Frozen Brain-MRI Foundation Models Are Site Fingerprints

Foundation models provide powerful representations for brain MRI analysis, but their predictions remain difficult to interpret in anatomically meaningful terms. Clinical assessment of brain MRI is commonly organized around anatomically defined structures and regional abnormalities, whereas conventional explanation methods typically produce voxel- or patch-level importance maps that do not explicitly quantify the contributions of individual brain regions. To address this mismatch, we propose RegionFM, an interpretable framework that integrates anatomical segmentation with brain MRI foundation-model embeddings. RegionFM first divides each MRI scan into anatomical regions and constructs a separate MRI volume for each region. A frozen foundation model then encodes each region into an embedding, and a region-additive logistic model combines these embeddings such that every anatomical region contributes an explicit scalar term to the final prediction. This formulation supports both subject-level and cohort-level analyses of regional contributions. We evaluate RegionFM on cognitive-impairment classification using embeddings from multiple pretrained brain MRI foundation models. The results show that RegionFM maintains performance comparable to less interpretable fine-tuning approaches while providing anatomically grounded explanations. Randomized embedding ablations yield near-chance performance, indicating that the predictions rely on meaningful structure captured by the foundation-model embeddings rather than simple feature statistics. Overall, RegionFM better aligns model explanations with anatomy-based clinical reasoning while maintaining competitive predictive performance.

A foundation encoder is pretrained once on a large image corpus and then reused with its weights frozen. Each new task is solved by training a small head on the features it produces. This setup is common in medical imaging, where labeled cases are scarce and a frozen encoder can be reused across findings. All downstream tasks then depend on the class separation present in that fixed feature space. Encoder selection usually uses the area under the receiver operating characteristic curve (AUROC) of a trained downstream head. AUROC measures the predictive information a head can extract from the features, but it does not measure class separation in the frozen feature space. A positive image that lies near a negative image in feature space has a bounded normalized margin under any Lipschitz head. Chance-calibrated neighborhood discordance (CANDOR) measures this feature-space separation without training a head. For a positive image, it compares the k nearest positive-label neighbors with the k nearest negative-label neighbors, among images acquired the same way. Discordance rate D is the share of positives whose opposite-label neighbors are nearer. Drawing the 2 candidate sets at equal size makes the labels interchangeable, so label-independent features have chance level D=1/2 without simulation. We apply CANDOR to 22 frozen encoders on 605,443 images from 20 public datasets, covering 8 binary tasks in 7 domains. On every task, the best encoder is below 1/2. A discordant image bounds the normalized margin of every Lipschitz head on that encoder. A selector that is shown the true label and chooses among 11 encoders reduces the miss rate from 0.359 to 0.028. Discordance is associated with occlusion retention and with none of pretraining objective, parameter count, release year, or finding size. The fixed chance level lets D be computed for a frozen encoder before any downstream head is trained.

Brain MRI foundation models learn rich representations of anatomy, but interpreting what clinical information they encode remains an open problem. Standard sparse autoencoders (SAEs) suffer from severe feature collapse in deep transformer layers, and in Alzheimer's disease (AD) research, aging confounds nearly every clinical variable, making naive annotation unreliable. We propose GeoSAE, a geometry-guided SAE framework that uses the foundation model's learned manifold structure to prevent feature collapse and annotates each surviving feature via age-deconfounded partial correlations. Applied to ~14k T1-weighted MRI scans from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and the Australian Imaging biomarkers and Lifestyle (AIBL) datasets, GeoSAE identifies a compact, fully interpretable feature set that predicts mild cognitive impairment (MCI)-to-AD conversion (AUC 0.746) using only 2% of the embedding dimensions, while comorbidity-annotated features achieve only chance-level performance. The identified features replicate across cohorts without retraining (r=0.97) and localize to neuroanatomically distinct regions consistent with Braak staging. This shows that geometry-guided SAEs can extract interpretable, biomarkers from frozen brain MRI foundation models.