Researchers at Mass General Brigham and Harvard Medical School have developed a new AI model that can predict multiple brain disorders based on MRI scans. According to the researchers, the so-called foundation model, Brain Imaging Adaptive Core (BrainIAC), performs better than many AI models that are trained for a single specific task.
Artificial intelligence has long been used to recognize patterns in medical imaging. By analyzing MRI scans, algorithms can find clues to tumors, strokes, or neurodegenerative disorders, for example. However, most existing models are designed to detect only one disorder.
BrainIAC, as described in Nature Neuroscience, takes a different approach. The model was pre-trained on nearly 49,000 MRI scans of the brain and functions as a generic AI base model that can then be adapted for various clinical applications.
Self-learning system
For the training, the researchers used self-supervised learning, a method in which AI learns patterns from largely unannotated data. As a result, the model developed a broad knowledge of the structure and organization of the brain.
After this pre-training, BrainIAC can be relatively easily adapted for specific tasks, such as detecting or monitoring neurological disorders. In experiments, the model was able to detect or predict Alzheimer's disease, Parkinson's disease, autism spectrum disorder, stroke, dementia, and brain tumors with a high degree of accuracy.
Remarkably, the model sometimes required up to ten times less training data to deliver comparable performance to AI models developed specifically for a single condition.
New generation of medical AI
According to the researchers, BrainIAC can serve as a generic basis for AI analyses of brain imaging. In the future, the model could be further expanded with larger datasets or applied to other imaging modalities, such as CT scans, microscopy images, retinal scans, or ultrasound.
The model has been made available as open source, allowing other research groups to already use it for studies on Alzheimer's disease and traumatic brain injury, among other things. BrainIAC can thus contribute to a broader use of AI in neurological research and possibly also to earlier detection of brain disorders in clinical practice.
MRI antenna improves scan quality
Just recently, we wrote about a new developed MRI antenna that improves image quality and shortens scan times without requiring changes to existing MRI systems. The innovation, created by scientists at the Max Delbrück Center in collaboration with Rostock University Medical Center, enhances the strength and direction of radiofrequency signals used in MRI. Traditional RF coils often struggle to capture signals from deep or complex anatomical regions, resulting in lower image quality and longer scans.
By integrating metamaterials into the antenna, researchers can guide electromagnetic waves more efficiently, improving spatial resolution. In tests with volunteers, the technology enabled detailed imaging of the eye and eye socket at 7 Tesla. The compact antenna could also increase patient comfort and may support additional applications such as imaging other organs, improving scans near implants, and enhancing MRI-guided therapies.
