Evaluation of Site Harmonization Strategies for Normative CentileBrain Models
Poster No:
1123
Submission Type:
Abstract Submission
Authors:
RUIYANG GE1, Paul Thompson2, Sophia Frangou3
Institutions:
1University of British Columbia, Vancouver, BC, 2University of Southern California, Los Angeles, CA, 3Icahn School of Medicine at Mount Sinai, New York, NY
First Author:
Co-Author(s):
Introduction:
The ongoing research interest in normative modeling on neuroimaging phenotypes has highlighted the need for robust and publicly available models pre-trained on data from large samples of healthy individuals [1,2,3]. To address this need we have previously released a series of normative CentileBrain models for regional morphometry [4], based on multi-site neuroimaging data from over 80 across global datasets. Multi-site neuroimaging data are critical for capturing variability across diverse populations, but site-effects pose significant challenges. In this work, we aimed to evaluate different strategies for mitigating site effects in FreeSurfer-derived morphometric features while using CentileBrain for normative modeling.
Methods:
The sample used in this study (N=3,000) was collated from 25 research sites (Figure 1A), consisting solely of healthy individuals. The morphometric measures comprised cortical thickness of 68 cortical regions, cortical surface area of 68 cortical regions (Desikan-Killiany atlas) and of the volume of 14 subcortical regions (FreeSurfer Aseg atlas). We established a classification benchmark using linear support vector machines to evaluate the optimal solution for handling site. Specifically, we performed site classification-a multiclass problem where we aimed to identify 25 distinct sites using a one-versus-rest multiclass scheme. This analysis is particularly significant as it provides insight into the degree to which site-specific information is encoded in the data, serving as a proxy for the presence of site-effects.
We conducted this benchmark using six types of datasets (Figure 1B).
(1) The observed FreeSurfer features without ComBat-GAM [5] harmonization.
(2) The harmonized FreeSurfer features with ComBat-GAM harmonization.
Normative deviation values (i.e., Z-scores) extracted using CentileBrain with four different site-effect harmonization strategies:
(3) Non-harmonized CentileBrain Z-scores, derived from applying CentileBrain models to observed features of each site, separately.
(4) Harmonized unseen datasets with ComBat-GAM before entering into CentileBrain for Z-score computation. This is the current CentileBrain Option.
(5) Harmonization of each unseen dataset separately with a reference sample before CentileBrain Z-score computation.
(6) Harmonization of all unseen datasets together with a reference sample.
The reference sample used for harmonization was the ComBat-GAM-harmonized training sample of the CentileBrain models, consisting of 37,407 healthy individuals (age range=3-90 years) [4].
We conducted this benchmark using six types of datasets (Figure 1B).
(1) The observed FreeSurfer features without ComBat-GAM [5] harmonization.
(2) The harmonized FreeSurfer features with ComBat-GAM harmonization.
Normative deviation values (i.e., Z-scores) extracted using CentileBrain with four different site-effect harmonization strategies:
(3) Non-harmonized CentileBrain Z-scores, derived from applying CentileBrain models to observed features of each site, separately.
(4) Harmonized unseen datasets with ComBat-GAM before entering into CentileBrain for Z-score computation. This is the current CentileBrain Option.
(5) Harmonization of each unseen dataset separately with a reference sample before CentileBrain Z-score computation.
(6) Harmonization of all unseen datasets together with a reference sample.
The reference sample used for harmonization was the ComBat-GAM-harmonized training sample of the CentileBrain models, consisting of 37,407 healthy individuals (age range=3-90 years) [4].
Results:
Among the six different scenarios, we found that using unharmonized data resulted in high accuracy in site prediction, with balanced accuracies of 0.286 for the observed FreeSurfer data and 0.537 for the non-harmonized Z-scores. In contrast, for the harmonized data, the site prediction accuracies were much lower, ranging from 0.020 to 0.036, indicating a minimal site effect in the multisite dataset. The current CentileBrain option achieved a balanced accuracy of 0.033, which is comparable to random guessing (1/25 = 0.04).
Conclusions:
Site-effect harmonization is essential for multi-site datasets in normative modeling of brain morphometry. The current CentileBrain data harmonization option offers an optimal solution for addressing site effects in morphometric data.
Lifespan Development:
Aging
Lifespan Development Other
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Neuroinformatics and Data Sharing:
Workflows 2
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Data analysis
Modeling
Morphometrics
MRI
NORMAL HUMAN
STRUCTURAL MRI
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
2. Dima D, Modabbernia A, Papachristou E, et al. Subcortical volumes across the lifespan: Data from 18,605 healthy individuals aged 3-90 years. Human Brain Mapping. 2021; 43, 452-469. DOI:10.1002/hbm.25320
3. RAI Bethlehem, J Seidlitz, SR White, etc. Brain charts for the human lifespan. Nature. 2022; 604, 525-533. DOI: 10.1038/s41586-022-04554-y
4. Ge R, Yu Y, Qi Y, Haas S, etc, Thompson P, Frangou S. Normative Modeling of Brain Morphometry Across the Lifespan using CentileBrain: Algorithm Benchmarking and Model Optimization. Lancet Digit Health. 2024; 6, e211-e221. DOI: 10.1016/S2589-7500(23)00250-9
5. Pomponio, R Erus, G Habes, et al. Harmonization of large MRI datasets for the analysis of brain imaging patterns throughout the lifespan. Neuroimage. 2020; 208, 116450. DOI: 10.1016/j.neuroimage.2019.116450