Abstract: Researchers developed an important model called the Brain Language Model (BrainLM) using artificial artificial intelligence to map brain activity and its implications for behavior and disease. BrainLM leverages 80,000 scans from 40,000 subjects to build a basic model that captures the dynamics of brain activity without requiring specific disease-specific data.
The model significantly reduces the cost and scale of data required for traditional brain studies, offering a robust framework that can more effectively assess conditions such as depression, anxiety and PTSD than other tools. Is. Brain LM has demonstrated robust use in clinical trials, potentially halving costs by identifying patients who would potentially benefit from new treatments.
Important facts:
- Creative AI model: BrainLM uses generative AI to analyze patterns of brain activity from vast datasets, learning the underlying dynamics without patient-specific details.
- Cost and efficiency in research: This model reduces the need to enroll large numbers of patients in clinical trials, potentially significantly reducing costs by using its predictive capabilities to select suitable study candidates.
- Wide application: Tested on different scanners and demographics, BrainLM has shown superior performance in predicting various mental health problems and promises to aid future research and treatment strategies.
Source: Baylor College of Medicine
A team of researchers from Baylor College of Medicine and Yale University incorporated generative artificial intelligence (AI) to create a fundamental model for brain activity. The Brain Language Model (BrainLM) was developed to model the brain in silico and determine how brain activity relates to human behavior and mental diseases.
This research was published as a conference paper at ICLR 2024.
“We've known for a long time that brain activity is related to a person's behavior and many diseases such as seizures or Parkinson's,” said Dr. Chadi Abdullah, associate professor in the Menninger Department of Psychiatry and Behavioral Sciences. Co-corresponding author of the paper.
“Functional brain imaging or functional MRIs allow us to see brain activity throughout the brain, but we previously could not fully capture the dynamics of this activity in time and space using traditional data analysis tools. .
“Recently, people have started using machine learning to capture brain complications and how they relate to specific diseases, but that requires enrolling and thoroughly examining thousands of patients with a particular behavior or disease. If needed, it's a very expensive process.”
The strength of new generative AI tools is that they can be used to create basic models independent of a particular task or specific patient population. Generative AI can act as a detective uncovering hidden patterns within a dataset.
By analyzing data points and the relationships between them, these models can learn the underlying dynamics—how and why things change or evolve.
These basic models are then fine-tuned to understand a range of topics. Researchers used generative AI to study how brain activity works regardless of a particular disorder or disease.
It can be applied to any population without needing to know information about the subject's behavior, illness, history or age. All it needs is brain activity to teach computers and AI models how brain activity evolves over space and time.
The team took 80,000 scans from 40,000 subjects and trained the model to figure out how these brain activities were related to each other over time, creating the BrainLM brain activity core model. Now, researchers can use BrainLM to refine a specific task and ask questions in other studies.
“If you want to do a clinical trial to develop a drug for depression, for example, it can cost hundreds of millions of dollars because you need to enroll a large number of patients and treat them for a long period of time. .
“With the power of BrainLM, we can potentially cut that cost in half by enrolling only half of the subjects using the power of BrainLM to select individuals who are most likely to benefit from the treatment. “Therefore, BrainLM can use the knowledge learned from the 80,000 scans to apply it to these specific study subjects,” Abdullah said.
The first step, preprocessing, abstracts signals and removes noise unrelated to brain activity. The researchers fed the summaries into a machine learning model and masked a percentage of each individual's data. When the model learned the dynamic, they tested it on a left-out testing group.
They also tested it on different samples to understand how well the model could generalize to data obtained from different scanners and different populations, such as older adults and younger people.
They found that BrainLM performed well in different samples. They also found that BrainLM predicted depression, anxiety and PTSD severity better than other machine learning tools that did not use generative AI.
“We found that BrainLM performs very well. It is predicting brain activity in a new sample that was hidden from it during training, as well as data from new scanners and a new population. Saath is doing well,” Abdullah said.
“These impressive results were obtained with a scan of 40,000 subjects. We are now working on significantly expanding the training dataset.
“The more robust models we can build, the more we can do to help patients care, such as developing new treatments for mental illnesses or guiding neurosurgery for seizures or DBS.”
The researchers plan to use this model for future research to predict brain-related diseases.
About this AI and neuroscience research news
the author: Huma Warren
Source: Baylor College of Medicine
contact: Huma Warren – Baylor College of Medicine
Image: This image is credited to Neuroscience News.
Original research: The results will be presented at ICLR 2024.