Hierarchical Bayesian Regression for Multi-Site Normative Modeling of Neuroimaging Data

Kia, Seyed Mostafa; Huijsdens, Hester; Dinga, Richard; Wolfers, Thomas; Mennes, Maarten; Andreassen, Ole A.; Beckmann, Christian F.; Marquand, André F.

doi:10.48550/arxiv.2005.12055

Cited by 2 publications

(5 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Thompson et al (2014); Casey et al (2018); Bethlehem et al (2022)). This would then require data harmonisation, which represents a separate issue and active field of research on its own, as the application of normative models and normative ranges in real datasets should be adjusted to effectively deal with site-effects (Kia et al, 2020, 2021). To make our simulations realistic we used an age distribution based on the UK Biobank, the biggest single-study dataset currently available, which has been already used in several normative modelling studies (Nobis et al, 2019; Janahi et al, 2021).…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Normative models for neuroimaging markers: Impact of model selection, sample size and evaluation criteria

Božek¹,

Griffanti

Lau

et al. 2022

Preprint

View full text Add to dashboard Cite

Modelling population reference curves or normative modelling is increasingly used with the advent of large neuroimaging studies. In this paper we assess the performance of fitting methods from the perspective of clinical applications and investigate the influence of the sample size. Further, we evaluate linear and nonlinear models for percentile curve estimation and highlight how the bias-variance trade-off manifests in typical neuroimaging data. We created plausible ground truth distributions of hippocampal volumes in the age range of 45 to 80 years, as an example application. Based on these distributions we repeatedly simulated samples for sizes between 50 and 50,000 data points, and for each simulated sample we fitted a range of normative models. We compared the fitted models and their variability across repetitions to the ground truth, with specific focus on the outer percentiles (1th, 5th, 10th) as these are the most clinically relevant. Our results quantify the expected decreasing trend in variance of the volume estimates with increasing sample size. However, bias in the volume estimates only decreases a modest amount, without much improvement at large sample sizes. The uncertainty of model performance is substantial for what would often be considered large samples in a neuroimaging context and rises dramatically at the ends of the age range, where fewer data points exist. Flexible models perform better across sample sizes, especially for nonlinear ground truth. Surprisingly large samples of several thousand data points are needed to accurately capture outlying percentiles across the age range for applications in research and clinical settings. Performance evaluation methods should assess both, bias and variance. Furthermore, extreme caution is needed when attempting to extrapolate beyond the age range included in the source dataset. To help with such evaluations of normative models we have made our code available to guide researchers developing or utilising normative models.

show abstract

Section: Discussionmentioning

confidence: 99%

mentioning

confidence: 99%

Normative models for neuroimaging markers: Impact of model selection, sample size and evaluation criteria

Božek¹,

Griffanti

Lau

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…An automatic quality check procedure based on the Freesurfer Euler characteristic was run on all data and samples with a value higher than five were removed. [53][54][55][56][57][58] Normative Modeling…”

Section: Mri Data Acquisition and Analysismentioning

confidence: 99%

“…We estimated a normative model for each region of interest (ROI, n=150) in the Freesurfer Destrieux atlas 52 , using HBR with age as a covariate, and sex and scanner id as batch effects, to predict cortical thickness 45,54,55,58 . This accommodated multi-site pooling using transfer learning and comparisons across scanners 54,55,59 . The deviations from these models were then used as features in the linear mixed models outlined below.…”

Section: Mri Data Acquisition and Analysismentioning

confidence: 99%

“…To account for site and scanner effects we used the Hierarchical Bayesian regression (HBR) approach for normative modeling 54,55,59 , which efficiently accommodates inter-site variation and provides computational scaling, which is useful for multi-cohort and longitudinal studies with data from different scanners. We estimated a normative model for each region of interest (ROI, n=150) in the Freesurfer Destrieux atlas 52 , using HBR with age as a covariate, and sex and scanner id as batch effects, to predict cortical thickness 45,54,55,58 . This accommodated multi-site pooling using transfer learning and comparisons across scanners 54,55,59 .…”

Section: (Which Was Not Certified By Peer Review)mentioning

confidence: 99%

See 1 more Smart Citation

A 10-Year Longitudinal Study of Brain Cortical Thickness in People with First-Episode Psychosis using Normative Models

Berthet,

Haatveit,

Kjelkenes

et al. 2024

Preprint

Self Cite

View full text Add to dashboard Cite

Background: Clinical forecasting models have potential to optimize treatment and improve outcomes in psychosis, but predicting long-term outcomes is challenging and long-term follow up data are scarce. In this 10-year longitudinal study we aimed to characterize the temporal evolution of cortical correlates of psychosis and their associations with symptoms. Design: Structural MRI from people with first-episode psychosis and controls (n=79 and 218) were obtained at enrollment, after 12 months (n=67 and 197), and 10 years (n=23 and 77), within the Thematically Organized Psychosis (TOP) study. Normative models for cortical thickness estimated on public MRI datasets (n=42983) were applied to TOP data to obtain deviation scores for each region and timepoint. Positive And Negative Syndrome Scale (PANSS) scores were acquired at each timepoint along with registry data. Linear mixed effects (LME) models assessed effects of diagnosis, time and their interactions on cortical deviations plus associations with symptoms. Results: LMEs revealed main effects of diagnosis and time x diagnosis interactions in a distributed cortical network, where negative deviations in patients attenuate over time. In patients, symptoms also attenuate over time. LMEs revealed effects of anterior cingulate on PANSS total scores, and insular and orbitofrontal regions on PANSS negative scores. Conclusions: This long-term longitudinal study revealed a distributed pattern of cortical differences which attenuated over time together with a reduction in symptoms. These findings are not in line with a simple neurodegenerative account of schizophrenia, and deviations from normative models offer a promising avenue to develop biomarkers to track clinical trajectories over time.

show abstract

Hierarchical Bayesian Regression for Multi-Site Normative Modeling of Neuroimaging Data

Cited by 2 publications

References 0 publications

Normative models for neuroimaging markers: Impact of model selection, sample size and evaluation criteria

Normative models for neuroimaging markers: Impact of model selection, sample size and evaluation criteria

A 10-Year Longitudinal Study of Brain Cortical Thickness in People with First-Episode Psychosis using Normative Models

Contact Info

Product

Resources

About