Transforming brain research with AI and machine learning

Using machine learning, researchers are capable of use data from the brain to glean deeper insights and apply this latest knowledge in clinical settings. The findings shall be presented on Monday, November 13, 2–3 p.m. EST at Neuroscience 2023, the annual meeting of the Society for Neuroscience and the world’s largest source of emerging news about brain science and health.

Machine learning is a branch of artificial intelligence (AI) that centers on enabling computers to research data in increasingly complex ways. Researchers use algorithms with adaptive models that may change and develop without outside intervention. In neuroscience research, machine learning can study large datasets with information from the human brain and likewise apply models to predict outcomes based on that information.

Recent findings show that:

• Researchers can successfully discover activity between the thalamus and motor-related regions as correlated with depressive symptoms, applying neuroimaging techniques to detect abnormal brain regions in psychiatric disorders. (Ayumu Yamashita, Advanced Telecommunications Research Institute International)
• Machine learning models using patient data can predict progression from mild cognitive impairment (MCI)to Alzheimer’s disease (AD); their study focuses on using kPCA (kernel Principal Components Evaluation). (Nikita Goel, University of Southern California)
• A brand new application of artificial neural networks can uncover how cells within the visual cortex interpret light signals from the retina and translate that data into vision. (Dan Butts, University of Maryland)
• Researchers applied machine learning models to deep brain recordings and combined that with traditional control theory methods to create a tool for predicting targets for deep brain stimulation (DBS). This optimization can assist with treatment for dystonia in children. (Maral Kasiri, University of California, Irvine)

“Advances in AI and machine learning are transforming brain research and clinical treatments,” said Terry Sejnowski, the Francis Crick Professor on the Salk Institute for Biological Studies and distinguished professor at UC San Diego, who will moderate the press conference. “Brain recordings produce huge datasets that could be analyzed with machine learning. Predictive modeling, machine-brain interfaces, and neuroimaging/neuromodulation are areas with particular promise in developing latest therapeutics and treatment plans for patients.”

This research was supported by national funding agencies including the National Institutes of Health and personal funding organizations. Discover more about AI and brain research on BrainFacts.org.

Monday, November 13, 2023

2–3 p.m. EST

Walter E. Washington Convention Center, Room 202B

Press conference summary

• These presentations explore the wealth of information that could be extracted and analyzed using machine learning. These topics range from predictive modeling that identifies brain activity for disorders resembling Alzheimer’s disease (AD) and depression (the primary two studies) to analyzing the brain’s complex visual system and applying models of deep brain networks for advancing therapies (the third and fourth studies, respectively).

Unsupervised-based feature selection method robustly extracted resting state functional connectivity related to major depressive disorder

Ayumu Yamashita, [email protected], Abstract PSTR099.15

• Researchers used neuroimaging techniques combined with machine learning to discover brain biomarkers. To find out how accurately and robustly these techniques detect abnormal brain regions in psychiatric disorders, the researchers used three forms of feature selection methods with regard to major depressive disorder (MDD).
• Out of the three methods, they found that the unsupervised-based feature selection method robustly extracted functional connections with larger effect sizes within the test data. Using MRI data from 1,162 patients, including 334 with depression, researchers extracted coordinated activity between parts of the brain (especially the thalamus and motor-related regions) which can be correlated with depressive symptoms.

Predicting future decline from mild cognitive impairment to Alzheimer’s disease with machine learning and 3D brain MRI

Nikita Goel, [email protected], Abstract NANO03.07

• Predictive models use machine learning to try to find out who will develop brain disorders resembling AD. Researchers used a variety of models to see what model might predict mild cognitive impairment (MCI) that may progress to AD.
• The researchers used MRI scans together with demographic and genetic data from greater than 2,448 people. Essentially the most functional model they developed used kPCA (kernel principal components evaluation) to generate latest features by combining the scan data with data about an AD risk gene in addition to basic patient data.

Biologically constrained deep neural networks to parse visual computations being performed in the first visual cortex

Dan Butts, [email protected], Abstract PSTR149.16

• Researchers used a novel application of artificial neural networks called convolutional deep neural networks (CNNs) to learn the way cells within the visual cortex of the brain interpret light signals captured by the retina.
• Using machine learning, researchers are starting to know how information concerning the visual world is carried across brain cells within the visual cortex.

Identifying the effective targets for deep brain stimulation: system identification using a straightforward recurrent neural network

Maral Kasiri, [email protected], Abstract PSTR154.16

• Clinicians face challenges with deep brain stimulation (DBS) in determining the optimal targets throughout the brain. In this fashion, treatment could be tailored to a selected patient to higher treat movement disorder symptoms.
• Based on intrinsic brain signals of brain areas (i.e., basal ganglia and thalamus) in children with dystonia, machine learning could be applied to develop models of deep brain networks combined with traditional control theory and statistical methods. The tip result’s a tool to discover essentially the most favorable DBS targets for patients.