cs.CVNov 16, 2026

SCAN: Site-Aware Collaborative Privacy-Preserving Adaptation of Vision-Language Models for Neurological Disease Classification

Authors: Nikhil J. Dhinagar, Rishab Pendam, Harmony S. Cen, Sophia I. Thomopoulos, John P. John, Mahir H. Khan, Sook-Lei Liew, Erik Kaestner, 

Organizations: Imaging Genetics Center, Mark & Mary Stevens Neuroimaging & Informatics Institute, Keck School of Medicine, University of Southern California, Los Angeles, CA, USA · Multimodal Brain Image Analysis Laboratory, ADBS Neuroimaging Centre, National Institute of Mental Health and Neurosciences (NIMHANS), Bangalore, India · Neuroscience Graduate Program, Mark & Mary Stevens Neuroimaging & Informatics Institute, Chan Division of Occupational Science & Occupational Therapy, Biomedical Engineering, University of Southern California, Los Angeles, CA, USA · Department of Radiation Medicine and Applied Sciences, University of California, San Diego, San Diego, CA, USA

Abstract

Worldwide neuroimaging data could strengthen diagnostic foundation models, but privacy rules often prevent centralizing brain MRI scans. We present SCAN, a site-aware collaborative adaptation framework for Neurological Disease Classification that lets new sites adapt a 3D medical vision-language model (VLM) without retraining or sharing patient scans. Each site keeps exact class prototypes for local prediction and can share privacy-protected class summaries: norm-bounded sums and counts perturbed once with calibrated noise. Using 7,641 scans from 67 sites across 12 countries, covering Alzheimer's disease (AD), stroke, epilepsy and controls, we assess local adaptation, cross-site benefits, and effects on existing sites. At seven unseen sites, the frozen VLM's zero-shot AUCs were 0.833-0.945 for AD and stroke. Training-free local prototypes improved epilepsy AUC at three sites from 0.598/0.625/0.508 to 0.684/0.729/0.762 and AD AUC at one site from 0.833 to 0.863, nearly matching a classifier trained at each site (mean AUC 0.757 vs. 0.759). With differential privacy (εε = 4), new sites contributed without materially reducing accuracy at existing sites (AUC change ≤\leq0.2%), although gains from sharing were limited, and membership attacks performed near chance. SCAN offers hospitals a worldwide framework to adapt and share brain MRI VLMs without exchanging scans or retraining.

Explore similar work

Oct 20, 2026

Convergent Performance, Divergent Mechanisms in LoRA-Adapted Vision Foundation Models for Brain MRI

Adapting Vision Foundation Models (VFMs) to medical imaging via LoRA has become standard practice, yet model selection is guided almost entirely by downstream task metrics, treating backbones with similar performance as interchangeable. We investigate whether equal performance implies equal internal representational shift. We analyse four ViT-B/16 backbones, DINOv3, SigLIP2, MerMED-FM, and BiomedCLIP, pre-trained with different objectives (self-distillation vs. contrastive vision-language) and on different data domains (natural vs. medical images). We adapt each to brain tumour segmentation and evaluate on six datasets covering both segmentation and Alzheimer's disease classification. Using attention divergence, weight-change analysis, and activation patching, we find that the four models converge to similar segmentation performance (mean Dice 78-81%) through markedly different internal dynamics, yet diverge in cross-task transfer: segmentation adaptation improves Alzheimer's disease classification for DINOv3 but degrades it for the other three backbones. We reveal a dissociation between modification magnitude and causal importance: the most heavily modified layers are not the causally relevant ones, with early-to-mid layers identified as sufficient for segmentation recovery. Restricting LoRA to these layers matches or approaches full-layer segmentation while reducing classification degradation in the three affected backbones, providing a mechanistically grounded strategy that is more parameter-efficient and more robust than standard full-layer adaptation.
Oct 26, 2026

Evaluating Federated Vision-Language and Vision-Only Models for Privacy-Aware Breast Cancer Imaging

Artificial intelligence (AI) shows promise for breast imaging, but limited labeled data, institutional data fragmentation, and privacy concerns hinder centralized training. We investigate a privacy-aware federated learning framework combining pretrained vision-language models (VLMs) with client-level differential privacy. We compare general-purpose CLIP models, biomedical BiomedCLIP, and an ImageNet-pretrained ViT-B/16 baseline on two public breast imaging datasets, with and without differential privacy. We also evaluate a CLIP ViT-B/16 ablation that removes the text pathway. Evaluation includes patient-level ROC-AUC with bootstrap 95% confidence intervals, baseline comparisons, threshold sensitivity, inter-model agreement, and communication cost. Under federated learning, all five models achieve high normal-versus-sick discrimination (ROC-AUC: 0.940-0.990). However, acquisition year alone achieves 0.974 on the same test partition, raising concerns about confounding rather than genuine pathology recognition. Benign-versus-malignant discrimination remains uncertain (ROC-AUC: 0.504-0.637), with all confidence intervals including 0.50. Removing the text pathway preserves classification accuracy while halving communication cost. Differential privacy reduces ROC-AUC by 0.375, attributable to perturbing the full model state rather than only trainable parameters. These findings highlight the importance of evaluating confounding, predictive reliability, privacy implementation, and deployment efficiency when developing federated AI systems for breast imaging.
Sep 21, 2026

Evaluating the Generalization of Neuroimaging Foundation Models on African Brain MRI

Neuroimaging foundation models pretrained on large, predominantly western cohorts are increasingly proposed as general-purpose backbones for brain MRI analysis. Yet, their ability to generalize to underrepresented clinical populations remains largely untested. We evaluate four recent foundation models (BrainIAC, Neuro-JEPA, NeuroVFM, and Primus) on a three-way diagnostic classification task (Control, Dementia, Parkinson's disease) using a cohort of 88 subjects from a Nigerian clinical brain MRI dataset, across four modality configurations (T1w, T2w, T1w+T2w, FLAIR), and compare against an end-to-end trained ViT3D baseline. The frozen backbones collapse to majority-class predictions, while Neuro-JEPA on FLAIR shows modest but still limited discrimination. In contrast, the end-to-end trained ViT3D achieves higher accuracy and MCC on every task (up to 53.4% accuracy, MCC=0.27) and is the only model with non-trivial recall. Our findings suggest that these frozen neuroimaging foundation models are insufficient for fine-grained diagnostic classification in small, non-western clinical cohorts, motivating parameter-efficient adaptation and broader multi-site external validation for equitable deployment in global health settings.