Authors: Jamie Linnea Luckhaus, Therese Scott Duncan, Uppsala University, Sweden

PATH: Participatory eHealth and Health Data Department of Women's and Children's Health Uppsala University

Reviewers: 

Stelios Hadjidimitriou, Signal Processing & Biomedical Technology Unit Department of Electrical & Computer Engineering Aristotle University of Thessaloniki, Greece

Kyproula Christodoulou, The Cyprus Institute of Neurology & Genetics, Cyprus

Paraskevi Chairta, The Cyprus Institute of Neurology & Genetics, Cyprus

Björn Falkenburger, TUD Dresden University of Technology, Dresden, Germany

Diagnosing Parkinson’s disease (PD) can be challenging, especially in the early stages. The early symptoms can be subtle or overlap with other disorders, and PD is defined clinically, so there is no lab test for diagnosing PD. As a result, delayed diagnosis is not uncommon. Artificial intelligence (AI) tools are being explored as potential means to strengthen Parkinson’s diagnosis by helping healthcare professionals to detect the disease earlier. To this end, in recent years, researchers have begun leveraging AI algorithms, such as machine learning models, to analyze a wide range of data, from wearable sensor data and voice recordings to genetic variations and brain scans, for telltale patterns of Parkinson’s [1].

How could AI be used for detecting Parkinson’s? 

Screening in daily life 

Patterns in wearable sensor data: PD leaves unique signatures in how a person moves and sleeps, with subtle changes in mobility, walking, and sleep quality. With wearable devices, such as smartwatches, enabling the passive collection of high volumes of human activity data in daily life, AI algorithms have been proven capable of detecting such unique signatures associated with Parkinson’s. A landmark study analysing data from the UK Biobank suggested that patterns in people’s daily mobility data recorded by a single wrist sensor can detect persons at the prodromal stage of the disease, up to seven years prior to clinical diagnosis [2]. Recently, AI-PROGNOSIS researchers used sleep, walking, and other activity metrics computed from wrist wearable data of the Parkinson’s Progression Markers Initiative study to train an AI model in distinguishing between persons with and without PD [3]. The model achieved high accuracy (over 80%) and seemed also capable of tracking measures of depletion of dopaminergic brain neurons, a hallmark of the disease. 

Patterns in smartphone interaction data and voice analysis: Smartphones and apps have been used to monitor typing, hand steadiness, and even voice, with machine learning models distinguishing persons with PD from healthy individuals by relevant metrics. For example, metrics of key tapping during touchscreen typing have been associated with fine motor skills impairment in early Parkinson’s patients [4], while data from the smartphone’s motion sensors have been used to detect the presence of tremor [5]. As Parkinson’s often affects speech, causing changes in volume, tone, or fluency even very early on, scientists have also trained AI models to analyze voice recordings for these subtle changes, including natural running speech excerpts from voice calls [6], as wells as recordings of prolonged enunciations of vowels [7]. AI-PROGNOSIS researchers are also attempting to use smartphone-captured video of persons doing movement tasks and AI-based body tracking to detect and quantify signs of agility, walking and balance impairment [8].  

Wearable and smartphone-based tracking methods are often unobtrusive, inexpensive, and have already shown encouraging results. These technologies could strengthen diagnosis by providing objective data points (not based on personal interpretation) and allowing for risk monitoring in daily life. Instead of relying only on a brief clinic exam, a doctor might incorporate a summary from a month’s worth of sensor data (or even longer), analyzed by AI, which could reveal activity patterns consistent with PD. 

Other key ways AI is enhancing Parkinson’s diagnosis 

Advanced brain imaging analysis: AI can uncover patterns in brain scans that might be too subtle for the human eye. For example, a 2025 study used a machine learning algorithm on specialized MRI images to differentiate Parkinson’s disease from two other parkinsonian syndromes, multiple system atrophy and progressive supranuclear palsy. The results were promising, as the AI approach could distinguish PD from atypical parkinsonism with about 86% sensitivity [9]. This kind of tool could help neurologists rule out other diseases than Parkinson’s diagnosis earlier in the disease course. Researchers envision that in the future, a patient might get an MRI and an AI system could instantly provide a probability of PD versus other conditions, assisting (not replacing) doctors in making the call. Such tools might be especially valuable in areas where movement disorder specialists are not available, as they can be made accessible online so that a community hospital can upload a brain scan and get expert-level diagnostic input. 

AI in medical records and genetic analysis: Another angle is using AI to crunch through medical records or lab data to flag at-risk persons. One study developed a machine learning model that scanned the past medical histories of thousands of people and identified those who would later develop Parkinson’s with good accuracy. The algorithm correctly predicted about 73% of future Parkinson’s cases (and correctly recognized 83% of people who would not get PD) by looking at patterns of diagnoses and health complaints in prior years [10]. Many of the predictive clues were symptoms already known to be linked to PD. For example, the model found that patients who had early reports of tremor, loss of smell, constipation, or certain sleep problems were more likely to be diagnosed with Parkinson’s down the line. Given access to electronic health records, tools like this could alert primary care doctors that a patient with a constellation of symptoms might warrant a neurologist referral.  

Beyond records, AI-based tools are used for the analysis and interpretation of genetic data [11]. Genetic research on Parkinson’s aim to identify genetic factors that are associated with an increased risk to develop the disease [12]. Machine learning is being applied in the analysis of genetic data, in some cases in combination with other data such as age or lifestyle factors. Such methods can be used to stratify risk levels, which can also aid in participant selection in future clinical trials of disease-modifying therapies [13-14]. 

Implications and future of AI in diagnosis   

The integration of AI into Parkinson’s diagnostics could significantly improve patient care. Earlier and more accurate diagnosis means patients can start medications and lifestyle interventions sooner, potentially slowing symptom progression and avoiding unnecessary tests or treatments for the wrong condition. It also means that when promising neuroprotective drugs do become available in the future, we can identify the right patients at the right time to receive them. Importantly, AI is seen as a support for clinicians, not a replacement. An AI might, for example, provide a second opinion on a scan or highlight abnormal activity patterns, but a neurologist will still make the final diagnosis in context of the whole patient. As these technologies develop, there’s a focus on making them explainable and user-friendly for doctors, so that an AI doesn’t just give a prediction, but also shows the reasoning. 

An aspect of predicting disease risk is the human factor: do people actually want to know what the AI predicts? The answer is not one-size-fits-all. In a recent survey of healthy people (average age 65, no diagnosis), 79% said they unconditionally wanted to be informed of their risk, mostly to prepare for the future or take preventative action [15]. In the same study, when asked about a hypothetical “highly reliable” predictive test, over 70% of participants said they would want detailed counselling and information on lifestyle changes if they tested positive, suggesting that, given actionable guidance, many welcome prognostic knowledge.  People with Parkinson’s also reported interest in personalized prognostic information—“the more personalized the better”—but stress that professional interpretation and coaching are vital. At the same time, certain persons with Parkinson’s note that an earlier diagnosis isn’t always valuable if no preventive options exist, suggesting predictions should be shared selectively [16, 17]. A study using the AI-PROGNOSIS project as context found that people with PD saw potential benefits in personalized recommendations and improved understanding of their disease, but also worried over possible psychological harm if predictions are given without adequate support [18].  

The healthcare system should create guidelines for how predictive information is shared with patients, taking into account personal preferences and emotional impact. Clinicians deploying AI tools will need to counsel patients about what a “prediction” really means – it’s a probability, not fate written in stone – and ensure psychological support is in place, since hearing a grim prediction can be stressful. Another challenge lies in communicating uncertainty. Even a very accurate AI will have a margin of error.  With no infallible biomarker to anchor these predictions, doctors and patients will need to interpret AI outputs cautiously.  

Most AI diagnostic tools for PD are currently in the research or experimental stage. Progress is rapid but there are challenges to overcome: ensuring these AI models are trained on diverse populations (so they generalize to all patients), protecting privacy in health data, and avoiding false positives that could worry patients unnecessarily. But the trend is clear, AI has immense potential to strengthen Parkinson’s diagnosis. The role of clinicians and technologists will be to support each individual’s choice, ensure predictions are as accurate as possible, and above all, keep the person with Parkinson’s at the center of any technological innovation. 

References 

[1] Shokrpour, S. et al. Machine learning for Parkinson’s disease: a comprehensive review of datasets, algorithms, and challenges. npj Parkinsons Dis. 11, 187 (2025). https://www.nature.com/articles/s41531-025-01025-9 

[2] Schalkamp, A. K., et al. Wearable movement-tracking data identify Parkinson’s disease years before clinical diagnosis. Nature Medicine 29, 2048-2056 (2023). https://doi.org/10.1038/s41591-023-02440-2  

[3] Sotirakis, H. et al. D3.2 / First report on predictive modelling for PD. Zenodo (2025). https://doi.org/10.5281/zenodo.15542465  

[4] Iakovakis, D., et al. Touchscreen typing-pattern analysis for detecting fine motor skills decline in early-stage Parkinson’s disease. Scientific reports 8, 1-13 (2018). https://doi.org/10.1038/s41598-018-25999-0  

[5] Papadopoulos, A., et al. Unobtrusive detection of Parkinson’s disease from multi-modal and in-the-wild sensor data using deep learning techniques. Scientific reports 10, 21370 (2020). https://doi.org/10.1038/s41598-020-78418-8  

[6] Laganas, C., et al. Parkinson’s disease detection based on running speech data from phone calls. IEEE Transactions on Biomedical Engineering 69, 1573-1584 (2021). https://doi.org/10.1109/tbme.2021.3116935 

[7] Shen, M., Mortezaagha, P. & Rahgozar, A. Explainable artificial intelligence to diagnose early Parkinson’s disease via voice analysis. Sci Rep 15, 11687 (2025). https://www.nature.com/articles/s41598-025-96575-6 

[8] Chatzichristos, C., et al. D3.1 / First report on digital biomarkers for PD. Zenodo 2025. https://doi.org/10.5281/zenodo.15542385  

[9] Vaillancourt, D. E. et al. Automated Imaging Differentiation for Parkinsonism. JAMA Neurol 82, 495 (2025). https://jamanetwork.com/journals/jamaneurology/fullarticle/2831631  

[10] Searles Nielsen, S. et al. A predictive model to identify Parkinson disease from administrative claims data. Neurology 89, 1448–1456 (2017). https://www.neurology.org/doi/10.1212/WNL.0000000000004536 

[11] Vilhekar, R. S., & Rawekar, A. (2024). Artificial Intelligence in Genetics. Cureus, 16(1), e52035. https://doi.org/10.7759/cureus.52035  

[12] Funayama, M., Nishioka, K., Li, Y., & Hattori, N. (2023). Molecular genetics of Parkinson's disease: Contributions and global trends. Journal of human genetics, 68(3), 125–130. https://doi.org/10.1038/s10038-022-01058-5  

[13] Sigala, R. E., Lagou, V., Shmeliov, A., Atito, S., Kouchaki, S., Awais, M., Prokopenko, I., Mahdi, A., & Demirkan, A. (2023). Machine Learning to Advance Human Genome-Wide Association Studies. Genes, 15(1), 34. https://doi.org/10.3390/genes15010034  

[14] Pihlstrøm, L., et al. Genetic stratification of age‐dependent parkinson’s disease risk by polygenic hazard score. Movement Disorders, 37(1), 62–69. (2021). https://doi.org/10.1002/mds.28808  

[15] Mahlknecht, P. et al. Preferences regarding Disclosure of Risk for Parkinson’s Disease in a Population‐based Study. Movement Disord Clin Pract 12, 203–209 (2025). https://movementdisorders.onlinelibrary.wiley.com/doi/10.1002/mdc3.14264 [16] Van Den Heuvel L, et al. Perspectives of people living with Parkinson’s disease on personalized prediction models. Health Expect. 2022 Aug;25(4):1580–90. 

[17] Schaeffer, E. et al. Patients’ views on the ethical challenges of early Parkinson disease detection. Neurology 94, e2037–e2044 (2020). 

[18] Luckhaus, J. L., et al.  Balancing hope and harm: A qualitative exploration of ethical aspects of using AI in Parkinson’s disease (Preprint). JMIR Preprints. (2025). https://doi.org/10.2196/preprints.74144