Meet NeuroVFM: A New Neuroimaging Foundation Model Trained With Vol-JEPA on Uncurated Clinical MRI and CT Volumes
Researchers at the University of Michigan unveiled NeuroVFM, a groundbreaking AI model trained on over 5 million uncurated MRI and CT scans to learn brain anatomy and pathology without relying on radiology reports.
- NeuroVFM is the first generalist neuroimaging foundation model trained on 5.24 million uncurated MRI and CT volumes without radiology-report labels.
- The Vol-JEPA framework extends I-JEPA and V-JEPA to volumetric medical imaging, enabling self-supervised learning from raw scan data.
- The model demonstrates strong generalization across imaging modalities and tasks, reducing reliance on manual annotations.
- Researchers at the University of Michigan led the development, positioning NeuroVFM as a potential backbone for future medical AI tools.
A team at the University of Michigan has introduced NeuroVFM, a neuroimaging foundation model designed to process volumetric medical imaging data such as MRI and CT scans. Unlike previous models that depend on labeled radiology reports, NeuroVFM leverages the Vol-JEPA framework, an extension of I-JEPA and V-JEPA, to learn directly from uncurated clinical volumes. The model was trained on a massive dataset of 5.24 million scans, enabling it to capture intricate details of brain anatomy and pathology without explicit supervision.
The innovation lies in its ability to generalize across different imaging modalities and tasks, potentially reducing the need for manual annotation in medical AI applications. This approach could streamline the development of diagnostic tools and improve accessibility in regions with limited radiology expertise. The researchers highlight that NeuroVFM’s performance on downstream tasks suggests it could serve as a versatile backbone for future neuroimaging AI systems.
Provides a new self-supervised framework for medical imaging AI, reducing annotation costs and enabling more scalable model training.
Offers a foundation for building diagnostic tools with lower data labeling requirements, potentially accelerating product development in healthcare AI.
Highlights a breakthrough in medical AI with broad commercial potential, particularly in regions with limited radiology resources.
Introduces a novel approach to self-supervised learning in medical imaging, relevant for AI and healthcare research.
Could improve accessibility to advanced diagnostic tools by reducing dependence on labeled medical data.
- Vol-JEPA
- A volumetric extension of the Joint Embedding Predictive Architecture (JEPA) framework, adapted for 3D medical imaging tasks.
- Foundation model
- A large AI model pre-trained on vast amounts of data, designed to be fine-tuned for various downstream tasks.
From NASA to the classroom: The engineer bringing AI to those left behind - UN News
The dark side of ‘artificial intelligence’: In conversation with Paul Tremblay - The Oxford Student
AI ResearchI Built An AI That Roasts Me. It Hurt.
AI ResearchAgainst Usefulness
AI ResearchLinkedIn is the undisputed king of long-form AI slop, according to a study spanning five platforms
Nvidia is getting more competition. Here's what that actually means - Yahoo Finance
Nvidia's dominance in AI chips is being challenged by new competitors, reshaping the high-performance computing landscape.
Turkcell Builds Artificial Intelligence Infrastructure Across Five Strategic Technology Layers - The Fast Mode
Turkcell has built an AI infrastructure across five strategic technology layers, a move that could enhance its services and competitiveness.
Apple (AAPL) Sues OpenAI Over AI Hardware Trade Secrets - Yahoo! Finance Canada
Apple has filed a lawsuit against OpenAI, accusing the company of stealing trade secrets related to AI hardware development.
Apple bites OpenAI with lawsuit - Boston Herald
Apple has filed a lawsuit against OpenAI, escalating legal disputes in the AI sector.
BusinessApple sues OpenAI alleging trade secret theft, says scheme was 'at every level'
Apple has filed a lawsuit against OpenAI, accusing the company of stealing trade secrets across multiple levels of its operations.
Singapore proposes mandatory notification when firms use personal data for AI training - The Straits Times
Singapore’s data protection authority is proposing new rules that would require companies to notify individuals when their personal data is used to train AI models.