Artificial intelligence (AI) is poised to revolutionize the care of neurological diseases, from detecting tumors the human eye can’t see and using implantable devices to manage various conditions to crunching data from thousands of people with the same condition to guide the development of new treatments, according to neurologists and biomedical researchers speaking at the Presidential Symposium – Present and Future Applications of AI in Neurological Care and Research at the 149th Annual Meeting of the American Neurological Association (ANA).

While used for years to analyze data, AI is on the verge of true breakthroughs in the diagnosis, prognosis, and treatment of neurological disorders, including Parkinson’s disease, Alzheimer’s disease, epilepsy, and stroke. One main benefit of AI is making predictions, which may help prevent some neurological conditions from developing – or make them less severe by ensuring earlier treatment – instead of just reacting, experts say.

“Our vision of the future includes everything from brain-machine integration for the treatment of movement disorders to powerful deep learning methods that combine massive data sets to analyze brain imaging along with clinical, genetic and protein data to obtain a detailed assessment of individual health and risks for complex neurological diseases such as Alzheimer’s,” said Elizabeth Ross, MD, PhD, FANA, ANA President, plenary session chair and Nathan Cummings Professor of Neurology and Neuroscience and director of the Center for Neurogenetics, Weill Cornell Medicine, New York.

“AI will improve care by assisting – not replacing – clinicians and researchers by providing a new perspective that integrates multiple layers and types of data into personalized clinical or preclinical predictions,” said Cassie Mitchell, PhD, plenary session co-chair and Assistant Professor, Wallace H. Coulter Department of Biomedical Engineering, Georgia Tech and Emory University School of Medicine, Atlanta.

Various forms of AI in use or in development for the benefit of neurological conditions, include:  

  • Machine learning, which can analyze large amounts of medical data to identify patterns, make predictions and support decision-making, including by significantly speeding up the development of anti-seizure drugs to deliver effective medicines to people with epilepsy who aren’t helped by existing drugs
  • Deep learning, a subset of machine learning, which can analyze complex data including medical images, and detect abnormalities no human eye can see – for example, a blood clot or the source of a stroke – based on information gleaned from very large datasets of thousands of patients
  • Natural language processing, which understands and processes human language and can be used in wearable and implantable devices that give patients and their doctors real-time feedback to help them better manage conditions such as Parkinson’s disease
  • Robotics, which can be used in various ways, from improving brain surgery for people with epilepsy to providing companions for people with Alzheimer’s disease

AI and Implantable Devices

Due to the brain’s complexity, the use of AI for neurological care lags behind cardiology and diabetes (think implantable defibrillators and insulin pumps), but there’s reason for optimism, particularly in the use of wearable and implantable devices.

“In the not-too-distant future, implantable devices will employ AI to help patients better manage their neurological conditions as well as give doctors more immediate feedback so that both patients and providers can adjust their care to optimize quality of life,” said Brian Litt, MD, who is presenting “AI Applications for Implantable Devices in Neuropsychiatric Diseases” and is the Perelman Professor of Neurology, Neurosurgery and Bioengineering at the University of Pennsylvania, Philadelphia. “As our ability to unify and interpret data globally matures, an eventual goal would be to use AI to assess someone, predict what kind of neurological diseases they may develop and initiate preventive therapy before they even show signs of the disease, whether it’s epilepsy or Alzheimer’s disease.”

The neurology clinic of 2040 likely will look very different than the one today, he said, with AI (along with advances in treatment, which AI also is facilitating) being most helpful to:

Provide real-time feedback to patients to help prevent seizures. For example, a device implanted under the skin in the scalp could alert a patient that if they have another drink of alcohol they will significantly increase their risk of having a seizure. If a patient has a bad night, the device could let them know the lack of REM sleep has led to an increase in abnormal brain activity, meaning it’s not safe for them to drive.

Predict and potentially prevent a neurological condition. For example, an AI analysis of a patient’s genetics, family history, basic tests (blood and imaging), where they live, their job and data from other people with the condition could determine if a patient has a high likelihood of developing a condition such as Alzheimer’s disease in the next few years. Based on this knowledge, the neurologist could prescribe therapy to potentially delay or prevent the disease, such as gene editing or disease-modifying medication.

Help patients manage a condition that doesn’t respond well to treatment. For example, a patient with Parkinson’s disease may have an implantable device that could stimulate parts of the brain. When the patient wants to take a walk, they could direct the device to stimulate their brain to improve their gait. If they want to write a letter, they can direct the device to stimulate the parts of their brain to reduce tremors in their hands.

AI has the potential to make care more accessible and available to all patients, Dr. Litt said. For example, a neurologist can implant a device in a patient with Parkinson’s disease and monitor them from hundreds of miles away, adjusting their medications if needed. This is already starting in some centers.

AI in Imaging Helps Identify Neurological Diseases and Conditions Early So They Can Be Treated

AI is playing an increasingly valuable role in neurological disease diagnosis, particularly in helping detect blood clots, stroke and disease pathology, such as Alzheimer’s disease. In some cases, it can detect problems in the brain (such as amyloid plaque, a marker for Alzheimer’s disease) that previously could only be confirmed through an autopsy.

“Think of AI as an extra pair of eyes that can see things we can’t and store knowledge beyond what any of us could remember. More advanced tools will help us diagnose diseases that we can’t easily diagnose today, such as different types of dementia, various psychiatric conditions and developmental disorders such as autism,” said Paul Thompson, PhD, who is presenting “AI for Discovery and Diagnosis of Brain Diseases using Deep Learning and Large-Scale Neuroimaging” and is a professor at the Keck School of Medicine at the University of Southern California, Los Angeles. “In the near future, I think AI’s use in radiology will be pretty exciting – we’ll see fewer cases that baffle doctors and much better diagnostic tools and the rate of missed key findings or treatable conditions via brain scan will drop.”

There are some promising new treatments for Alzheimer’s disease and Parkinson’s disease, and more in the pipeline. Using pattern recognition from enormous databases, AI could help identify the right treatments for the right patients more quickly using brain imaging in combination with other test results to improve early detection and diagnosis. One such database is Enhancing Neuro Imaging Genetics Through Meta Analysis (ENIGMA), a brain network of medical images and data from more than 50,000 patients with more than 30 brain diseases.

Using AI, experts are beginning to understand what Alzheimer’s disease risk genes are doing to the brain, such as reducing the volume of the hippocampus, the part of the brain responsible for learning, memory and emotional response. This could pave the way for the development of new drugs.

AI Could Enhance Drug Development

About a third of people with epilepsy have a refractory form, meaning neither medication nor surgery effectively controls their seizures. This is particularly problematic for children because seizures can interrupt development, including cognition and language.

Current epilepsy drug development is time intensive, involving videotaping mice who have been given the drug while electrodes in their brains monitor electrical activity for days or weeks. But using machine learning, researchers have developed automated, AI-aided motion sequencing technology for epilepsy, a faster, cheaper and more accurate method of evaluating anti-seizure medications in one hour, even if the mouse doesn’t have seizures during that time.

“Motion sequencing could make a significant difference in how quickly new drugs are identified and developed,” said Ivan Soltesz, PhD, who is presenting “AI-Aided Analyses of Seizure and Interictal Phenotypes and Drug Responses in Epilepsy Models: Possibilities for Clinical Applications” and is the James R. Doty Professor of Neurosurgery and Neurosciences at Stanford University School of Medicine, Stanford, California. “The goal is to get effective drugs to children and adults with refractory epilepsy much sooner so they can grow and have a better quality of life.”

Personalizing Healthcare Through AI

ChatGPT and similar technologies paved the way for large-scale use of Large Language Modules (which involve deep learning and natural language processing), and healthcare and research are following this trend. There is great excitement in the biomedical and data science communities about the applications of generative AI. However, safeguards to ensure privacy are needed as we learn more about pros and cons.

Early examples of AI being used to personalize healthcare with significant oversight by humans include:

Ambient AI, which involves listening to patient and doctor visits and summarizing what was said (which can save the doctor time and be more detailed). In the near future we expect translation to happen seamlessly from many languages and dialects.

Applications of deep learning, such as predictive models for intensive care unit (ICU) outcomes, which predict the probability of a patient developing sepsis or other conditions (so they can be prevented), how long they are likely to stay in the ICU, as well as optimize workflows to allow better planning of resources.

AI is already being used to document processes and to check if the patient’s insurance plan authorizes the recommended labs and treatment, said Lucila Ohno-Machado, MD, PhD, MBA, who is presenting “The Power of Informatics and Machine Learning Applications to Personalized Healthcare Delivery” and is the Waldemar von Zedtwitz Professor of Medicine, and chair of the Department of Biomedical Informatics & Data Science, and Deputy Dean for Biomedical Informatics at Yale School of Medicine, New Haven, Conn. She anticipates that these AI applications will significantly expand during the next few years, and also envisions robotics being used in a variety of ways including helping neurological disease patients with daily activities, such as bringing them the remote control, finding their glasses or cleaning.

“Neurologists should be optimistic about the development of new models that assist them in the care of patients, in addition to managing paperwork,” she said. “Saving time from administrative work and having a co-pilot to make sure they don’t forget important details will increase time spent with patients and further improve the quality of care.”

Featured image: Dreamstime