🤖 AI Summary
Current public brain MRI datasets lack systematic evaluation in scale, diversity, and consistency, hindering foundation model generalizability. To address this, we conduct the first cross-dataset quantitative profiling study across 54 publicly available datasets (>530,000 scans), establishing a multi-level assessment framework spanning dataset-, image-, and feature-level analyses. Our methodology integrates modality distribution statistics, voxel spacing and intensity quantification, preprocessing pipeline variability analysis, and validation within the 3D DenseNet121 feature space. Results reveal critical issues: dominance of healthy controls, uneven disease coverage, substantial geometric and intensity heterogeneity, and residual covariate shift post-preprocessing. We identify the need for perceptually informed preprocessing combined with domain-adaptive modeling. This work provides a reproducible, structured evaluation paradigm to guide data curation and algorithm design for brain MRI foundation models.
📝 Abstract
The development of foundation models for brain MRI depends critically on the scale, diversity, and consistency of available data, yet systematic assessments of these factors remain scarce. In this study, we analyze 54 publicly accessible brain MRI datasets encompassing over 538,031 to provide a structured, multi-level overview tailored to foundation model development. At the dataset level, we characterize modality composition, disease coverage, and dataset scale, revealing strong imbalances between large healthy cohorts and smaller clinical populations. At the image level, we quantify voxel spacing, orientation, and intensity distributions across 15 representative datasets, demonstrating substantial heterogeneity that can influence representation learning. We then perform a quantitative evaluation of preprocessing variability, examining how intensity normalization, bias field correction, skull stripping, spatial registration, and interpolation alter voxel statistics and geometry. While these steps improve within-dataset consistency, residual differences persist between datasets. Finally, feature-space case study using a 3D DenseNet121 shows measurable residual covariate shift after standardized preprocessing, confirming that harmonization alone cannot eliminate inter-dataset bias. Together, these analyses provide a unified characterization of variability in public brain MRI resources and emphasize the need for preprocessing-aware and domain-adaptive strategies in the design of generalizable brain MRI foundation models.