Scientists from Harvard Medical School and Mass General Brigham have developed an innovative artificial intelligence system called BrainIAC (Brain Imaging Adaptive Core).
This is a so-called “foundation model” that processes data from millions of MRI scans and is capable of diagnosing various neuropsychiatric conditions faster and more accurately than previously existing specialized algorithms.
The study, published in the journal Nature Neuroscience, describes a system trained using a “self-supervised” learning method on nearly 50,000 brain MRI scans. Unlike older models created specifically to identify a single disease, BrainIAC possesses a general baseline knowledge of brain structure. This allows it to recognize a wide spectrum of diseases after only minimal additional training.
What is the technology capable of?
The BrainIAC system is distinguished by its exceptional versatility, enabling it to predict complex conditions such as Alzheimer’s disease, autism, dementia, brain tumors, Parkinson’s, and stroke with high precision.
The model’s efficiency is further evidenced by the fact that it requires 10 times less data to learn new tasks compared to traditional AI, significantly accelerating the research process.
Furthermore, the system performs in-depth analysis and identifies hidden patterns in MRI scans that are imperceptible to the human eye. This capability allows clinicians to accurately determine the mutational status of tumors and predict future risks of disease development.
The BrainIAC algorithm is open-source and available to researchers worldwide. Scientists anticipate that similar “foundation models” will be developed in the future for imaging other organs (CT, ultrasound, retinal scans), completely transforming the standards of early diagnosis.
