derived from brain scans for prediction is relatively low. One reason for this limitation is the clinical and biological heterogeneity inherent to such conditions. Recent studies have implicated aberrations in the cerebellum – a relatively understudied brain region – in these clinical conditions. Here, we used machine learning to test the value of individual deviations from normative cerebellar development across the lifespan (based on trained data from >27k participants) for prediction of autism spectrum disorder (ASD) (n=317), bipolar disorder (BD) (n=238), schizophrenia (SZ) (n=195), mild cognitive impairment (MCI) (n=122), and Alzheimer’s disease (AD) (n=116). We applied several atlases and derived median, variance, and percentages of extreme deviations within each region of interest. Our results show that lobular and voxel-wise cerebellar data can be used to discriminate healthy controls from ASD and SZ with moderate accuracy (the area under the receiver operating characteristic curves ranged from 0.56 to 0.64), The strongest contributions to these predictive models were from posterior regions of the cerebellum, which are more strongly linked to higher cognitive functions than to motor control.
- 22 August 2024
- Published by Springer Nature in Nature Mental Health
- Vol. 2 (10) , 1196-1207
Mapping cerebellar anatomical heterogeneity in mental and neurological illnesses
- 22 August 2024
- Published by Springer Nature in Nature Mental Health
- Vol. 2 (10) , 1196-1207
Mapping cerebellar anatomical heterogeneity in mental and neurological illnesses
- Open Access
- 4 July 2024
- Published by Center for Open Science in PsyArXiv
Predicting and Understanding Vaccine Hesitancy in Adults: Insights into Modifiable Risk Factors from a Mixed Methods Multiple Population Study Combining Machine Learning and Thematic Analysis
BackgroundVaccine hesitancy, the delay in acceptance or reluctance to vaccinate, ranks among the top threats to global health and undermines efforts to control vaccine-preventable diseases. The identification of modifiable risk factors and psychological processes underlying vaccine hesitancy is imperative to inform targeted interventions aimed at increasing vaccination uptake.MethodsLeveraging gradient boosting machines and thematic analysis, this mixed methods multiple population study aimed to identify modifiable predictors of vaccine hesitance in the general adult population. Predictors of vaccine hesitance was investigated in 2926 Norwegian adults (Mean age = 37.91, 79.69% female) during the COVID-19 pandemic, before the predictive utility of these variables was investigated in an independent sample of 734 adults in the United Kingdom (Mean age = 40.34, 57.08% female), enabling an examination of generalizability across populations. Two teams of authors independently conducted the machine learning and thematic analyses, blind to the analytic procedure and results of each other.ResultsThe machine learning model performed well in discerning vaccine hesitant (n = 248, 8.48% and n = 109, 14.85%, Norway and UK, respectively) from vaccine uptaking individuals (n = 2678, 91.52% and n = 625, 85.15%), achieving an AUC of 0.94 (sensitivity = 0.81; specificity = 0.98) in the Norwegian sample, and an AUC of 0.98 (sensitivity = 0.83; specificity = 0.97) in the out-of-sample replication in the UK. The mixed methods investigation identified five categories of modifiable risk factors and psychological processes tied to vaccine hesitancy, including illusion of invulnerability, doubts about vaccine efficacy, mistrust in official entities, minimization of the societal impact of COVID-19, and health-related fears tied to vaccination. Moreover, the media’s portrayal of rare incidents as fear amplifiers, and stigmatizing presentation of unvaccinated individuals were provided as additional underlying motives leading to vaccine reluctance and polarization. The thematic analysis further revealed information overload, fear of needles, previous negative vaccination experiences, fear of not getting healthcare follow-up after vaccination if needed, and vaccine aversion due to underlying (psychiatric) illness (e.g., eating disorders) as motives underlying vaccine hesitance.ConclusionsThe identified influential predictors were consistent across the two European samples, highlighting their generalizability across European populations. These predictors offer insights about modifiable factors that could be adapted by public health campaigns in mitigating misconceptions and fears related to vaccination toward increasing vaccine uptake. Moreover, the results highlight the media’s responsibility, as mediators of the public perception of vaccines, to minimize polarization and provide accurate portrayals of rare vaccine-related incidents, reducing the risk aggravating fear and reactance to vaccination.
- 4 July 2024
- Published by Center for Open Science in PsyArXiv
Open AccessPredicting and Understanding Vaccine Hesitancy in Adults: Insights into Modifiable Risk Factors from a Mixed Methods Multiple Population Study Combining Machine Learning and Thematic Analysis
BackgroundVaccine hesitancy, the delay in acceptance or reluctance to vaccinate, ranks among the top threats to global health and undermines efforts to control vaccine-preventable diseases. The identification of modifiable risk factors and psychological processes underlying vaccine hesitancy is imperative to inform targeted interventions aimed at increasing vaccination uptake.MethodsLeveraging gradient boosting machines and thematic analysis, this mixed methods multiple population study aimed to identify modifiable predictors of vaccine hesitance in the general adult population. Predictors of vaccine hesitance was investigated in 2926 Norwegian adults (Mean age = 37.91, 79.69% female) during the COVID-19 pandemic, before the predictive utility of these variables was investigated in an independent sample of 734 adults in the United Kingdom (Mean age = 40.34, 57.08% female), enabling an examination of generalizability across populations. Two teams of authors independently conducted the machine learning and thematic analyses, blind to the analytic procedure and results of each other.ResultsThe machine learning model performed well in discerning vaccine hesitant (n = 248, 8.48% and n = 109, 14.85%, Norway and UK, respectively) from vaccine uptaking individuals (n = 2678, 91.52% and n = 625, 85.15%), achieving an AUC of 0.94 (sensitivity = 0.81; specificity = 0.98) in the Norwegian sample, and an AUC of 0.98 (sensitivity = 0.83; specificity = 0.97) in the out-of-sample replication in the UK. The mixed methods investigation identified five categories of modifiable risk factors and psychological processes tied to vaccine hesitancy, including illusion of invulnerability, doubts about vaccine efficacy, mistrust in official entities, minimization of the societal impact of COVID-19, and health-related fears tied to vaccination. Moreover, the media’s portrayal of rare incidents as fear amplifiers, and stigmatizing presentation of unvaccinated individuals were provided as additional underlying motives leading to vaccine reluctance and polarization. The thematic analysis further revealed information overload, fear of needles, previous negative vaccination experiences, fear of not getting healthcare follow-up after vaccination if needed, and vaccine aversion due to underlying (psychiatric) illness (e.g., eating disorders) as motives underlying vaccine hesitance.ConclusionsThe identified influential predictors were consistent across the two European samples, highlighting their generalizability across European populations. These predictors offer insights about modifiable factors that could be adapted by public health campaigns in mitigating misconceptions and fears related to vaccination toward increasing vaccine uptake. Moreover, the results highlight the media’s responsibility, as mediators of the public perception of vaccines, to minimize polarization and provide accurate portrayals of rare vaccine-related incidents, reducing the risk aggravating fear and reactance to vaccination.
- Open Access
- 2 May 2024
- Published by Springer Nature in npj Digital Medicine
- Vol. 7 (1) , 1-14
Constructing personalized characterizations of structural brain aberrations in patients with dementia using explainable artificial intelligence
Deep learning approaches for clinical predictions based on magnetic resonance imaging data have shown great promise as a translational technology for diagnosis and prognosis in neurological disorders, but its clinical impact has been limited. This is partially attributed to the opaqueness of deep learning models, causing insufficient understanding of what underlies their decisions. To overcome this, we trained convolutional neural networks on structural brain scans to differentiate dementia patients from healthy controls, and applied layerwise relevance propagation to procure individual-level explanations of the model predictions. Through extensive validations we demonstrate that deviations recognized by the model corroborate existing knowledge of structural brain aberrations in dementia. By employing the explainable dementia classifier in a longitudinal dataset of patients with mild cognitive impairment, we show that the spatially rich explanations complement the model prediction when forecasting transition to dementia and help characterize the biological manifestation of disease in the individual brain. Overall, our work exemplifies the clinical potential of explainable artificial intelligence in precision medicine.
- 2 May 2024
- Published by Springer Nature in npj Digital Medicine
- Vol. 7 (1) , 1-14
Open AccessConstructing personalized characterizations of structural brain aberrations in patients with dementia using explainable artificial intelligence
Deep learning approaches for clinical predictions based on magnetic resonance imaging data have shown great promise as a translational technology for diagnosis and prognosis in neurological disorders, but its clinical impact has been limited. This is partially attributed to the opaqueness of deep learning models, causing insufficient understanding of what underlies their decisions. To overcome this, we trained convolutional neural networks on structural brain scans to differentiate dementia patients from healthy controls, and applied layerwise relevance propagation to procure individual-level explanations of the model predictions. Through extensive validations we demonstrate that deviations recognized by the model corroborate existing knowledge of structural brain aberrations in dementia. By employing the explainable dementia classifier in a longitudinal dataset of patients with mild cognitive impairment, we show that the spatially rich explanations complement the model prediction when forecasting transition to dementia and help characterize the biological manifestation of disease in the individual brain. Overall, our work exemplifies the clinical potential of explainable artificial intelligence in precision medicine.
- Open Access
- 15 April 2024
- Published by Wiley in Human Brain Mapping
- Vol. 45 (6) , e26685
Dissecting unique and common variance across body and brain health indicators using age prediction
Ageing is a heterogeneous multisystem process involving different rates of decline in physiological integrity across biological systems. The current study dissects the unique and common variance across body and brain health indicators and parses inter‐individual heterogeneity in the multisystem ageing process. Using machine‐learning regression models on the UK Biobank data set (N = 32,593, age range 44.6–82.3, mean age 64.1 years), we first estimated tissue‐specific brain age for white and gray matter based on diffusion and T1‐weighted magnetic resonance imaging (MRI) data, respectively. Next, bodily health traits, including cardiometabolic, anthropometric, and body composition measures of adipose and muscle tissue from bioimpedance and body MRI, were combined to predict ‘body age’. The results showed that the body age model demonstrated comparable age prediction accuracy to models trained solely on brain MRI data. The correlation between body age and brain age predictions was 0.62 for the T1 and 0.64 for the diffusion‐based model, indicating a degree of unique variance in brain and bodily ageing processes. Bayesian multilevel modelling carried out to quantify the associations between health traits and predicted age discrepancies showed that higher systolic blood pressure and higher muscle‐fat infiltration were related to older‐appearing body age compared to brain age. Conversely, higher hand‐grip strength and muscle volume were related to a younger‐appearing body age. Our findings corroborate the common notion of a close connection between somatic and brain health. However, they also suggest that health traits may differentially influence age predictions beyond what is captured by the brain imaging data, potentially contributing to heterogeneous ageing rates across biological systems and individuals.
- 15 April 2024
- Published by Wiley in Human Brain Mapping
- Vol. 45 (6) , e26685
Open AccessDissecting unique and common variance across body and brain health indicators using age prediction
Ageing is a heterogeneous multisystem process involving different rates of decline in physiological integrity across biological systems. The current study dissects the unique and common variance across body and brain health indicators and parses inter‐individual heterogeneity in the multisystem ageing process. Using machine‐learning regression models on the UK Biobank data set (N = 32,593, age range 44.6–82.3, mean age 64.1 years), we first estimated tissue‐specific brain age for white and gray matter based on diffusion and T1‐weighted magnetic resonance imaging (MRI) data, respectively. Next, bodily health traits, including cardiometabolic, anthropometric, and body composition measures of adipose and muscle tissue from bioimpedance and body MRI, were combined to predict ‘body age’. The results showed that the body age model demonstrated comparable age prediction accuracy to models trained solely on brain MRI data. The correlation between body age and brain age predictions was 0.62 for the T1 and 0.64 for the diffusion‐based model, indicating a degree of unique variance in brain and bodily ageing processes. Bayesian multilevel modelling carried out to quantify the associations between health traits and predicted age discrepancies showed that higher systolic blood pressure and higher muscle‐fat infiltration were related to older‐appearing body age compared to brain age. Conversely, higher hand‐grip strength and muscle volume were related to a younger‐appearing body age. Our findings corroborate the common notion of a close connection between somatic and brain health. However, they also suggest that health traits may differentially influence age predictions beyond what is captured by the brain imaging data, potentially contributing to heterogeneous ageing rates across biological systems and individuals.