System Integration: How AI Is Weaving Itself into Neurology

Artificial intelligence (AI) refers to the use of computer systems or machines to simulate human intelligence processes such as learning, reasoning, problem-solving, and decision-making. AI systems often rely on algorithms, data analysis, and machine learning to identify patterns and perform tasks that traditionally require human input. AI technologies include natural language processing, computer vision, robotics, neural networks, and machine learning, among others.

Article Highlights

Meet the Experts

Impact of AI in Neurology

Barriers to AI

AI's Future in Neurology

Emerging Trends in AI

Disease Specific Insights

In recent years, we’ve seen AI expand its role in healthcare processes across areas such as diagnosis and risk prediction, personalized medicine, administrative efficiency, surgical and robotic assistance, and improved access to care. It has not only made its way into medicine. AI has taken on an even larger presence in society, with an estimated 3.5 million industrial robots in operation globally, with around 590,000 new industry robots installed just this year. It is estimated that the healthcare robot market is expected to be worth more than $16 billion by 2025.

To better understand some of the potential benefits, as well as limitations, of AI in neurology, NeurologyLive^® gathered the opinions of several experts in their respective fields. As part of HCPLive’s This Year in Medicine series, these medical experts provided thoughts on how AI impacted their clinical and research work in 2024, some of the challenges/barriers the medical community faces with its incorporation, where its greatest potential lies, and the most intriguing emerging concepts with AI on the horizon. Additionally, at the end of the piece, the commentators voice their opinions on an AI-focused question related to their specific disease specialty.

Epilepsy (Partamian): The use of Artificial Intelligence (AI) in epilepsy care has significantly changed the field in 2024, bringing new possibilities for patients and their families. AI-driven wearable devices and mobile applications now monitor electroencephalography (EEG) data in real time, identifying subtle patterns that can predict seizures before they happen. This capability allows patients to take preventive measures, lowering the risk of injury and greatly enhancing their quality of life.

AI has also improved the accuracy of epilepsy treatments. It has refined the process of identifying the epileptogenic zone—the area of the brain that triggers seizures—resulting in more effective and targeted surgery. With AI, personalized treatment plans can be developed by analyzing individual patient data, such as seizure patterns and responses to medications. This leads to optimized drug dosages and helps find alternative therapies for those who do not respond to standard medications.

Furthermore, AI's ability to combine data from various imaging modalities has enhanced the diagnosis and prediction of seizures, as well as the identification of epileptogenic tissue. Additionally, AI-assisted surgical tools, including robotic systems and laser ablation techniques, have improved the safety and effectiveness of epilepsy surgeries, reduced complications and leading to better outcomes for patients.

Headache and Migraine (Petrušić): In 2024, AI has only begun to make an impact in the headache field. Researchers have increasingly applied machine learning-driven data analysis, emphasizing the importance of involving AI scientists in study design and data interpretation. As a result, significant progress has been made in utilizing neuroimaging data to differentiate migraines and their subtypes from other headache disorders. Additionally, prediction models for calcitonin gene-related peptide (CGRP) monoclonal antibody treatment response have been improved, marking a critical step toward personalized treatment approaches.

MS and Demyelinating Disorders (Zivadinov & Dwyer): I think one of the biggest impacts has been the roll-out by MRI manufacturers of AI-related improvements to image reconstruction, improving scan time and reconstruction quality for a huge proportion of patients with MS. We also seen more translational roll-out of AI-based MRI-quantification tools that are allowing clinicians to track individual patient's lesion and brain volume changes over time. There are still challenges there, but this will hopefully make a big difference as the techniques get more accurate and more able to account for individual variation and confounders. The AI application to acquisition and reconstruction is applicable to a variety of neurological disorders, not just MS. The scanner manufacturers are also proposing AI algorithms that require that only part of images are acquired further reducing speed of acquisitions by half or more.

Movement Disorders (Mahajan): In movement disorders, we still have not seen AI make a sizable, real-world impact, in my opinion. We have seen some very interesting work being published in this space using methodology that I did not think was possible. The most visible study has been the use of nocturnal breathing signals to assess patients with Parkinson disease (PD). AI has been increasingly used to analyze and supplement collected data from wearables, imaging and other technologies: to separate signal from noise, and identify novel signals.

Neuromuscular (Morren): In 2024, there have been several key inroads in enhancing our neuromuscular (NM) diagnostic tools with AI technology: Various machine learning (ML) and deep learning (DL) models have been leveraged to distinguish between electromyography (EMG) signals in normal vs Amyotrophic Lateral Sclerosis (ALS) vs myopathy patients, with diagnostic accuracy often approaching 100%!

Similarly, with NM ultrasound images, DL models using segmentation techniques achieved diagnostic accuracy of >90% for nerve entrapment disorders, and 87% for inflammatory myopathies. ML models achieved carpal tunnel syndrome (CTS) diagnostic accuracy of up to 95%, with DL applications in CTS getting up to 98%. A DL model using facial movements video for the diagnosis of myasthenia gravis (MG) showed 87% for diagnostic accuracy, 94% for MG severity assessment (compared to 4 neurology experts’ accuracy of 63%). Another ML model for prediction of ICU admission for MG patients demonstrated an accuracy of 94%.

DL models exhibited the ability to differentiate between facioscapulohumeral muscular dystrophy (FSHD) and myositis using muscle MRI, on par with radiology experts. Similar imaging applications allow a “virtual brain biopsy” AI model using MRI brain data for ALS diagnosis, attaining sensitivity and specificity of 90%.

Brain-computer interface (BCI) using DL in ALS patients allow scalp EEG input to be processed and transferred external devices with a movement accuracy ranging from 88 to 98%.

Additional recent innovative AI clinical applications in the NM space include the prediction of cognitive impairment in ALS patients based on genetic data, the diagnosis of ALS using surface EMG signals, and the identification of gait features specific to Duchenne Muscular Dystrophy (DMD) via accelerometer data.

These AI applications not only enhance diagnosis, they empower patients in managing their conditions and enable healthcare providers to deliver more responsive and effective care.

Sleep Disorders (Verma): The development of various auto-scoring AI tools that helps with sleep study scoring.The development of various wearable smart watches and rings that provide insightful data and trends about the sleep stages, sleep quality, nocturnal oxygenation, EKG alerting about the possibility of sleep apnea.There is advancement in AI tools that help patient understand their risks of sleep disorders.The AI may help in diagnosis of narcolepsy by analyzed seeing the pattern of polysomnography. These AI tools are also very helpful for patient education. There are various AI tools that helps with interpretation in different languages widening the reach to people speaking different languages.

Stroke (Chen): In 2024, the most significant impact of AI in stroke care remains centered on neuroimaging. AI-powered platforms like Viz AI and Rapid AI have earned widespread trust among neurologists and clinicians for their ability to rapidly identify large vessel occlusions, analyze CT perfusion, and detect intracranial hemorrhages. These tools have revolutionized acute stroke management by enabling early stroke team activation, expediting critical decision-making, and facilitating timely initiation of treatments for ischemic and hemorrhagic strokes. Beyond neuroimaging, AI has shown promise in other aspects of stroke care, including predicting stroke recurrence, estimating functional outcomes, and providing decision support for thrombolysis and thrombectomy tailored to individual patient and imaging profiles. While these advancements remain largely in the experimental stage and are yet to be widely implemented in clinical practice, their potential to streamline workflows and improve patient outcomes offers an exciting glimpse into the future of stroke care.

NeurologyLive: What are some of the challenges/barriers we still currently face with its incorporation and use?

Epilepsy (Partamian): Despite significant progress, several challenges still hinder the full integration of AI in epilepsy care. One major issue is the inconsistency in how hospitals collect and manage data, making it difficult to share and compare information across institutions. Additionally, many AI models lack transparency since the logic behind AI-model predictions may be unknown and doctors have a hard time trusting their predictions.

Another obstacle is integrating AI tools into daily hospital routines, which can be difficult due to limited resources. Training efficient AI models depends on labeled data. Yet, collecting large amounts of accurately labeled data is time-consuming and expensive. Additionally, some cases of epilepsy exhibit rare patterns that cannot be distinguished well by AI models trained on data that has not been trained on such cases. Moreover, having data from all types of epilepsy cases is a nearly impossible task since data is not readily shared between institutions. AI tools must also pass strict approval processes before they can be used in hospitals, leading to significant delays in implementation. Furthermore, if AI models are trained on data from only certain types of epilepsy, they can become biased, resulting in unfair treatment recommendations that may not benefit all patients equally.

Addressing these challenges is crucial to ensuring that AI tools in epilepsy care are effective, ultimately benefiting all patients.

Headache and Migraine (Petrušić): Migraine is the second most common disorder causing disability worldwide. However, it remains one of the most misdiagnosed, undertreated, and stigmatized conditions. Current standards indicate that only 50% of migraine patients experience at least a 50% improvement in symptoms with any given treatment. These unsatisfactory outcomes stem from the highly heterogeneous phenotypes of migraine, which create significant diagnostic and treatment challenges. Moreover, these phenotypes are poorly characterized in neuroimaging, neurophysiological, and biomedical/omics studies, and there is no validated biomarker for migraine or its subtypes.

A major bottleneck in migraine research is that studies have largely relied on unimodal analyses, failing to integrate multiple data types. Additionally, the lack of standardized protocols for data collection and analysis, coupled with the limited number of patients involved in studies, hinders AI utilization in headache research. AI tools require large, well-organized databases, which are challenging to establish even across a few coordinated headache centers. Tackling these challenges is possible only by industry investing more into forming a comprehensive international database for various primary and secondary headache disorders respecting all ethical and regulatory considerations regarding emerging next-generation AI technologies. Moreover, a collaboration between headache experts, multidisciplinary neuroscientists, and AI scientists who share the same goal and yet form multiple hubs each focusing on a specific challenge in the headache field is paramount.

Furthermore, building trust in the physician-AI technology-patient loop is pivotal for the successful implementation of AI-driven healthcare. To achieve that goal, addressing concerns about data use, privacy, bias, and safety is crucial for fostering patient trust in AI-driven healthcare. Moreover, investment in AI education of physicians specialized in headache research and treatment will ensure responsible AI use and foster trust among patients and their physicians. In addition, there is also an urgent need for policymakers to create frameworks that optimize the use of AI in making decisions while supporting clinicians who will remain in control of improved clinical processes and workflows. Finally, high costs and limited access to AI technology limit the adoption of AI tools in resource-limited settings.

MS and Demyelinating Disorders (Zivadinov & Dwyer): I believe the biggest challenges remain transparency and interpretability of AI models. There are also challenges how to regulate them effectively when they are constantly changing - how do we balance the desire for continual improvement with the desire to ensure that we are using validated tools? The challenge is also the expertise in doing it. There are no available platforms where clinicians or non-computer savvy scientists can use the AI without help of experts. Moreover, use of graphics processing units (GPU) and servers needed for it is still expensive and requires technical expertise. I would add that lack of large gold standard datasets is a major problem, especially including multi-center data.

Movement Disorders (Mahajan): Movement disorders, as a field, is perhaps more centered on history and examination than any other in medicine. Diagnostic accuracy and consequence of error are at least, equally important when it comes to AI and neurodegenerative disease. It is imperative that we not get overtly enamored by AI and hasten its incorporation into clinical care but ensure that algorithms are studied extensively in the real world. Such a discrepancy between training datasets and the real world is a known issue with AI. A recent study in the Movement disorders journal reported that only 20% of studies on AI using neuroimaging in Parkinson’s disease passed minimal quality criteria, with only 8% using external test sets. In those using external test sets, the accuracy was even lower. This raises serious concerns about the accuracy of training data sets in a field where the vast majority of what we observe and feel in clinic may otherwise never be documented.Large language models, the most common form of AI studied, have regularly demonstrated error. There are concerns regarding biases, issues with replication, “hallucinating data”, under or overestimation (due to overcompensation) in rare disorders, amongst others. There have been articles raising concerns about the relative absence of published data with negative results and the low bar of regulatory approval. The WHO has advised caution, and the FDA has called for regulation and the likely need for constant oversight. The field of movement disorders, unlike our colleagues in cardiology and vascular neurology, has the unenviable task of diagnosing and prognosticating disorders which may change over time and have a substantial degree of heterogeneity. Finally, the most important aspect of incorporation into direct clinical care is its acceptance by patients. Trust, touch, and the human experience is a huge part of the practice of medicine. In a study in Nature, that while comprehensibility was similar between a human physician and AI, empathy, reliability and willingness to follow advice was substantially better with a human physician.11 In a separate survey of 1400 US adults, 69% reported being uncomfortable being diagnosed by AI.

Neuromuscular (Morren): One significant issue is the limited availability of large, diverse datasets necessary for robust AI model development. Current datasets often lack sufficient representation of different populations and NM conditions, leading to potential biases and limited generalizability. Models trained on narrowly focused data may perform suboptimally when applied to broader clinical settings or underrepresented groups, potentially exacerbating health disparities. Accordingly, AI can perpetuate, amplify or mitigate healthcare disparities, depending on what guardrails are in place and how AI is used. Furthermore, these biases can result in diagnostic inaccuracies, undermining trust and reliability in AI systems.

Regarding the studies of ML & DL applications using EMG data for NM disease diagnosis (classification), many studies are further limited by “overfitting” effects in small, curated samples. Accordingly, a current study at Cleveland Clinic aims to improve upon these limitations using the largest known clinical EMG database pipeline for AI research to establish performance in classifying up to 5 NM conditions.

Another challenge lies in the regulatory and ethical landscape surrounding AI in healthcare. Existing frameworks often fail to keep pace with the rapid development of AI technologies, leaving gaps in governance, oversight and steering. Questions around accountability and liability persist, particularly when AI-generated recommendations contribute to an unintended outcome with patient harm. Additionally, there is concern about over-reliance on AI, which could erode the clinician-patient relationship and diminish the humanistic aspects of care. Transparent communication about AI's role and limitations is essential to promote trust and maintain the integrity of medical practice.

Integration into clinical workflows also poses practical challenges. Many healthcare providers face significant learning curves in adopting AI systems, with usability and interoperability issues further complicating implementation. AI tools need to seamlessly integrate with existing electronic health records and other systems, but such integration is often hampered by technical incompatibilities or workflow disruptions. Overcoming these barriers will require ongoing education, interdisciplinary collaboration, and iterative refinement of AI tools to align with the real-world needs of clinicians and patients.

A core tenet in the strategy to mitigate many of the limitations outlined is Human-Centered AI (HCAI) which focuses on stakeholder (including patients, clinicians, ethicists etc.) involvement in AI tool design. HCAI addresses biases at each stage of AI tool development: data collection, annotation, model development, evaluation, deployment, operationalization, monitoring, and feedback integration. Addressing these challenges is essential to unlocking AI's full potential while safeguarding equitable and ethical patient care.

Sleep Disorders (Verma): These devices have about 60-80% accuracy but getting better day by day as the AI is evolving.These findings could be very confusing and overwhelming for lay person with so much data. Due to HIPPA and privacy laws, these devices may not have been trained on vast variety of clinical data limiting the generalization of these devices.We still need more transparency about different algorithms used by different companies. There is still need more collaboration between researcher, clinicians, AI engineers and institution with large data set to develop strong AI models help the field of sleep Medicine. More public education is needed in the field of Sleep Medicine that may help people to understand the data trends possible clinical implications.

Stroke (Chen): AI holds immense promise, particularly in its ability to make highly accurate predictions using advanced techniques such as deep learning. However, several challenges still hinder its integration into clinical practice. A major concern is the "black box" nature of AI algorithms—they often lack transparency, making it difficult for clinicians to understand the logic or data underlying their predictions. This opacity can undermine trust, especially when patient lives are at stake. Additionally, while AI is undoubtedly exciting, its true value lies in its ability to deliver tangible benefits, such as improving patient outcomes or addressing other clinically meaningful endpoints. Demonstrating such benefits often requires rigorous validation through randomized trials or other robust experimental designs. These processes are complex, resource-intensive, and time-consuming, posing significant barriers to widespread adoption in healthcare.

NeurologyLive: What do you see for the potential future it holds from both a clinical and research perspective?

Epilepsy (Partamian): The future of AI in epilepsy is promising. AI has the potential to help doctors diagnose epilepsy more accurately and develop personalized treatment plans. It will also enable remote monitoring, reducing the need for frequent hospital visits and making care more accessible for people in in centers with less expertise.

In research, AI could revolutionize how we understand and treat epilepsy. For example, virtual models of patients' brains created using AI could allow scientists to simulate seizures and test various treatments in a risk-free environment. Surgeons might also use these models to practice procedures virtually, improving outcomes and minimizing risks. Additionally, AI may help discover new biomarkers for epilepsy, enabling earlier diagnosis and more effective interventions.

A key focus for future development will be making AI more explainable. By designing AI systems that clearly show how they arrive at decisions, doctors can better understand and trust these tools, leading to greater adoption in clinical settings.

Our group is working on the development of sophisticated AI tools that identify the epileptogenic brain regions automatically by incorporating information from various modalities. Current work focuses on invasive data analysis that can automatically identify the epileptogenic areas and predict surgical outcomes. In the future, our work will extend to developing tools that extract biomarkers from noninvasive data and approximate these epileptogenic brain areas with high accuracy and precision reducing the risks associated with long term invasive data acquisition and enhancing patient outcome. We aim to develop tools that could potentially be used in centers that lack multidisciplinary expertise and could potentially be integrated into 3D surgical navigation systems.

Overall, AI has the potential to revolutionize epilepsy care and research, allowing reversing psychological and social comorbidities.

Headache and Migraine (Petrušić): Deep phenotyping of migraine subtypes using AI-driven analysis of multimodal data, including neuroimaging, multiomics, and clinical history, will significantly advance our understanding of multi-layered pathophysiology, boost the discovery of novel digital biomarkers through deep learning models applied to omics and neuroimaging data, and open new avenues for investigating next-generation specific targets for acute and preventive treatment. Furthermore, advances in headache research will be followed by the acceleration of clinical trial design by identifying ideal patient cohorts and better monitoring clinical trials with the help of multi-agent chatbots. This will enable personalized medicine for headache disorders by using AI to tailor treatments based on genetic, behavioral, and environmental factors. In addition, large language models fine-tuned for migraine research, diagnosis, and treatment, could assist in study protocol design and provide augmented decision-making support for the diagnosis and management of headache disorders

MS and Demyelinating Disorders (Zivadinov & Dwyer): From a clinical perspective, obviously the new MS criteria that are coming out soon and including paramagnetic rim lesions (PRLs) and central vein sign (CVS), as discussed at ECTRIMS in Copenhagen, will really benefit from good AI to help identify PRLs. There is also a lot of good research going into earlier prediction of MS, and treatment response monitoring and prediction. I hope these will be incorporated into easy tools for every clinician. From a research perspective, I think we are still only just starting to reap the benefits of AI in terms of unsupervised exploration of datasets - we still have a lot of questions about potential heterogeneous subgroups of MS, and AI can start to help us tease this apart in a non-biased way. Real-word validation and in-use validation has even not began using AI tools. There is a lot of potential in clinical trials for better selection of patients, randomization, as well as in the clinical routine prediction of long-term outcomes.

Movement Disorders (Mahajan): Once we better understand the strengths and weaknesses of AI, I think it can be used to great effect. Perhaps the best use of AI can be to create smarter health systems to offload regulatory and administrative burdens on physicians. This aspect of the use of AI has been termed “boring AI”.13 In clinical care, it pertains to activities such as scheduling, addressing basic MyChart questions, prior authorization, patient risk stratification for return visits and resource allotment. A recent study in NEJM reported the use of AI in reporting hospital quality measures which otherwise may cost a hospital greater than 5 million dollars annually to generate.Researchers at UCSD generated draft abstractions in < 1 hour with >90% accuracy.14 In research, these can be used for monitoring symptoms. This is already being done in place of the use of diaries to document fluctuations in symptoms and capture subtle changes. It will additionally help with remote trials and inclusion of patients who are otherwise unable to travel to an academic health center for clinical trials and other research, thereby increasing much-needed inclusivity.

Neuromuscular (Morren): Clinically, AI's growing capabilities will lead to earlier and more accurate diagnoses through integration with wearable technologies and real-time data analysis. For example, smart devices embedded with AI algorithms could monitor patients' motor functions continuously, enabling early detection of disease progression or treatment responses. Additionally, AI could facilitate real-time decision support in clinical settings, offering recommendations tailored to individual patients by synthesizing data from electronic health records, imaging studies, and genetic analyses, among other relevant data sources. This precision medicine approach will help optimize treatment plans, improve patient outcomes, and reduce healthcare cost.

From a research perspective, AI is poised to accelerate drug discovery and the development of targeted therapies for neuromuscular disorders. By analyzing vast datasets of genetic, proteomic, and clinical information, AI models can identify novel therapeutic targets, predict drug efficacy, and even simulate clinical trials, significantly reducing the time and cost associated with traditional drug development pipelines (overall drug development can go from >10 yrs to one-tenth of the time, and drug discovery cost reduced by nearly 70%). Moreover, AI's ability to process unstructured data, such as imaging and electrophysiological signals, opens avenues for discovering new biomarkers and disease mechanisms. These advancements can deepen our understanding of complex neuromuscular conditions and support the development of innovative treatments.

Prospectively, the synergy between AI and emerging technologies like quantum computing and advanced neuroimaging will further enhance its potential. For example, AI-powered "virtual biopsies" could become standard for diagnosing conditions like ALS or DMD, eliminating the need for invasive procedures. Additionally, BCIs driven by DL may offer transformative solutions for patients with severe motor impairments, restoring movement or communication capabilities. To realize this future, interdisciplinary collaboration, ethical stewardship, and continued innovation will be essential in ensuring AI is harnessed responsibly and equitably for the benefit of all patients.

Sleep Disorders (Verma): These devices with AI tools will be helpful in treatment personalization, therapy response monitoring, and engagement with therapy in terms of clinical setting. On research standpoint, while comparing two arms for treatment efficacy, these devices can be used to track the possibility of confounding data due to underlying sleep apnea poor quality of sleep impacting the clinical trial results. These devices may be more helpful in research related to cardiac and neurological illnesses as unrecognized sleep apnea or nocturnal hypoxemia may impact the overall outcome of clinical trials.

Stroke (Chen): AI is poised to revolutionize clinical medicine and research, transforming how we diagnose, treat, and study diseases. Clinically, AI can streamline administrative burdens, such as generating detailed clinical documentation from physician-patient conversations or responding to patient inquiries with speed and accuracy. It holds the potential to predict stroke or related complications before they occur, enabling proactive intervention to prevent disease progression. AI can also support decision-making in acute stroke care by predicting outcomes and tailoring recommendations for treatments like thrombolysis or thrombectomy to each patient’s unique characteristics. Additionally, it can forecast stroke recovery outcomes, facilitating early and informed discussions with patients and their families.

From a research perspective, AI offers unprecedented capabilities to analyze electronic medical records (EMRs) efficiently and accurately. It can extract critical data such as stroke timelines, treatment metrics, and patient outcomes, accelerating retrospective studies and driving impactful quality improvement initiatives.

NeurologyLive: What emerging AI concepts or ideas do you believe will come more into the fold over the next year or so?

Epilepsy (Partamian): The next year is expected to witness significant advancements in AI, particularly in the fields of deep learning (such as multimodal data analysis and large language models) and quantum computing, and multimodal AI models. Deep learning will continue to evolve, enabling more accurate analysis of complex data in epilepsy research. Quantum computing will start to make a substantial impact, facilitating faster processing of large datasets and more precise predictions. Multimodal AI models will become more common as well, combining diverse data sources like EEG, MRI, and clinical records to provide a more comprehensive understanding of epilepsy. Additionally, large language models will begin to play a larger role in analyzing unstructured clinical data, uncovering new insights and patterns in data.Large language models (LLMs) like GPT and similar AI systems are well-suited to analyzing unstructured clinical data, such as physician notes, patient histories, and medical reports. These models are able to identify patterns from data, extracting meaningful insights, and even detecting subtle trends in large volumes of text that would be challenging and time-consuming for humans to process manually.

New software integrated with large language models and multimodal analysis tools will be available soon allowing more accurate epilepsy management and providing real-time, non-invasive monitoring of patients outside clinical settings.Other tools will revolutionize the presurgical planning for epilepsy patients and provide valuable information to predict the candidate epileptogenic brain regions enhancing the accuracy of surgical outcomes.Finally, as these technologies evolve, there will be a stronger focus on Ethical AI and Regulatory frameworks to address issues like data privacy and algorithmic bias, ensuring safe and trustworthy deployment in clinical environments. Therefore, we expect significant advancements in AI-driven epilepsy diagnosis, treatment, and research.

Headache and Migraine (Petrušić): Digital twin, as a dynamic digital representation of patients, platforms will revolutionize the headache field. It will be used to perform digital experiments, that will foster the discovery of biomarkers for specific headache disorders and enable groundbreaking steps towards personalized medicine and better optimization of treatment. However, the quality of a digital twin crucially depends on the data available to the model. Therefore, addressing the previously mentioned challenges and barriers is essential. In addition, digital twin decisions must be explainable, so that physicians, patients, and policymakers can understand how recommendations are made.

In addition, AI-driven wearable devices equipped with next-generation biosensors will significantly improve the quantity and quality of collected data, further empowering the digital twin concept. Furthermore, a collaborative ecosystem created among headache experts and AI scientists will lead to the development of improved telemedicine tools that offer AI-supported diagnostics and personalized recommendations.

MS and Demyelinating Disorders (Zivadinov & Dwyer): Multimodal AI is a big one. Historically, we've seen a lot of AI doing specific things in specific modalities - e.g., interpreting a medical record, or quantifying an MRI, or evaluating OCT. The bigger leap is to allow AI to put all that together into a single model that can better understand the full patient, and the tools for this are emerging rapidly. Similarly, digital twins are a big one - if we can really model patients with AI, then we can simulate what will happen to them with different treatments. Finally, I really hope that we see more progress in federated learning - allowing more large-scale national and international collaboration to train larger and more complex models while preserving patient privacy.

Movement Disorders (Mahajan): Honestly, it is tough to predict that. There are two themes: to increase efficiency of systems in healthcare and research and discovery. I wholeheartedly support both. As mentioned above, healthcare can greatly benefit from better use of AI to improve health systems and ease the burden off physicians. I hope that happens soon. I foresee greater improvement in AI based algorithms using wearables and home-based devices studying movement, sleep and other patterns. Where I would like to exercise caution is in rapid deployment in diagnostics and direct patient care without exhaustive testing and physician-input. I would strongly advise that researchers seek to partner with clinicians in developing and testing research, as well.

Neuromuscular (Morren): A notable development is the rapid integration of AI with BCIs, which has shown promise in restoring communication abilities for individuals with conditions like ALS. For instance, the biotech company Synchron, has successfully implanted AI-driven BCIs that enable patients to control digital devices using their thoughts, thereby enhancing their autonomy and quality of life.

Another emerging trend is the application of AI in neurorehabilitation through neuromorphic neuromodulation. This approach utilizes AI algorithms embedded in neural devices to provide responsive neurostimulation, aiding in motor recovery for patients with neurological impairments. By processing neural signals in real-time, these intelligent systems can deliver personalized therapeutic interventions, thereby promoting neuroplasticity and functional restoration.

Additionally, AI is being leveraged to create personalized voice clones for patients at risk of losing their voice due to progressive NM disorders like ALS. This would represent an attractive alternative to other options, including traditional voice banking. Companies like ElevenLabs have developed AI-generated voice synthesis technologies that replicate an individual's unique voice characteristics. This advancement enables patients to maintain their vocal identity, facilitating more natural and emotionally resonant communication even as their voice-altering condition progresses.

Collectively, these emerging AI applications are set to enhance patient autonomy, improve rehabilitation outcomes, and preserve personal identity in the face of substantial NM deficits. As these technologies continue to evolve, they hold the potential to transform the landscape of NM medicine, offering innovative solutions to longstanding and previously refractory impairments.

Sleep Disorders (Verma): These AI tools will get better and better over the time and will play a very important role in development of various clinical prediction models, which will impact the clinical practice and improve outcome. In future, in-lab sleep study may not be needed for the diagnosis sleep disorders due to advancement in AI tools. These AI tools will help with remote monitoring of physiological and treatment response data of sleep disorder patients. There will be more and more AI integration in clinical workflow that we will lead to improved efficiency and outcome in sleep disorder field.

Stroke (Chen): In the rapidly evolving field of artificial intelligence, two distinct types of AI have emerged: "traditional" AI and "foundation" AI. Traditional AI, often referred to as task-specific AI, excels at performing one specific function with high precision. For instance, machine learning models like those used by Viz AI and Rapid AI are trained to identify large vessel occlusions from CTA head and neck scans. These algorithms are highly effective but limited to the single task they were designed for.

In contrast, foundation AI represents the next frontier of innovation. These models are trained on vast and diverse datasets—such as text, images, or other multimodal inputs—not to master a single task but to understand patterns, relationships, and underlying structures in the data. This approach mimics human learning, where insights are gained by observing and interpreting the world without explicit instructions. Once trained, foundation AI models are remarkably versatile, capable of excelling in a wide range of tasks, even those they were not specifically designed for. The most well-known example of this is ChatGPT, which demonstrates the immense potential of foundation AI to adapt and generalize across domains. Since gaining traction in 2023, foundation AI has begun to redefine the landscape of artificial intelligence, positioning itself as the future of the field and paving the way for transformative applications in healthcare and beyond.

Disease Specific

NeurologyLive: Are there any ways AI can improve seizure detection or epilepsy diagnosis?

Epilepsy (Partamian) - AI is transforming the way seizures are detected, and epilepsy is diagnosed, offering numerous benefits to patients and healthcare providers. AI-based algorithms can analyze brain wave patterns from EEG tests with remarkable accuracy and speed, enabling quicker and more precise seizure detection. These technologies can also assist doctors in diagnosing epilepsy by analyzing medical images, such as MRI scans, to identify hidden patterns indicative of the condition.

AI-powered wearable devices could continuously track brain activity and detect seizures in real-time, with the ability to send alerts to caregivers or doctors. Beyond detection, AI can analyze large datasets from electronic health records, genetic tests, and other sources to uncover potential causes of epilepsy and predict the most effective treatments for individual patients.

Additionally, AI-powered chatbots and virtual assistants can support patients by helping them track seizures, manage medications, and stay in touch with their healthcare teams. These tools enhance daily management and empower patients to take an active role in their care.

In conclusion, the integration of AI in epilepsy diagnosis and treatment has the potential to revolutionize the field. It promises better patient outcomes, improved quality of life, and greater efficiency in healthcare delivery. As AI technology continues to advance, it will drive even more innovative solutions, ultimately transforming the lives of people living with epilepsy.

NeurologyLive: Are there any ways AI can help in drug development for migraine/headache disorders?

Headache and Migraine (Petrušić): Yes, AI can significantly streamline drug development for headache patients. AI techniques have already facilitated virtual screening, drug design, and drug-target interaction modeling, establishing novel paradigms for predicting pharmacodynamic and pharmacokinetic properties. Tension-type headache, cluster headache, and trigeminal autonomic cephalalgias continue to lack targeted therapeutic options, representing an area of high unmet need. AI can accelerate the discovery of novel therapeutic targets for these conditions by leveraging large datasets from genomics, proteomics, and clinical data to identify potential drug targets that may have been previously overlooked. Furthermore, advanced machine learning and deep learning algorithms could analyze existing drugs for potential efficacy in headache management and propose multidrug combinations for a more personalized and effective approach in non-responsive patients. Additionally, AI can analyze real-world data to monitor long-term drug safety and effectiveness.

NeurologyLive: Are there any ways AI can help in drug development for MS, including progressive forms of the disease?

MS and Demyelinating Disorders (Zivadinov & Dwyer): Absolutely, this is one of the big success stories of AI, recently, with part of a Nobel prize awarded this year for predicting protein folding with AI. These kinds of techniques are also being used to predict drug targets and interactions and will hopefully greatly speed up the pipeline for pharma companies. Less directly, but perhaps equally importantly, I think the non-biased way unsupervised AI can be used can allow us to potentially identify disease mechanisms we might be missing and point us toward different avenues for therapies that we might be overlooking right now. This is a huge area that will grow exponentially. Trials can be better designed and simulated respect to the outcomes and endpoints when using AI, the mechanism of action of different drugs can be better explored in relation to pathophysiological mechanisms. Finally, new pathways can be discovered.

NeurologyLive: Are there any ways AI can help in management or drug development of movement disorders?

Movement Disorders (Mahajan): I am excited about the possibility of drug repurposing for rare movement disorders. We now know that patients with some rare disorders may respond specifically to certain known medications, better. As an example, ataxia has no FDA approved medications for symptomatic improvement. We know that a newly discovered mutation (FGF14 in SCA 27b) may be responsible for a number of patients previously diagnosed as idiopathic late-onset cerebellar ataxia. Through clinical experience published in case series, these patients respond symptomatically to Dalfampridine, a medication that is available and FDA-approved for walking issues in multiple sclerosis. Similarly, Docosahexaenoic acid is a beneficial replacement treatment for SCA-38.15 AI can help establish such symptomatic relationships. Long term, my hope is for the use of AI to identify such drugs for disease modification in neurodegenerative disease.

NeurologyLive: Are there ways that AI can help in treating more rare, genetic neuromuscular disorders?

Neuromuscular (Morren): AI offers promising avenues for advancing the treatment of rare genetic neuromuscular disorders, addressing challenges like limited patient populations and complex disease presentations. One key area is drug discovery and development. AI algorithms can analyze large-scale genomic and proteomic data to identify novel therapeutic targets for rare diseases. For example, ML models can pinpoint specific mutations or pathways driving these disorders, enabling the design of highly targeted therapies.

Another application is personalized treatment planning. AI can integrate diverse data sources, including genetic profiles, disease phenotypes, and patient-specific clinical information, to generate tailored treatment recommendations. For instance, in conditions like DMD, AI tools can assess muscle imaging data and predict responses to emerging therapies, helping clinicians refine their approach to individual patients. Similarly, AI-powered predictive models can forecast disease progression, allowing for earlier intervention and optimized care strategies, which are crucial in managing the rapid decline sometimes seen in rare, genetic neuromuscular disorders.

AI also facilitates the design and execution of clinical trials for rare diseases. Traditional trials are often hampered by small patient populations and logistical challenges. AI can help overcome these barriers by identifying suitable candidates through advanced pattern recognition and by simulating trials to predict outcomes before live testing. Furthermore, AI-driven digital biomarkers derived from wearable devices or remote monitoring systems can provide continuous data on patient status, reducing the need for frequent in-person assessments. These innovations not only improve trial efficiency but also expand access to experimental therapies for patients with rare neuromuscular disorders, potentially transforming the treatment landscape.

NeurologyLive: Are there any ways AI can help in drug development for sleep disorders or even personalization of treatment?

Sleep Disorders (Verma): AI has capability to analyze a large amount of data including clinical, bio-psychosocial, genetic, biochemical, sleep study, therapy response data all together leading to clinical predictions which can be helpful for personalization of treatment for individuals. The area is helping in drug development in other fields, and it could be helpful in sleep disorders as well especially for insomnia, hypersomnia, narcolepsy, and leg movement disorders. The AI tool can alert clinician early enough to monitor therapy response data and intervene appropriately.

NeurologyLive: What sort of large-scale impact can AI have on triage and efficiency of stroke care?

Stroke (Chen): Artificial intelligence holds immense potential to transform triage and efficiency in stroke care, a field where every second can mean the difference between recovery and permanent disability. By combining unparalleled speed, precision, and scalability, AI can bridge the gap between resource-rich urban centers and under-resourced rural hospitals, ensuring equitable access to high-quality care. Its ability to process complex data rapidly and consistently could revolutionize acute stroke management, enabling faster, safer decisions in high-stakes situations.

System Integration: How AI Is Weaving Itself into Neurology

Key Takeaways

Article Highlights

Disease Specific