Researchers have developed SCAN, a novel framework for adapting vision-language models (VLMs) for neurological disease classification without requiring the sharing of sensitive patient MRI scans. This site-aware collaborative approach allows new medical sites to fine-tune existing VLMs using differential privacy, enabling them to contribute to a global diagnostic foundation model while maintaining patient privacy. The framework has demonstrated effectiveness in improving classification accuracy for conditions like Alzheimer's disease and epilepsy across multiple international sites, even with privacy-preserving data summaries. AI
IMPACT Enables broader collaboration and data sharing for medical AI research while preserving patient privacy.
RANK_REASON Academic paper detailing a new method for adapting AI models. [lever_c_demoted from research: ic=1 ai=1.0]
- Alzheimer's disease
- differential privacy
- epilepsy
- Neurological Disease Classification
- stroke
- vision-language models
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →