Mass General Brigham investigators have developed a robust new AI foundation model that is capable of analyzing brain MRI datasets to perform numerous medical tasks, including identifying brain age, predicting dementia risk, detecting brain tumor mutations, and predicting brain cancer survival. The tool, known as BrainIAC, outperformed other, more task-specific AI models and was especially efficient when limited training data were available. BrainIAC is a foundation model trained and validated on nearly 49,000 diverse brain MRI scans to identify key neurological health indicators without needing extensively labeled datasets. Results are published in Nature Neuroscience.
“BrainIAC has the potential to accelerate biomarker discovery, enhance diagnostic tools, and speed the adoption of AI in clinical practice,” says corresponding author Benjamin Kann, MD, of the Artificial Intelligence in Medicine Program at Mass General Brigham. “Integrating BrainIAC into imaging protocols could help clinicians better personalize and improve patient care.”
Despite recent advances in medical AI approaches, there is a lack of publicly available models that focus on broad, brain MRI analysis. Most conventional frameworks perform specific tasks and require extensive training with large, annotated datasets that can be hard to obtain. Furthermore, brain MRI images from different institutions can vary in appearance and be based on their intended applications (such as in neurology vs oncology care), making it challenging for AI frameworks to learn similar information from them.
To address these limitations, the research team designed a brain imaging adaptive core, or BrainIAC. The tool uses a method called self-supervised learning to identify inherent features from unlabeled datasets, which can then be adapted to a range of applications. After pretraining the framework on multiple brain MRI datasets, the researchers validated its performance on 48,965 diverse brain MRI scans across seven distinct tasks of varying clinical complexity.
They found that BrainIAC could successfully generalize its learnings across healthy and abnormal images and, subsequently, apply them to both relatively straightforward tasks, such as classifying MRI scan types, and very challenging tasks, such as detecting brain tumor mutation types. The model also outperformed three conventional, task-specific AI frameworks at these applications and others.
The authors note that BrainIAC was especially good at predicting outcomes when training data was scarce or task complexity was high, suggesting that the model could adapt well to real-world settings where annotated medical datasets are not always readily available. Further research is needed to test this framework on additional brain imaging methods and larger datasets.
— Mass General Brigham
Improving Brain Surgery Outcomes with Noninvasive Advanced Imaging
Researchers have found a promising new way to predict which patients with a congenital brain malformation would respond well to surgical intervention. The results of this preliminary application of a noninvasive advanced imaging approach, which were published this month in the Journal of Neurosurgery, could significantly improve the quality of care for patients with Chiari malformation type-I (CM-I), a congenital condition in which the lower part of the cerebellum bulges through the normal opening at the base of the skull and into the spinal canal.
Although a congenital condition, many people don’t begin to experience symptoms such as headaches, dizziness, impaired balance, and more, until adulthood. Surgical intervention is successful in about 75% of cases but, because of the risks associated with the surgery, an effective way to identify the conditions that would predict positive surgical outcomes is vital.
The study was conducted by a team working through the Center for Systems Imaging Core at Emory University and led by John Oshinski, PhD, a professor of radiology and imaging sciences and biomedical engineering; Daniel Barrow, MD, a professor of neurosurgery; and Grace McIlvain, PhD, now an assistant professor of biomedical engineering and radiology at Columbia University.
Measuring Brain Motion and Cerebrospinal Fluid Flow
The researchers tested two measures. They first used phase-contrast MRI to measure cerebrospinal fluid (CSF) flow/stroke volume, which is the amount of CSF passing through the cerebral aqueduct during a heart cycle. CM-I obstructs normal flow, causing pressure variations in and around the intracranial space. This impaired CSF flow also increases brain motion, as brain tissue tries to displace CSF to maintain cerebral homeostasis. The researchers measured brain motion using a technique called cine displacement encoding with stimulated echoes (DENSE) in MR imaging. DENSE can quantify submillimeter displacements associated with brain tissue motion. They also assessed tonsillar descent, or the amount of the lower part of the cerebellum that descends into the spinal canal. This measure has previously been used as a standard clinical presurgical assessment metric but has produced mixed results.
Promising Results
The researchers found that presurgical measures of cerebral dynamics were more predictive of improvements to CSF flow and brain motion after surgery than the conventional measure of presurgical tonsillar descent.
“The measurement of neural dynamics, such as brain motion and CSF flow, rather than static markers, such as tonsillar descent, are a new approach to understanding the pathophysiology of brain disease and represent a new method to improve patient treatment options,“ Oshinski says.
The next step is to validate the work through a larger, blinded clinical trial, the work for which is already underway as the team recruits additional sites for clinical trials.
— Emory University School of Medicine