The Role of AI in the Management of Movement Disorders

• Abstract
• AI-aided Deep Phenotyping
• AI-enhanced Movement Disorders Diagnostics
• AI in the Identification of Neuronal α-synuclein Disease
• AI-guided Clinical Decisions and Treatment Selection
• AI and Deep Brain Stimulation (DBS)
• AI in Symptom Monitoring
• AI in the Movement Disorders Clinic
• AI in Telemedicine for Movement Disorders
• Ethical Considerations and Quality of AI Research
• Conclusions
• References

Abstract

Artificial intelligence (AI) has emerged as a transformative force in the management of movement disorders. This review explores the various applications of AI across the spectrum of care, from diagnosis to clinical workflows, treatment, and monitoring. Recent advancements include deep phenotyping tools like the Next Move in Movement Disorders (NEMO) project for hyperkinetic disorders, diagnostic platforms such as DystoniaNet, and biomarker identification systems for early Parkinson's disease detection. AI may revolutionize treatment selection through technologies like DystoniaBoTXNet and adaptive deep brain stimulation systems. For symptom monitoring, innovations like the Emerald device and smartphone-based assessment tools enable continuous, objective evaluation. AI may also enhance patient care through improved telemedicine capabilities and ambient listening. Despite these promising developments, recent critiques highlight methodological concerns in AI research, emphasizing the need for rigorous validation and transparency. The future of AI in movement disorders requires balancing technological innovation with clinical expertise to improve patient outcomes.

Artificial intelligence (AI) has emerged as a transformative force in the field of movement disorders, offering new possibilities for identifying correct phenomenology, limiting diagnostic error, offering personalized treatment, and improving overall patient care. This article explores the various applications of AI in managing movement disorders, highlighting recent advancements and their potential impact on clinical practice.

AI-aided Deep Phenotyping

One of the greatest challenges in the field of movement disorders is the correct identification of complex phenomenologies. Deep phenotyping has emerged as a crucial approach in advancing our understanding and management of these disorders, particularly Parkinson's disease (PD),[1] [2] which is the most commonly seen movement disorder in neurology practices.[3] This comprehensive method integrates multiple assessment tools, including clinical evaluations, biological markers, genetic information, imaging techniques, and sensor-based technologies, to provide a more detailed and nuanced characterization of these complex neurological conditions that can exhibit a broad phenotypic spectrum.

The potential benefits of advancing deep phenotyping in movement disorders are multifaceted. It offers enhanced understanding by providing new insights into various aspects of these disorders, examining how they affect multiple domains of a patient's life and health. This approach also improves diagnostic accuracy, helping clinicians differentiate between various movement disorders and potentially identify subtypes within conditions, like PD, where hypokinesia is the only consistent feature among all patients. Furthermore, deep phenotyping enables more personalized treatment strategies, allowing clinicians to tailor interventions based on individual disease characteristics.[2]

AI has arisen as a valuable tool in accelerating our efforts to characterize a myriad of movement disorders. The Next Move in Movement Disorders (NEMO) project,[4] led by Dr. Marina de Koning-Tijssen, has made significant strides in applying AI to hyperkinetic movement disorders, in particular. Initiated in 2018 at the University Medical Center Groningen, the NEMO project aims to develop a computer-aided classification tool for precise phenotyping of hyperkinetic movement disorders and to enhance our understanding of brain pathophysiology through advanced neuroimaging techniques.

This group has employed a comprehensive, multimodal approach to data collection, integrating electromyography (EMG), accelerometry, three-dimensional video recordings, and various neuroimaging techniques including T1-weighted structural MRI, resting-state fMRI, motor task fMRI, and 18F-FDG PET scans. This holistic approach allows for a more thorough analysis of movement disorders, capturing both physical manifestations and underlying physiology.[4] [5]

The computer-aided classification tool being developed has the potential to reduce inter- and intra-observer variability in phenotyping, improve diagnostic accuracy, enhance treatment selection and monitoring, and facilitate personalized medicine approaches. The NEMO project is ongoing, with plans to evaluate which modalities, features, and models are most efficient to implement in clinical settings, build machine learning approaches to assist neurologists in phenotype classification, and address more complex mixed movement disorder phenotypes.

AI-enhanced Movement Disorders Diagnostics

Many other groups are taking different approaches toward better application of AI in diagnosing movement disorders. Perhaps one of the most exciting developments in the realm of automated image analysis and pattern recognition is the work led by Dr. Kristina Simonyan on DystoniaNet,[6] DystoniaBoTXNet,[7] and DystoniaDBSNet,[8] all of which will be discussed in greater detail below. These AI-powered platforms have the potential to revolutionize the diagnosis and treatment planning for patients with dystonia.

Dystonia is one of the most common and disabling hyperkinetic movement disorders,[9] and is characterized by involuntary muscle contractions that lead to abnormal postures and movements.[10] It is also one of the hardest movement disorders to diagnose correctly, with some patients waiting up to a decade for a correct diagnosis.[6] Although specific clinical scales exist to characterize the distribution and severity of dystonia,[11] they do not aid in the accurate diagnosis of dystonia, nor do they distinguish dystonia from its many mimics.

DystoniaNet can potentially address these shortcomings. This AI-powered system analyzes structural brain MRIs to identify subtle microstructural changes associated with dystonia, effectively identifying a neural network biomarker for dystonia. In the study by Valeriani and Simonyan,[6] DystoniaNet achieved a 98.8% accuracy in diagnosing dystonia, significantly outperforming experienced clinicians. The system can process and analyze an MRI scan in just 0.36 seconds. A prospective randomized study (NCT05317390) is currently ongoing, aiming to validate the performance of this algorithm. If positive, DystoniaNET will allow the accurate, objective, and fast diagnosis of dystonia in the clinical setting.

A similar approach has been employed by the Automated Imaging Differentiation for Parkinsonism (AIDP) Study Group, and the results of their work was recently published in JAMA Neurology. In what is likely to become a landmark publication, Vaillancourt et al report on their prospective multicenter cohort study that evaluated the discriminative performance of AIDP using 3-T diffusion MRI and support vector machine learning.[12] The study involved 249 patients with established diagnoses of PD, multiple system atrophy, and progressive supranuclear palsy, with a subset of autopsy cases for neuropathological validation. AIDP demonstrated robust differentiation capabilities, achieving high area under the receiver operating characteristic curve values for distinguishing PD from atypical parkinsonism (0.96), multiple system atrophy from progressive supranuclear palsy (0.98), PD from multiple system atrophy (0.98), and PD from progressive supranuclear palsy (0.98). The model's predictions were neuropathologically confirmed in 93.9% of cases. These findings suggest that AIDP could be a valuable tool in the diagnostic workup for these parkinsonian syndromes, with the potential to change clinical practice in the very near future.

There is also great enthusiasm with the development of wireless technologies for the early diagnosis of PD,[13] perhaps even at the prodromal stage, when disease-modifying interventions may be most effective. In a landmark study published in Nature Medicine, the group led by Dr. Dina Katabi developed an AI model to detect PD and track its progression using nocturnal breathing signals.[13] This model was evaluated on a large dataset of 7,671 individuals from multiple hospitals in the United States and public datasets. The AI system demonstrated impressive performance, with an area under the curve of 0.90 and 0.85 on held-out and external test sets, respectively, for PD detection. Moreover, the model could estimate PD severity and progression in alignment with the Movement Disorder Society Unified Parkinson's Disease Rating Scale (MDS-UPDRS).

The group led by Dr. Brit Mollenhauer has also made significant strides in PD diagnostics with their work on proteomics and AI for biomarker identification.[14] Their study utilized machine learning techniques to analyze plasma proteomic data, identifying a panel of eight proteins that could accurately distinguish PD patients from healthy controls. Moreover, this biomarker panel was able to predict the development of PD in individuals with REM sleep behavior disorder up to 7 years before motor symptom onset.

Using a different approach to the one employed by the Mollenhauer group, the work from a few years ago by Makarious et al[15] is yet another impressive example of the potential of machine learning to predict PD. The researchers developed a model combining clinical, demographic, genetic, and transcriptomic data from the Parkinson's Progression Markers Initiative (PPMI) dataset. They used an automated machine learning approach called GenoML to compare various algorithms and select the best performing one. The final model, which used an AdaBoost classifier, achieved high accuracy in predicting PD diagnosis, outperforming models based on single data modalities.

This study highlights the importance of smell identification tests and polygenic risk scores in predicting PD, while also incorporating numerous smaller-effect genetic variants and transcripts. The model was validated on an external dataset (PDBP) and showed good generalizability. The researchers also conducted additional analyses, including the construction of gene expression networks and exploration of drug–gene interactions. Although the model shows promise for identifying high-risk individuals in large-scale settings, the authors do acknowledge limitations, such as the lack of diversity in the sample and the need for further validation in more diverse populations.

Whereas most of these technologies are still undergoing development and validation, large language models (LLMs) are now ubiquitous, publicly available, and free, at least in its basic versions. As of the time of publication of this manuscript, LLMs can already be used to increase the efficiency of movement disorders practice as it relates to diagnosis. From a practical, strictly clinical perspective, LLMs (like ChatGPT) can assist the neurologist (regardless of whether they are movement specialists or not) in expanding the differential diagnosis for challenging cases.[16] Indeed, by leveraging vast databases of medical information, AI tools can already help clinicians consider rare conditions or atypical presentations that might otherwise be overlooked.[17] This is particularly valuable in the field of neurogenetics, for example, where genotype–phenotype correlations are increasingly complex.[18]

AI in the Identification of Neuronal α-synuclein Disease

The field of movement disorders is undergoing a significant paradigm shift with the emergence of new biological definitions for synucleinopathies, particularly PD and dementia with Lewy bodies (DLB). This new framework[19] introduces the concept of “neuronal α-synuclein disease” (NSD), anchored on two key biological components: (1) evidence of synuclein pathology using seed amplification assays (SAAs) in cerebrospinal fluid or skin samples, and (2) evidence of dopaminergic denervation, typically assessed through dopamine transporter imaging (DAT-SPECT), which can identify neurodegeneration in the nigrostriatal pathway.

Artificial intelligence is accelerating the shift toward widespread adoption[20] of this new definition. Gibbons et al, for example, have developed a novel, AI-driven method for the continuous, quantitative analysis of phosphorylated alpha-synuclein (P-SYN) within cutaneous axons, aiming to improve the diagnosis and monitoring of synucleinopathies.[21] The study involved 100 participants, including 80 patients with various synucleinopathies and 20 healthy controls, all of whom underwent skin sampling. These samples were immunostained for P-SYN before being digitized using confocal microscopy. The results were impressive, with a 99.3% diagnostic concordance between traditional pathologic review and AI-augmented reading. P-SYN was detected in all 80 synucleinopathy cases and in none of the healthy controls. The AI-assisted method proved highly reproducible and enhanced the ability to differentiate between synucleinopathy subtypes. Looking ahead, the success of this AI-driven approach to skin biopsy analysis opens new possibilities for early diagnosis and monitoring of synucleinopathies. In their paper, Gibbons et al suggest that this method could serve as a quantifiable biomarker for disease severity and progression, potentially accelerating drug development for these disorders. Replication of their findings by other groups is still required.

Another powerful tool to ascertain the presence of synuclein in biospecimens is through SAAs, as mentioned previously.[19] A current shortcoming of this assay, however, is that test results are interpreted through manual, time-consuming, and potentially inconsistent methods. AI-QuIC, an AI-driven platform, has been reported to automate this analysis, potentially eliminating any subjective bias.[22] This deep-learning-based model, trained on a well-labeled real-time, quaking-induced conversion dataset of over 8,000 wells, achieved over 98% sensitivity and 97% specificity in classifying true positive, false positive, and negative reactions. By learning from raw fluorescence data, AI-QuIC simplified the SAA analysis workflow, providing a robust, scalable, and consistent diagnostic solution, and opening a potential path for the widespread application of this technology in the clinical space.

Finally, within the realm of functional neuroimaging, AI has also demonstrated potential in improving the interpretation of DAT-SPECT for the diagnosis of parkinsonian syndromes. Several machine learning algorithms have been developed to automatically quantify dopamine transporter binding, potentially reducing inter-rater variability and improving diagnostic accuracy.[23] [24] [25]

AI-guided Clinical Decisions and Treatment Selection

AI-powered electronic medical records (EMRs) could soon generate therapeutic suggestions based on comprehensive patient data.[16] [26] This could include analyzing test results, medication history, and comparing individual patient data to aggregated, deidentified data from other patients. Such capabilities could help identify outliers in disease progression and medication response, potentially distinguishing conditions like early multiple system atrophy from idiopathic PD.

One of the barriers preventing this goal is the fact that EMR data is typically incomplete or fragmented. To potentially address this gap, Huang et al have introduced VisAGE (Visualization Assisted by Knowledge Graph Enrichment),[27] a platform that enriches patient records with a knowledge graph built from external databases that include protein–protein interactions, genomic data, and drug–chemical associations. The evaluation of VisAGE using the PPMI dataset demonstrated its ability to generate visualizations that cluster similar patients more effectively than baseline methods that do not modify the original database. Further assessment using drug and symptom enrichments revealed that VisAGE can create more detailed and nuanced groupings of PD patients. This innovative approach exemplifies how bioinformatics is poised to transform the assessment and treatment of patients with movement disorders.

Implementation of technologies like VisAGE, however, does pose several privacy risks that must be considered. One important risk is the possibility of patient re-identification, as the linkage of clinical and genomic data (which is inherently identifying) could serve as a “fingerprint” for an individual, particularly if the deidentified data is cross-referenced with external sources. Furthermore, without safeguards, VisAGE's enriched visualizations could expose sensitive conditions such as mental health, HIV status, or genetic predispositions, among others, either directly or through inferences made from linked data (for example, through the identification of medications that have very specific indications). Allowing access to VisAGE's visualizations to multiple user types could also infringe upon the privacy rule of the Health Insurance Portability and Accountability Act (HIPAA). Finally, technologies like VisAGE may not strictly adhere to HIPAA's security rule, and health systems would need to be prepared in case of breaches.

Despite any potential barriers or concerns, it is likely that EMR–AI integrations will become a reality in the near future.[26] [28] [29] In the interim, however, the use of AI in guiding treatment selection for movement disorders is an area of rapidly growing interest. The work by the Simonyan group with DystoniaBoTXNet[7] and DystoniaDBSNet[8] exemplifies this trend, offering personalized treatment recommendations for dystonia patients. Building on the success of DystoniaNet,[6] they developed DystoniaBoTXNet,[7] an AI system designed to predict the efficacy of botulinum toxin treatment in dystonia patients. This tool analyzes brain MRI data to identify biomarkers associated with treatment response, potentially allowing clinicians to personalize treatment plans more effectively. DystoniaBoTXNet[7] identified key brain regions as components of the treatment biomarker, including the superior parietal lobule, inferior and middle frontal gyri, and corpus callosum. Interestingly, five of the eight regions identified by DystoniaBoTXNet overlapped with the diagnostic biomarker regions found by DystoniaNet,[6] providing additional evidence for their role in dystonia pathophysiology. With an average processing speed of 19.2 seconds, the system has been reported to demonstrate 100% sensitivity, 86.1% specificity, and high overall accuracy in predicting treatment outcomes (96.3%), offering a valuable tool for clinical decision-making, once publicly available.

AI and Deep Brain Stimulation (DBS)

There have been recent advances in the application of AI to assess candidacy for DBS, as well as to optimize DBS settings. Regarding the former, the latest tool introduced by the Simonyan group is DystoniaDBSNet,[8] a deep learning algorithm developed to predict the efficacy of DBS in dystonia patients using preoperative structural brain MRIs. Trained and validated on a cohort of 130 patients with various dystonia phenotypes, DystoniaDBSNet has shown to identify a neural biomarker for DBS treatment outcome, encompassing regions like the precentral and middle frontal gyri, superior frontal gyrus, anterior cingulate cortex, thalamus, and postcentral gyrus. The model achieved an overall accuracy of 96.0%, with 100% sensitivity and 85.7% specificity in predicting DBS treatment outcome.[8] This fully automated, objective tool has the potential to enhance clinical decision-making in DBS candidate selection and improve clinical care for dystonia patients, who have very limited treatment options at the present time.[30]

AI is now also capable of informing and personalizing the delivery of DBS therapy to patients with PD. In a recent proof-of-concept study[31] exploring adaptive deep brain stimulation (aDBS) for PD, AI played a pivotal role in enabling personalized therapy. First, AI was critical for biomarker identification, where researchers employed a data-driven analysis pipeline that combined non-parametric cluster-based permutation analysis with machine learning techniques. This sophisticated approach allowed them to identify individualized neural biomarkers of PD symptoms by searching the frequency space of field potentials in both the subthalamic nucleus and the sensorimotor cortex, ultimately pinpointing physiological signals that optimally predicted the occurrence of each patient's most bothersome motor signs. The AI component proved invaluable in predicting medication states in clinical settings and tracking symptom fluctuations in at-home environments, with the AI models outperforming conventional analyses. Second, after having established individualized biomarkers, they were then leveraged to develop personalized aDBS algorithms. The machine learning techniques were again crucial in creating these algorithms, enabling them to record, analyze, and respond to the unique brain activity associated with each patient's symptom state. Finally, this personalized AI-powered system was able to detect and respond to abnormal brain rhythms associated with PD in real time. Based on the patient's current symptom state, the aDBS system automatically adjusted stimulation amplitude or other parameters. This study opens the door towards precision approaches in DBS for PD, but technical, regulatory, cost-related, and ethical barriers will need to be resolved before pursuing the wide-scale clinical deployment and adoption of this potentially revolutionary advancement.[32]

AI in Symptom Monitoring

Continuous symptom monitoring is crucial for effective management of movement disorders, and AI has opened new possibilities in this area. As discussed previously, the work[13] by the Katabi group on what is now known as the Emerald device represents an example of a significant advance in passive, at-home monitoring of movement disorder symptoms.

The Emerald device (https://emeraldinno.com/) uses radio waves to monitor a patient's movements, gait, and even breathing patterns without requiring the patient to wear any sensors. This technology can provide continuous, objective data on symptom fluctuations and disease progression in the home setting. This group demonstrated the device's ability to detect subtle gait changes in PD patients, potentially allowing for earlier intervention and more precise medication adjustments.[33] The non-invasive nature of this technology addresses privacy concerns that can be associated with video-based monitoring systems (see [Table 1]), making it a promising tool for both clinical practice and research. The real-time passive and continuous gathering of enormous amounts of raw data over prolonged periods of time, however, can raise privacy concerns of its own, mainly including: (1) the possibility that monitoring could occur without conscious awareness or consent at every moment; (2) the collection of data from anyone moving within its range (not just the intended subject); (3) the potential to infer health status, mood, or even private activities; (4) access by unauthorized parties if the data is not sufficiently encrypted, particularly in the absence of clear regulation and oversight for these type of novel technologies; and (5) the possibility of re-identification, given the granularity of the data being collected.

Whereas the work by the Katabi group relies on the interpretation of radio signals, many different platforms for video analysis are in development, which have the advantage of being widely accessible to the general population without the need to purchase and install hardware. The automated analysis of image and video data that enables machines to “see” and make decisions based on sensed images is now referred to as computer vision.[11]

Pose estimation, a computer vision method based on convolutional neural networks (CNNs), is particularly relevant to the study of movement disorders. This technique extends object detection by incorporating a topological model of key human body joints, enabling markerless and unobtrusive tracking of body configuration and movement. The application of computer vision in movement disorders has been diverse, encompassing the analysis of various tremors (including essential, dystonic, and parkinsonian), bradykinesia, and facial movement disorders such as blepharospasm and tics. Additionally, it has been employed to study gait and postural disorders, evaluate handwriting in neurodegenerative conditions, and assess head deviations in cervical dystonia.[11]

As for recent notable examples within the movement disorders space, in February 2025, researchers from the Fixel Institute at the University of Florida published details about VisionMD, an open-source software for automated video-based analysis of motor tasks.[34] VisionMD uses deep learning to track body movements and compute kinematic features that quantify symptom severity in PD and other movement disorders. This tool offers a user-friendly, customizable framework that enables clinicians and researchers to objectively evaluate motor symptoms without the need for specialized hardware. Islam et al from the movement disorders group at the University of Rochester have also previously reported on using smartphone-based assessments to quantify PD severity.[35] By analyzing data from finger-tapping tests and gait assessments performed on a smartphone, their AI algorithm could accurately estimate MDS-UPDRS scores, potentially allowing for more frequent and convenient disease monitoring.

Whereas computer vision methods may offer significant clinical benefits to patients with movement disorders, they also raise several notable privacy concerns. These concerns stem from the sensitive nature of video data and the context in which it is collected. [Table 1] summarizes the key privacy concerns pertaining to this technology, as well as possible strategies to address them.

AI in the Movement Disorders Clinic

Beyond diagnostics, several AI tools are available as of the publication of this article to improve the efficiency and workflow of movement disorders clinics.[16] Ambient listening is one of these notable innovations and refers to the use of AI systems that passively capture, interpret, and transcribe spoken interactions between movement specialists and patients in real time, without requiring typing into the EMR.[36] Unlike traditional dictation tools, ambient listening operates seamlessly in the background during medical encounters, leveraging natural language processing and voice recognition technologies to generate structured, clinically relevant documentation directly from the conversation. This approach not only alleviates the administrative burden but also enhances the accuracy and completeness of clinical records, allowing providers to focus more fully on patient engagement and care. As ambient listening becomes increasingly integrated into movement disorders workflows, it offers the potential to reduce burnout and foster a more patient-centered clinical environment, while simultaneously raising important considerations regarding data privacy and security. [Table 2] depicts a few ambient listening clients that are currently available.

Large language model–based responses to patients' in-basket messages represent another promising AI innovation for improving movement disorders workflows, as demonstrated by a recent cross-sectional study at NYU Langone Health. In this study,[37] 16 primary care physicians evaluated 344 patient message–response pairs—some drafted by generative AI chatbots and others by healthcare professionals (HCPs)—without knowing the source of each response.

Physicians rated both AI and HCP responses favorably, but AI drafts consistently scored higher for communication style and empathy. Although both types of responses were similar in informational accuracy, completeness, and relevance, AI drafts were often described as more personalized and empathetic. However, AI responses were also 38% longer, more complex, and written at a higher reading level (eighth grade versus sixth grade for HCPs), raising concerns about accessibility for patients with lower health or English literacy.

Overall, these findings suggest that EMR-integrated AI chatbots can produce high-quality, empathetic, and usable draft responses to patient messages, potentially reducing clinician inbox burden and improving patient–provider communication. However, the increased length and complexity of AI-generated messages highlight the need for further refinement to ensure accessibility and health equity, especially for patients with limited literacy. Future research should focus on optimizing AI outputs for patient comprehension, quantifying efficiency gains, and addressing potential biases and risks of inaccurate information.

AI in Telemedicine for Movement Disorders

The COVID-19 pandemic accelerated the adoption of telemedicine in movement disorders care, and AI has played a crucial role in enhancing these remote assessments. AI has the potential to augment telemedicine consultations for movement disorders by analyzing video recordings of patient movements in real time to provide objective measures of symptom severity, which would improve the accuracy of remote assessments.[11] [16] As an example, Li et al reported in 2018[38] a video-based system to objectively assess levodopa-induced dyskinesias in patients with PD, which was able to provide an objective means of quantifying dyskinesia severity, and had a responsiveness that was as good as or better than the validated Unified Dyskinesia Rating Scale (UDysRS).

As the field waits for technologies like this to become widely available, a feasible model that can be employed quickly and easily is the use of validated wearable sensors and biosensors, which offer a powerful means to augment neurological examinations during virtual consultations and enable remote monitoring of both motor and non-motor symptoms in patients with movement disorders. AI can play a crucial role in transforming the wealth of self-tracking data generated by these devices into actionable insights, assisting general neurologists (as well as movement disorders specialists) in making informed clinical decisions, such as optimizing medication timing and dosage.[39]

Ethical Considerations and Quality of AI Research

Although the potential of AI in movement disorders management is immense, it is crucial to approach this rapidly growing body of literature with a critical eye. Several voices within the movement disorders space have argued for the importance of maintaining human oversight in clinical decision-making and ensuring that AI tools are developed and implemented in ways that enhance, rather than replace, the clinician–patient relationship.[40] [41] A summary of the ethical considerations and their implications pertaining the implementation of AI in movement disorders care can be found in [Table 3].

Furthermore, it is critical that the quality of the research published is upheld to the highest standards, particularly at a time when it is difficult to keep up with the rate of publication.[42] Dzialas et al, for example, provided a recent critical assessment of the quality of neuroimaging-based AI studies in PD over the last few years.[43] They conducted a comprehensive review of 244 studies that applied AI to neuroimaging for PD diagnosis, prognosis, or intervention. They evaluated these studies based on five minimal quality criteria (MQC) related to data handling, model complexity, and reporting practices. Their findings revealed significant quality issues in the field, with only 20% of the evaluated studies passing all five MQC.

The findings have significant implications for the broader field of AI applications in movement disorders. Although many studies focused on distinguishing PD from healthy controls or atypical parkinsonian syndromes, the quality issues identified suggest that the reported accuracies may be overly optimistic. This calls into question the readiness of these AI tools for clinical application in diagnosis. The review also found that prognostic and interventional studies were sparse, highlighting a gap in the current research landscape. The lack of external validation in most studies also raises concerns about the generalizability of AI models to diverse clinical settings and patient populations, particularly minorities. Finally, the poor adherence to quality criteria makes it difficult to reproduce and validate the findings of many studies, hindering the translation of AI research into clinical practice.

To ensure the reliability and clinical applicability of AI-driven neuroimaging tools for PD, the authors advocate for enhanced quality control mechanisms throughout the research process. They argue for rigorous methodologies and interdisciplinary collaborations between clinicians and AI experts to produce robust, interpretable, and clinically relevant tools. This critical assessment by Dzialas et al should serve as a wake-up call for the field, emphasizing the need for improved methodological rigor in AI-based neuroimaging studies for movement disorders.

Even with high-quality AI research, several concerns and precautions must be considered when deciding to incorporate AI tools into the management of patients with movement disorders.[16] Patient privacy is paramount, as the technology presents a risk of unauthorized access or data breaches involving sensitive protected health information. Homegrown, intramural large language models can address this concern, although the possibility of a data breach can never be completely excluded.[44]

Equity is another very important factor to keep in mind. The complexity of abnormal movements, combined with incomplete or biased training data, can limit the accuracy of AI tools and lead to non-generalizable or inappropriate treatment recommendations. The design, training, and deployment of AI systems can have particularly profound and complex implications for racial, ethnic, gender, or sexual minority populations, among others, as these systems often reflect and reinforce stereotypes and discrimination present in their training data. As a result, biased AI algorithms can perpetuate inequities, such as over-diagnosing functional movement disorders in trans African American women, limiting the referral of Hispanic patients for deep brain stimulation surgery, or requiring more advanced disease in Asian Americans before recognizing conditions like parkinsonism, to list a few examples. Fortunately, recognizing these shortcomings also presents the opportunity to design AI systems with equity in mind, which should help to uncover and address hidden patterns of discrimination, meet specific cultural and social needs, and increase access to information, care, and services for underrepresented populations.

Currency of the training data must also be considered, as AI models have knowledge cutoff dates, potentially hindering their ability to provide up-to-date information on rapidly evolving fields like neurogenetics or recent clinical trial results. Lastly, the accessibility of AI tools to patients raises concerns about self-diagnosis and the potential for patients to rely on these tools instead of seeking professional medical care,[45] particularly when the phenomenology is ambiguous and demands a specialist's expertise.

Conclusions

The integration of AI into the management of movement disorders represents a paradigm shift in how we approach diagnosis, treatment, and patient care. From improving diagnostic accuracy and treatment selection to enhancing symptom monitoring and patient safety, AI technologies offer numerous opportunities to improve outcomes for individuals with movement disorders. However, as highlighted by recent critiques, it is crucial that we approach this rapidly evolving field with both enthusiasm and caution. As we continue to develop and refine AI tools for movement disorders, we must prioritize rigorous methodology, transparency, and a focus on clinical applicability.

The future of AI in movement disorders management is bright, but realizing its full potential will require ongoing collaboration between clinicians, researchers, and technology experts. It is important that the field works in unison not only to prevent the duplication of efforts that are geared toward a common goal, but also to learn from each other's victories and failures. By combining clinical expertise with the power of AI, we may be at the forefront of an unprecedented revolution that can improve the lives of individuals living with these disabling neurologic conditions.

Conflict of Interest

A.D. has received personal compensation for serving as a Consultant for Manifold Bio, SPARK-NS, and Genus Lifesciences. He has received personal compensation for serving on a Scientific Advisory Board for Supernus Therapeutics, Abbvie, and Amneal. The University of Pennsylvania (A.D.'s employer) has received research support from Teva Pharmaceuticals, Cerevel Therapeutics, Prevail Therapeutics, Lundbeck Therapeutics, AskBio, Annovis Bio, Ono pharmaceuticals, and Amylyx therapeutics. He has served as Site PI for clinical trials funded by these companies. A.D. has received publishing royalties from UpToDate and Elsevier for publications relating to health care.

• References

• 1 Dorsey ER, Omberg L, Waddell E. et al. Deep phenotyping of Parkinson's disease. J Parkinsons Dis 2020; 10 (03) 855-873
  CrossrefPubMedSearch in Google Scholar
• 2 Vizcarra JA, Yarlagadda S, Xie K, Ellis CA, Spindler M, Hammer LH. Artificial intelligence in the diagnosis and quantitative phenotyping of hyperkinetic movement disorders: a systematic review. J Clin Med 2024; 13 (23) 7009
  PubMedSearch in Google Scholar
• 3 Dorsey ER, Sherer T, Okun MS, Bloem BR. The emerging evidence of the Parkinson pandemic. J Parkinsons Dis 2018; 8 (s1): S3-S8
  CrossrefPubMedSearch in Google Scholar
• 4 van der Stouwe AMM, Tuitert I, Giotis I. et al. Next move in movement disorders (NEMO): developing a computer-aided classification tool for hyperkinetic movement disorders. BMJ Open 2021; 11 (10) e055068
  PubMedSearch in Google Scholar
• 5 Dalenberg JR, Peretti DE, Marapin LR, van der Stouwe AMM, Renken RJ, Tijssen MAJ. Next move in movement disorders: neuroimaging protocols for hyperkinetic movement disorders. Front Hum Neurosci 2024; 18: 1406786
  PubMedSearch in Google Scholar
• 6 Valeriani D, Simonyan K. A microstructural neural network biomarker for dystonia diagnosis identified by a DystoniaNet deep learning platform. Proc Natl Acad Sci U S A 2020; 117 (42) 26398-26405
  PubMedSearch in Google Scholar
• 7 Yao D, O'Flynn LC, Simonyan K. DystoniaBoTXNet: novel neural network biomarker of botulinum toxin efficacy in isolated dystonia. Ann Neurol 2023; 93 (03) 460-471
  PubMedSearch in Google Scholar
• 8 Simonyan K, Yao D. DystoniaDBSNet: a novel deep learning biomarker of predictive treatment outcomes in dystonia. Brain Stimul 2025; 18 (01) 274
  PubMedSearch in Google Scholar
• 9 Junker J, Hall J, Berman BD. et al; Dystonia Coalition Study Group. Longitudinal predictors of health-related quality of life in isolated dystonia. J Neurol 2024; 271 (02) 852-863
  PubMedSearch in Google Scholar
• 10 Albanese A, Bhatia K, Bressman SB. et al. Phenomenology and classification of dystonia: a consensus update. Mov Disord 2013; 28 (07) 863-873
  CrossrefPubMedSearch in Google Scholar
• 11 Friedrich MU, Relton S, Wong D, Alty J. Computer vision in clinical neurology: a review. JAMA Neurol 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 12 Vaillancourt DE, Barmpoutis A, Wu SS. et al; AIDP Study Group. Automated imaging differentiation for Parkinsonism. JAMA Neurol 2025; e250112 . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 13 Yang Y, Yuan Y, Zhang G. et al. Artificial intelligence-enabled detection and assessment of Parkinson's disease using nocturnal breathing signals. Nat Med 2022; 28 (10) 2207-2215
  PubMedSearch in Google Scholar
• 14 Hällqvist J, Bartl M, Dakna M. et al. Plasma proteomics identify biomarkers predicting Parkinson's disease up to 7 years before symptom onset. Nat Commun 2024; 15 (01) 4759
  PubMedSearch in Google Scholar
• 15 Makarious MB, Leonard HL, Vitale D. et al. Multi-modality machine learning predicting Parkinson's disease. NPJ Parkinsons Dis 2022; 8 (01) 35
  PubMedSearch in Google Scholar
• 16 Deik A. Potential benefits and perils of incorporating ChatGPT to the movement disorders clinic. J Mov Disord 2023; 16 (02) 158-162
  CrossrefPubMedSearch in Google Scholar
• 17 Mohajer B. Faster, more practical, but still accurate: deep learning for diagnosis of progressive supranuclear palsy. Radiol Artif Intell 2024; 6 (03) e240181
  PubMedSearch in Google Scholar
• 18 Zampatti S, Farro J, Peconi C. et al. AI-powered neurogenetics: supporting patient's evaluation with Chatbot. Genes (Basel) 2024; 16 (01) 29
  PubMedSearch in Google Scholar
• 19 Simuni T, Chahine LM, Poston K. et al. A biological definition of neuronal α-synuclein disease: towards an integrated staging system for research. Lancet Neurol 2024; 23 (02) 178-190
  CrossrefPubMedSearch in Google Scholar
• 20 Dietiker C, Tanner C. Evolving perspectives on α-synuclein testing. JAMA Neurol 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 21 Gibbons C, Bellaire B, Levine T, Freeman R. A novel diagnostic method for detection and quantitation of cutaneous phosphorylated alpha-synuclein (S26.004). Neurology 2024; 102 (7, supplement 1)
  CrossrefPubMedSearch in Google Scholar
• 22 Howey KD, Li M, Christenson PR, Larsen PA, Oh SH. AI-QuIC: machine learning for automated detection of misfolded proteins in seed amplification assays. bioRxiv 2024. Epub ahead of print
  CrossrefPubMed
• 23 Magesh PR, Myloth RD, Tom RJ. An explainable machine learning model for early detection of Parkinson's disease using LIME on DaTSCAN imagery. Comput Biol Med 2020; 126: 104041
  PubMedSearch in Google Scholar
• 24 Mushta I, Koks S, Popov A, Lysenko O. Exploring the potential imaging biomarkers for Parkinson's disease using machine learning approach. Bioengineering (Basel) 2024; 12 (01) 11
  PubMedSearch in Google Scholar
• 25 Khachnaoui H, Khlifa N, Mabrouk R. Machine learning for early Parkinson's disease identification within SWEDD group using clinical and DaTSCAN SPECT imaging features. J Imaging 2022; 8 (04) 97
  PubMedSearch in Google Scholar
• 26 Gupta R, Kumari S, Senapati A, Ambasta RK, Kumar P. New era of artificial intelligence and machine learning-based detection, diagnosis, and therapeutics in Parkinson's disease. Ageing Res Rev 2023; 90: 102013
  PubMedSearch in Google Scholar
• 27 Huang EW, Wang S, Zhai C. VisAGE: integrating external knowledge into electronic medical record visualization. Pac Symp Biocomput 2018; 23: 578-589
  PubMedSearch in Google Scholar
• 28 Ayala Solares JR, Diletta Raimondi FE, Zhu Y. et al. Deep learning for electronic health records: a comparative review of multiple deep neural architectures. J Biomed Inform 2020; 101: 103337
  PubMedSearch in Google Scholar
• 29 Smith CM, Weathers AL, Lewis SL. An overview of clinical machine learning applications in neurology. J Neurol Sci 2023; 455: 122799
  PubMedSearch in Google Scholar
• 30 Cloud LJ, Jinnah HA. Treatment strategies for dystonia. Expert Opin Pharmacother 2010; 11 (01) 5-15
  CrossrefPubMedSearch in Google Scholar
• 31 Oehrn CR, Cernera S, Hammer LH. et al. Chronic adaptive deep brain stimulation versus conventional stimulation in Parkinson's disease: a blinded randomized feasibility trial. Nat Med 2024; 30 (11) 3345-3356
  PubMedSearch in Google Scholar
• 32 Cole ER, Miocinovic S. Are we ready for automated deep brain stimulation programming?. Parkinsonism Relat Disord 2025; 134: 107347
  PubMedSearch in Google Scholar
• 33 Liu Y, Zhang G, Tarolli CG. et al. Monitoring gait at home with radio waves in Parkinson's disease: a marker of severity, progression, and medication response. Sci Transl Med 2022; 14 (663) eadc9669
  PubMedSearch in Google Scholar
• 34 Acevedo G, Lange F, Calonge C, Peach R, Wong JK, Guarin DL. VisionMD: an open-source tool for video-based analysis of motor function in movement disorders. NPJ Parkinsons Dis 2025; 11 (01) 27
  PubMedSearch in Google Scholar
• 35 Islam MS, Rahman W, Abdelkader A. et al. Using AI to measure Parkinson's disease severity at home. NPJ Digit Med 2023; 6 (01) 156
  PubMedSearch in Google Scholar
• 36 Duggan MJ, Gervase J, Schoenbaum A. et al. Clinician experiences with ambient scribe technology to assist with documentation burden and efficiency. JAMA Netw Open 2025; 8 (02) e2460637
  PubMedSearch in Google Scholar
• 37 Small WR, Wiesenfeld B, Brandfield-Harvey B. et al. Large language model-based responses to patients' in-basket messages. JAMA Netw Open 2024; 7 (07) e2422399
  PubMedSearch in Google Scholar
• 38 Li MH, Mestre TA, Fox SH, Taati B. Automated assessment of levodopa-induced dyskinesia: evaluating the responsiveness of video-based features. Parkinsonism Relat Disord 2018; 53: 42-45
  PubMedSearch in Google Scholar
• 39 Fereshtehnejad SM, Lökk J. Challenges of teleneurology in the care of complex neurodegenerative disorders: the case of Parkinson's disease with possible solutions. Healthcare (Basel) 2023; 11 (24) 3187
  PubMedSearch in Google Scholar
• 40 Mahajan A, Lees AJ. “The machine will see you now”: a clinician's perspective on artificial “intelligence” in clinical care. Mov Disord Clin Pract 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 41 Landers M, Saria S, Espay AJ. Will artificial intelligence replace the movement disorders specialist for diagnosing and managing Parkinson's disease?. J Parkinsons Dis 2021; 11 (s1): S117-S122
  PubMedSearch in Google Scholar
• 42 Karpov OE, Pitsik EN, Kurkin SA. et al. Analysis of publication activity and research trends in the field of AI medical applications: network approach. Int J Environ Res Public Health 2023; 20 (07) 5335
  PubMedSearch in Google Scholar
• 43 Dzialas V, Doering E, Eich H. et al; International Parkinson Movement Disorders Society-Neuroimaging Study Group. Houston, we have AI problem! Quality issues with neuroimaging-based artificial intelligence in Parkinson's disease: a systematic review. Mov Disord 2024; 39 (12) 2130-2143
  PubMedSearch in Google Scholar
• 44 Wiest IC, Leßmann ME, Wolf F. et al. Anonymizing medical documents with local, privacy preserving large language models: the LLM-Anonymizer. medRxiv 2024. Epub ahead of print
  CrossrefPubMed
• 45 Mandl KD, How AI. How AI could reshape health care—rise in direct-to-consumer models. JAMA 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar

Address for correspondence

Publication History

Accepted Manuscript online:
29 April 2025

Article published online:
26 May 2025

© 2025. Thieme. All rights reserved.

Thieme Medical Publishers, Inc.
333 Seventh Avenue, 18th Floor, New York, NY 10001, USA

• References

• 1 Dorsey ER, Omberg L, Waddell E. et al. Deep phenotyping of Parkinson's disease. J Parkinsons Dis 2020; 10 (03) 855-873
  CrossrefPubMedSearch in Google Scholar
• 2 Vizcarra JA, Yarlagadda S, Xie K, Ellis CA, Spindler M, Hammer LH. Artificial intelligence in the diagnosis and quantitative phenotyping of hyperkinetic movement disorders: a systematic review. J Clin Med 2024; 13 (23) 7009
  PubMedSearch in Google Scholar
• 3 Dorsey ER, Sherer T, Okun MS, Bloem BR. The emerging evidence of the Parkinson pandemic. J Parkinsons Dis 2018; 8 (s1): S3-S8
  CrossrefPubMedSearch in Google Scholar
• 4 van der Stouwe AMM, Tuitert I, Giotis I. et al. Next move in movement disorders (NEMO): developing a computer-aided classification tool for hyperkinetic movement disorders. BMJ Open 2021; 11 (10) e055068
  PubMedSearch in Google Scholar
• 5 Dalenberg JR, Peretti DE, Marapin LR, van der Stouwe AMM, Renken RJ, Tijssen MAJ. Next move in movement disorders: neuroimaging protocols for hyperkinetic movement disorders. Front Hum Neurosci 2024; 18: 1406786
  PubMedSearch in Google Scholar
• 6 Valeriani D, Simonyan K. A microstructural neural network biomarker for dystonia diagnosis identified by a DystoniaNet deep learning platform. Proc Natl Acad Sci U S A 2020; 117 (42) 26398-26405
  PubMedSearch in Google Scholar
• 7 Yao D, O'Flynn LC, Simonyan K. DystoniaBoTXNet: novel neural network biomarker of botulinum toxin efficacy in isolated dystonia. Ann Neurol 2023; 93 (03) 460-471
  PubMedSearch in Google Scholar
• 8 Simonyan K, Yao D. DystoniaDBSNet: a novel deep learning biomarker of predictive treatment outcomes in dystonia. Brain Stimul 2025; 18 (01) 274
  PubMedSearch in Google Scholar
• 9 Junker J, Hall J, Berman BD. et al; Dystonia Coalition Study Group. Longitudinal predictors of health-related quality of life in isolated dystonia. J Neurol 2024; 271 (02) 852-863
  PubMedSearch in Google Scholar
• 10 Albanese A, Bhatia K, Bressman SB. et al. Phenomenology and classification of dystonia: a consensus update. Mov Disord 2013; 28 (07) 863-873
  CrossrefPubMedSearch in Google Scholar
• 11 Friedrich MU, Relton S, Wong D, Alty J. Computer vision in clinical neurology: a review. JAMA Neurol 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 12 Vaillancourt DE, Barmpoutis A, Wu SS. et al; AIDP Study Group. Automated imaging differentiation for Parkinsonism. JAMA Neurol 2025; e250112 . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 13 Yang Y, Yuan Y, Zhang G. et al. Artificial intelligence-enabled detection and assessment of Parkinson's disease using nocturnal breathing signals. Nat Med 2022; 28 (10) 2207-2215
  PubMedSearch in Google Scholar
• 14 Hällqvist J, Bartl M, Dakna M. et al. Plasma proteomics identify biomarkers predicting Parkinson's disease up to 7 years before symptom onset. Nat Commun 2024; 15 (01) 4759
  PubMedSearch in Google Scholar
• 15 Makarious MB, Leonard HL, Vitale D. et al. Multi-modality machine learning predicting Parkinson's disease. NPJ Parkinsons Dis 2022; 8 (01) 35
  PubMedSearch in Google Scholar
• 16 Deik A. Potential benefits and perils of incorporating ChatGPT to the movement disorders clinic. J Mov Disord 2023; 16 (02) 158-162
  CrossrefPubMedSearch in Google Scholar
• 17 Mohajer B. Faster, more practical, but still accurate: deep learning for diagnosis of progressive supranuclear palsy. Radiol Artif Intell 2024; 6 (03) e240181
  PubMedSearch in Google Scholar
• 18 Zampatti S, Farro J, Peconi C. et al. AI-powered neurogenetics: supporting patient's evaluation with Chatbot. Genes (Basel) 2024; 16 (01) 29
  PubMedSearch in Google Scholar
• 19 Simuni T, Chahine LM, Poston K. et al. A biological definition of neuronal α-synuclein disease: towards an integrated staging system for research. Lancet Neurol 2024; 23 (02) 178-190
  CrossrefPubMedSearch in Google Scholar
• 20 Dietiker C, Tanner C. Evolving perspectives on α-synuclein testing. JAMA Neurol 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 21 Gibbons C, Bellaire B, Levine T, Freeman R. A novel diagnostic method for detection and quantitation of cutaneous phosphorylated alpha-synuclein (S26.004). Neurology 2024; 102 (7, supplement 1)
  CrossrefPubMedSearch in Google Scholar
• 22 Howey KD, Li M, Christenson PR, Larsen PA, Oh SH. AI-QuIC: machine learning for automated detection of misfolded proteins in seed amplification assays. bioRxiv 2024. Epub ahead of print
  CrossrefPubMed
• 23 Magesh PR, Myloth RD, Tom RJ. An explainable machine learning model for early detection of Parkinson's disease using LIME on DaTSCAN imagery. Comput Biol Med 2020; 126: 104041
  PubMedSearch in Google Scholar
• 24 Mushta I, Koks S, Popov A, Lysenko O. Exploring the potential imaging biomarkers for Parkinson's disease using machine learning approach. Bioengineering (Basel) 2024; 12 (01) 11
  PubMedSearch in Google Scholar
• 25 Khachnaoui H, Khlifa N, Mabrouk R. Machine learning for early Parkinson's disease identification within SWEDD group using clinical and DaTSCAN SPECT imaging features. J Imaging 2022; 8 (04) 97
  PubMedSearch in Google Scholar
• 26 Gupta R, Kumari S, Senapati A, Ambasta RK, Kumar P. New era of artificial intelligence and machine learning-based detection, diagnosis, and therapeutics in Parkinson's disease. Ageing Res Rev 2023; 90: 102013
  PubMedSearch in Google Scholar
• 27 Huang EW, Wang S, Zhai C. VisAGE: integrating external knowledge into electronic medical record visualization. Pac Symp Biocomput 2018; 23: 578-589
  PubMedSearch in Google Scholar
• 28 Ayala Solares JR, Diletta Raimondi FE, Zhu Y. et al. Deep learning for electronic health records: a comparative review of multiple deep neural architectures. J Biomed Inform 2020; 101: 103337
  PubMedSearch in Google Scholar
• 29 Smith CM, Weathers AL, Lewis SL. An overview of clinical machine learning applications in neurology. J Neurol Sci 2023; 455: 122799
  PubMedSearch in Google Scholar
• 30 Cloud LJ, Jinnah HA. Treatment strategies for dystonia. Expert Opin Pharmacother 2010; 11 (01) 5-15
  CrossrefPubMedSearch in Google Scholar
• 31 Oehrn CR, Cernera S, Hammer LH. et al. Chronic adaptive deep brain stimulation versus conventional stimulation in Parkinson's disease: a blinded randomized feasibility trial. Nat Med 2024; 30 (11) 3345-3356
  PubMedSearch in Google Scholar
• 32 Cole ER, Miocinovic S. Are we ready for automated deep brain stimulation programming?. Parkinsonism Relat Disord 2025; 134: 107347
  PubMedSearch in Google Scholar
• 33 Liu Y, Zhang G, Tarolli CG. et al. Monitoring gait at home with radio waves in Parkinson's disease: a marker of severity, progression, and medication response. Sci Transl Med 2022; 14 (663) eadc9669
  PubMedSearch in Google Scholar
• 34 Acevedo G, Lange F, Calonge C, Peach R, Wong JK, Guarin DL. VisionMD: an open-source tool for video-based analysis of motor function in movement disorders. NPJ Parkinsons Dis 2025; 11 (01) 27
  PubMedSearch in Google Scholar
• 35 Islam MS, Rahman W, Abdelkader A. et al. Using AI to measure Parkinson's disease severity at home. NPJ Digit Med 2023; 6 (01) 156
  PubMedSearch in Google Scholar
• 36 Duggan MJ, Gervase J, Schoenbaum A. et al. Clinician experiences with ambient scribe technology to assist with documentation burden and efficiency. JAMA Netw Open 2025; 8 (02) e2460637
  PubMedSearch in Google Scholar
• 37 Small WR, Wiesenfeld B, Brandfield-Harvey B. et al. Large language model-based responses to patients' in-basket messages. JAMA Netw Open 2024; 7 (07) e2422399
  PubMedSearch in Google Scholar
• 38 Li MH, Mestre TA, Fox SH, Taati B. Automated assessment of levodopa-induced dyskinesia: evaluating the responsiveness of video-based features. Parkinsonism Relat Disord 2018; 53: 42-45
  PubMedSearch in Google Scholar
• 39 Fereshtehnejad SM, Lökk J. Challenges of teleneurology in the care of complex neurodegenerative disorders: the case of Parkinson's disease with possible solutions. Healthcare (Basel) 2023; 11 (24) 3187
  PubMedSearch in Google Scholar
• 40 Mahajan A, Lees AJ. “The machine will see you now”: a clinician's perspective on artificial “intelligence” in clinical care. Mov Disord Clin Pract 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar
• 41 Landers M, Saria S, Espay AJ. Will artificial intelligence replace the movement disorders specialist for diagnosing and managing Parkinson's disease?. J Parkinsons Dis 2021; 11 (s1): S117-S122
  PubMedSearch in Google Scholar
• 42 Karpov OE, Pitsik EN, Kurkin SA. et al. Analysis of publication activity and research trends in the field of AI medical applications: network approach. Int J Environ Res Public Health 2023; 20 (07) 5335
  PubMedSearch in Google Scholar
• 43 Dzialas V, Doering E, Eich H. et al; International Parkinson Movement Disorders Society-Neuroimaging Study Group. Houston, we have AI problem! Quality issues with neuroimaging-based artificial intelligence in Parkinson's disease: a systematic review. Mov Disord 2024; 39 (12) 2130-2143
  PubMedSearch in Google Scholar
• 44 Wiest IC, Leßmann ME, Wolf F. et al. Anonymizing medical documents with local, privacy preserving large language models: the LLM-Anonymizer. medRxiv 2024. Epub ahead of print
  CrossrefPubMed
• 45 Mandl KD, How AI. How AI could reshape health care—rise in direct-to-consumer models. JAMA 2025; . Epub ahead of print
  CrossrefPubMedSearch in Google Scholar