Artificial intelligence and machine learning can help make fundamental as well as translational discoveries in the field of neuroscience.
TThe development of better and more reliable artificial intelligence (AI) technology has immeasurable applications in the field of scientific research. AI gives scientists powerful extra eyes and hands, helping them sift through large amounts of data in seconds, guide experiments, and write better manuscripts.
“We are seeing the emergence of subfields that are AI plus X, and X is basically every scientific field, and neuroscience is no exception,” said Georgia Tech neuroengineer Christopher Rozelle, who moderated the AI press conference at the 2025 Society for Neuroscience Conference.
During the session, five panelists discussed the applications of AI in biology and how machine learning can enhance clinical practice and the field of neuroscience, from data analysis to clinical diagnosis.
Modified artificial neural networks provide clues about how the brain integrates sensory information
The human brain integrates various sensory inputs to ensure that we are aware of our surroundings.1 “Today this challenge is also very successfully solved by artificial neural networks (ANNs), which are actually inspired by the brain,” said Marcel Oberlander, a neurobiologist at the Max Planck Institute for Behavioral Neurobiology.
This motivated Oberlander and his team to explore whether ANNs could help improve our understanding of brain function, particularly perception. However, ANNs lack many properties of the brain, such as neuronal diversity and connectivity.
Incorporating these elements into the ANN resulted in better performance than traditional models. Brain-like ANNs require less data and time to achieve the same results. Incorporating characteristics from the brain to generate ANNs could help neuroscientists better understand how these characteristics contribute to brain functions such as perception, Oberlander said.
Reverse engineering neurons with AI
The inability of neurons to send electrical signals to each other underlies nearly every neurological disease, from epilepsy to schizophrenia.2 Although patch-clamp electrophysiology is useful for measuring the electrical output of neurons, it cannot provide information about the ion channels responsible for the altered electrical signal.
Classical computational models integrate ion channels and neuron morphology to predict a cell’s electrical output. To reverse this process, neurobiologist Roy Ben Shalom and his team at the University of California, Davis built a deep learning model called NeuroInverter. AI tools have successfully analyzed and predicted the ion channel composition of more than 170 different types of neurons.
“NeuroInverter opens the door to a deeper understanding of brain disorders,” said Ben-Shalom. “We can now create a ‘digital twin’ of any neuron just by knowing its voltage response. This will be a very powerful tool for disease modeling and discovery.”
AI tools help analyze gait disorders
Aging and neurological diseases such as stroke and multiple sclerosis impair the ability to walk.3,4 To treat and rehabilitate patients, clinicians must first accurately measure gait impairment. Clinician assessments can be subjective, while objective tools like motion capture systems require specialized and expensive equipment.
These limitations have led researchers to look for practical and cost-effective alternatives. Trisha Kesar, a rehabilitation medicine researcher at Emory University, and her team used machine learning algorithms to analyze smartphone videos of normal and impaired gait patterns. This allowed us to classify clinically relevant gait disorders with more than 85% accuracy.
“Overall, our goal is to make accurate and objective gait analysis available to clinicians in a variety of community settings, which can aid in more accurate, effective, and more individualized rehabilitation,” Kesar said.
AI detects freezing gait in Parkinson’s disease patients early on
People with Parkinson’s disease may suddenly find themselves unable to take a step, as if their feet are stuck to the floor.5 Although deep brain stimulation has proven to be a promising treatment for other conditions, treatment of freezing gait remains limited because the onset of symptoms is unknown.
Using virtual reality, Cleveland Clinic scientist Jay Alberts and his team found that scenarios that caused freezing of gait activated unique neural signatures in participants’ brains. Alberts and his team trained a machine learning model based on data from each of these trials to predict the probability that an individual would experience freezing gait. The AI model was able to accurately detect gait freezing before it occurred.
Paul Cantlay, a scientist on Alberts’ team, said: “This could potentially treat freezing gait before it actually develops using an adaptive deep brain stimulation paradigm.”
AI tool that deciphers the meaning of words from brain activity
Brain-computer interfaces (BCIs) can help restore communication to severely disabled patients.6 Current approaches decode phonetic aspects of speech, but can confuse similar-sounding words.
To overcome this, neurophysiologist Matthew Nelson and his team at the University of Alabama at Birmingham recorded brain activity as people thought about words from different categories, such as clothes and animals. They used machine learning algorithms to decipher these entire semantic categories based on an individual’s brain activity. The AI-based tool was able to accurately decode categories 77% of the time.
“Overall, we believe this is an important step toward language BCIs that will ultimately combine semantic and phonetic information, as well as information from other areas of language, to provide the richest, most robust, and best overall language decoding in BCIs,” Nelson said.
