A Comparative Evaluation of Structural MRI Foundation Models for Age, Sex, and Body-Mass Index Predictions
- 1Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health
Abstract
Foundation models pre-trained on large neuroimaging datasets offer a promising approach to overcome the limited sample sizes typical of clinical imaging studies, yet their generalization across diverse populations remains unclear. We present the first systematic benchmark of four publicly available structural MRI foundation models: AnatCL, BrainIAC, 3D-Neuro-SimCLR, and SwinBrain. Using T1-weighted MRIs from the Parkinson's Progression Markers Initiative (PPMI), Healthy Brain Network (HBN), and Nathan Kline Institute (NKI) datasets, we evaluate these models on sex classification, brain age prediction, and body mass index prediction, comparing against models trained from FreeSurfer-derived cortical thickness and cortical surface area features. Submitted models are evaluated using a standardized frozen feature probing framework. The evaluation methods are available in BrainFMBench, a living benchmark for structural brain MRI foundation models hosted on GitHub, where new models can be added through pull requests. Although some foundation models outperformed FreeSurfer on particular tasks and datasets, 3D-Neuro-SimCLR and AnatCL outperformed the baselines overall, with 3D-Neuro-SimCLR demonstrating the most consistent performance (with the notable exception of HBN sex classification). The remaining models did not consistently outperform the baselines, indicating that the advantage of learned representations over morphometric features was not consistent across datasets and tasks for these models. In addition, cross-model feature correlation analysis reveals that foundation model representations correlate differently with traditional cortical measurements. These findings position structural MRI foundation models, particularly 3D-Neuro-SimCLR and AnatCL, as promising avenues to boost the performance of predictive models in neuroimaging.
Related
- OmniEEG-Bench: A Standardized Evaluation Benchmark for EEG Foundation Models
- CAMELIA: Enhancing species assemblage predictions by integrating structural and functional indices: A case study on plant communities in the French Alps
- Structural characteristics and policy differences of digital health management policies in China: A quantitative analysis based on the PMC index model
- A comparative evaluation of thiol/disulfide homeostasis and ischemia-modified albumin levels in patients with heroin use disorder in remission and detoxification
- Comparative evaluation of metabolomic, microbial, and integrated features for distinguishing a rat model of heatstroke death from cervical-dislocation controls and estimating postmortem interval across organs
BibTeX
@misc{encin2026comparative,
title = {{A Comparative Evaluation of Structural MRI Foundation Models for Age, Sex, and Body-Mass Index Predictions}},
author = {A. Encin and A. Gilmore and A. Rokem and E. Dickie and T. Glatard},
howpublished = {bioRxiv},
year = {2026},
month = may,
doi = {10.64898/2026.05.15.725427},
url = {https://www.biorxiv.org/content/10.64898/2026.05.15.725427v4},
note = {CC BY},
}
RIS
TY - UNPB
TI - A Comparative Evaluation of Structural MRI Foundation Models for Age, Sex, and Body-Mass Index Predictions
AU - A. Encin
AU - A. Gilmore
AU - A. Rokem
AU - E. Dickie
AU - T. Glatard
PY - 2026
DA - 2026/05/15
PB - bioRxiv
DO - 10.64898/2026.05.15.725427
UR - https://www.biorxiv.org/content/10.64898/2026.05.15.725427v4
AB - Foundation models pre-trained on large neuroimaging datasets offer a promising approach to overcome the limited sample sizes typical of clinical imaging studies, yet their generalization across diverse populations remains unclear. We present the first systematic benchmark of four publicly available structural MRI foundation models: AnatCL, BrainIAC, 3D-Neuro-SimCLR, and SwinBrain. Using T1-weighted MRIs from the Parkinson's Progression Markers Initiative (PPMI), Healthy Brain Network (HBN), and Nathan Kline Institute (NKI) datasets, we evaluate these models on sex classification, brain age prediction, and body mass index prediction, comparing against models trained from FreeSurfer-derived cortical thickness and cortical surface area features. Submitted models are evaluated using a standardized frozen feature probing framework. The evaluation methods are available in BrainFMBench, a living benchmark for structural brain MRI foundation models hosted on GitHub, where new models can be added through pull requests. Although some foundation models outperformed FreeSurfer on particular tasks and datasets, 3D-Neuro-SimCLR and AnatCL outperformed the baselines overall, with 3D-Neuro-SimCLR demonstrating the most consistent performance (with the notable exception of HBN sex classification). The remaining models did not consistently outperform the baselines, indicating that the advantage of learned representations over morphometric features was not consistent across datasets and tasks for these models. In addition, cross-model feature correlation analysis reveals that foundation model representations correlate differently with traditional cortical measurements. These findings position structural MRI foundation models, particularly 3D-Neuro-SimCLR and AnatCL, as promising avenues to boost the performance of predictive models in neuroimaging.
ER -