AI Models BrainIAC and Prima Advance Neurological Diagnosis with Enhanced MRI Analysis

Researchers at Mass General Brigham and the University of Michigan have independently developed artificial intelligence (AI) models designed to improve the analysis of brain MRI datasets and aid in neurological diagnosis. These models, named BrainIAC and Prima, respectively, demonstrate capabilities in identifying various conditions, predicting risks, and potentially streamlining clinical workflows by processing medical imaging data.

BrainIAC: A Foundation Model for Diverse Brain MRI Analysis

Investigators at Mass General Brigham have developed BrainIAC, an AI foundation model aimed at analyzing brain MRI datasets for a range of medical tasks. The model's capabilities include identifying brain age, predicting dementia risk, detecting brain tumor mutations, and forecasting brain cancer survival rates.

BrainIAC reportedly outperformed other task-specific AI models, particularly in scenarios with limited training data.

Benjamin Kann, MD, of the Artificial Intelligence in Medicine (AIM) Program at Mass General Brigham, indicated that BrainIAC has the potential to accelerate biomarker discovery, enhance diagnostic tools, and support the integration of AI into clinical practice for personalized patient care.

Traditional AI approaches for brain MRI analysis frequently require extensive training with large, annotated datasets, which can be challenging to acquire. BrainIAC addresses these challenges by employing self-supervised learning, allowing it to identify inherent features from unlabeled datasets and adapt to various applications.

The research team pretrained BrainIAC on multiple brain MRI imaging datasets and subsequently validated its performance on 48,965 diverse brain MRI scans across seven distinct clinical tasks. The model demonstrated the ability to generalize its learnings across both healthy and abnormal images, applying them to tasks ranging from classifying MRI scan types to detecting brain tumor mutation types. Further research is planned to test the framework on additional brain imaging methods and larger datasets.

Prima: High-Accuracy AI for Neurological Diagnosis and Treatment Urgency

A separate study published in Nature Biomedical Engineering details an AI-powered model named Prima, developed at the University of Michigan, which analyzes brain MRIs to diagnose neurological conditions and assess the urgency of treatment.

Key findings reported by the University of Michigan researchers include:

• Prima detected neurological conditions with up to 97.5% accuracy.
• The model predicted the urgency of patient treatment.
• Researchers suggest this technology could impact neuroimaging in health systems.

Dr. Todd Hollon, a neurosurgeon at University of Michigan Health and assistant professor of neurosurgery at U-M Medical School, stated that the model has the potential to reduce the burden on physicians and health systems by improving diagnosis and treatment through rapid information processing.

Prima was tested on over 30,000 MRI studies over one year and demonstrated diagnostic performance over 50 radiologic diagnoses of major neurological disorders.

Prima is designed to determine which cases require higher priority, such as brain hemorrhages or strokes. It can automatically alert providers and recommend the appropriate subspecialty provider, such as a stroke neurologist or neurosurgeon.

Prima functions as a vision language model (VLM), processing video, images, and text simultaneously. Unlike some previous models that relied on manually curated subsets of MRI data, Prima was trained on a comprehensive dataset comprising over 200,000 MRI studies and 5.6 million sequences collected since radiology digitization began at University of Michigan Health. The system also integrated patients' clinical histories and the reasons physicians ordered medical imaging studies.

Samir Harake, a data scientist involved in the research, explained that Prima integrates patient medical history and imaging data to develop a comprehensive understanding of health, contributing to its performance across prediction tasks.

The researchers suggest that Prima could help address challenges such as neuroradiology workforce shortages and diagnostic errors by providing rapid results. While the research is in its initial evaluation stage, future work aims to integrate more detailed patient information and electronic medical record data for enhanced diagnostic accuracy. The model could also potentially be adapted for other imaging modalities, including mammograms, chest X-rays, and ultrasounds.

Dr. Hollon described Prima as a "co-pilot" for interpreting medical imaging studies.