AI Model Predicts Brain Disorders from MRI Scans | Medical Xpress

A new artificial intelligence platform, called BrainIAC (Brain Imaging Adaptive Core), is demonstrating a remarkable ability to extract multiple risk signals from routine brain MRI scans. Researchers have shown it can estimate a person’s “brain age,” predict dementia risk, detect brain tumor mutations and even forecast survival rates for individuals with brain cancer. The findings, published in Nature Neuroscience, represent a significant step forward in applying AI to complex neurological and oncological challenges.

Developed by a collaborative team at Mass General Brigham, Harvard Medical School, and other institutions, BrainIAC differs from many existing AI models in its approach. Instead of being trained for a single, specific task, it’s a “foundation model” – pre-trained on a vast dataset of nearly 49,000 brain MRI scans. This broad training allows it to adapt to a wide range of applications with minimal additional training, a crucial advantage when dealing with limited or specialized datasets.

How BrainIAC Learns from Images

The key to BrainIAC’s versatility lies in a technique called self-supervised learning. Unlike traditional AI models that require large amounts of labeled data (where humans identify specific features in the images), BrainIAC learns patterns and structures directly from unlabeled MRI scans. This is particularly important in medical imaging, where obtaining accurately labeled datasets can be time-consuming, expensive, and require specialized expertise. By identifying inherent features within the images themselves, the model builds a foundational understanding of brain anatomy and pathology.

“There is a vast trove of data within the millions of brain MRIs performed each year in the United States,” explains Benjamin H. Kann, senior author of the study. “Typically, these scans are analyzed by humans for a particular reason, but this only scratches the surface of the story that these scans might advise us about our patients. With AI and advanced computational imaging techniques, People can unlock much more information from these scans than ever before—which may lead to potent, clinically useful ways to track a variety of acute and chronic conditions, from stroke, to cancer, to dementia, as well as predict future risks for patients.”

Outperforming Specialized Models

Researchers rigorously tested BrainIAC’s performance across seven distinct clinical tasks, ranging from classifying MRI scan types to identifying specific brain tumor mutations. In many cases, the model outperformed more conventional, task-specific AI frameworks, particularly in scenarios where limited training data were available. This ability to generalize from a broad base of knowledge is a major advantage, as it reduces the need for extensive, specialized datasets for each new application. The pre-training process, utilizing a method called contrastive learning, allows the model to establish a core baseline understanding before tackling specific challenges.

The model’s capabilities extend beyond diagnosis. As illustrated in a sample output from the study, BrainIAC can highlight areas of the brain it focuses on when determining the mutational status of a brain tumor, offering a degree of interpretability that is often lacking in “black box” AI systems. This transparency is crucial for building trust and facilitating clinical adoption.

Potential Applications and Future Directions

The potential applications of BrainIAC are far-reaching. Beyond the initial tasks demonstrated in the study – predicting brain age, dementia risk, tumor mutations, and cancer survival – the model could be adapted to assist in the diagnosis and monitoring of a wide range of neurological and psychiatric conditions, including Alzheimer’s disease, autism, Parkinson’s disease, and stroke. Researchers believe that this technology could be particularly valuable in settings where access to specialized expertise is limited.

Looking ahead, the researchers envision further improvements to BrainIAC through training on even larger and more diverse datasets. They also suggest that the principles behind BrainIAC could be applied to develop similar foundation models for other types of medical imaging, such as CT scans and ultrasound recordings. The ultimate goal is to create a suite of AI tools that can assist clinicians in making more accurate and timely diagnoses, leading to improved patient outcomes.

Open Access and Collaboration

To accelerate research and development in this field, the BrainIAC algorithm is being made openly available to the research community through a dedicated website: www.brainiac-platform.com. Researchers are already collaborating with the development team to investigate BrainIAC’s potential for various brain disorders, including Alzheimer’s disease and traumatic brain injury. This collaborative approach underscores the importance of open science in driving innovation and translating research findings into real-world benefits.

The development of BrainIAC represents a significant advancement in the application of AI to medical imaging. By leveraging the power of foundation models and self-supervised learning, researchers are unlocking new insights from existing data and paving the way for more accurate, efficient, and accessible healthcare.

This article is the result of careful human work by our author Ingrid Fadelli, edited by Stephanie Baum, and fact-checked and reviewed by Robert Egan. We rely on readers like you to preserve independent science journalism alive. If this reporting matters to you, please consider a donation.

Journal information: Nature Neuroscience (http://www.nature.com/neuro/)