# How Artificial Intelligence Is Reshaping Brain Disease Research: A 25-Year Scientific Map A 25-year review of 1,402 scientific publications examined research using artificial intelligence (AI — computer systems that learn from data) in neurodegenerative diseases such as Alzheimer's and Parkinson's. This research has grown explosively since 2017. This growth was driven by deep learning and the combining of multiple types of patient data. The United States (25.96%, about 1 in 4 publications) and China (24.11%, also about 1 in 4) produced the most research. The United Kingdom had the greatest international reach (centrality 0.24) and the highest average citations per publication (31.68). The most studied topics were Alzheimer's disease, Parkinson's disease, MRI brain scans, convolutional neural networks (CNN — an AI method for analyzing images), and biomarkers (biological signals of disease). Mild cognitive impairment (MCI — memory problems that are worse than normal aging but not yet dementia) was also among the most studied topics. The authors conclude that AI is reshaping how these diseases are diagnosed and how new treatments are discovered. The authors conclude that closer teamwork between computer scientists and doctors is needed before these tools reach routine patient care. # Artificial intelligence in neurodegenerative diseases research: a bibliometric analysis since 2000. ## Table of Contents - Key Points - Background: Why This Research Matters - What Is a Bibliometric Analysis? - How the Study Was Conducted - Publication Trends: A Slow Start, Then an Explosion - Which Countries Lead the Field - Leading Institutions - Where the Research Is Published - How Different Scientific Fields Connect - The Most Active Researchers and Research Groups - The Most Influential Papers - Citation Bursts: When Certain Papers Suddenly Mattered - What Topics Researchers Study Most - Fastest-Rising Topics - What the Authors Concluded - Why Some Countries Publish More Than Others - What This Means for Patients - Limitations of This Analysis - Priorities and Recommendations Going Forward - Frequently Asked Questions - Source Information ## Key Points - A 25-year review of 1,402 publications found AI research in neurodegenerative diseases grew explosively after 2017, driven by deep learning and combining multiple types of patient data. - The most studied topics were Alzheimer's disease, Parkinson's disease, MRI brain scans, convolutional neural networks, biomarkers, and mild cognitive impairment. - Keyword bursts shifted from support vector machines (2019–2022) toward neural networks and deep neural networks (2022–2023), reflecting a move to deep learning. - The authors concluded AI is reshaping diagnosis and treatment discovery but closer teamwork between computer scientists and doctors is needed before routine patient care. ## Background: Why This Research Matters Neurodegenerative diseases (conditions in which nerve cells progressively stop working and die) include Alzheimer's disease and Parkinson's disease. They cause worsening problems with thinking, memory, and movement. As populations age worldwide, these conditions are becoming more common and represent a growing global health challenge. Existing treatments cannot stop or reverse these diseases. That gap has created urgent demand for new ways to detect them earlier and to design better therapies. Researchers have turned to artificial intelligence for help. AI has already shown promise in several areas of medicine: - **Deep learning models** for segmenting brain scans (identifying and outlining structures on imaging) - **Predictive algorithms** that estimate how drugs will interact with their biological targets - **Multimodal frameworks** that combine genetic information with clinical data (multimodal means using several types of data at once) The problem is that AI research in brain diseases requires teamwork across very different specialties. Neurobiologists (scientists who study the nervous system), computer modelers, and clinicians often work in separate silos. That separation creates fragmented methods and slows progress. What was missing, according to the authors, was a big-picture view. No one had systematically mapped who collaborates with whom, how the knowledge base has shifted, and which technologies are emerging in this field. This study is the first bibliometric analysis of AI applications in neurodegenerative disease research since the year 2000. ## What Is a Bibliometric Analysis? A bibliometric analysis is a statistical study of published research itself. Instead of examining patients, it examines papers — counting them, tracking who wrote them, and mapping which papers are cited together. These analyses can reveal: - Collaboration networks among countries, institutions, and individual authors - Which keywords appear together most often (co-occurrence), showing how topics cluster - Co-citation patterns — when two papers are cited together by a third paper, suggesting they are conceptually linked - Which journals carry the greatest influence in a field By combining authorship, citation, and keyword data with visual maps, bibliometrics describes both the structure and the movement of a research field. This helps identify new research frontiers, guide how funding is allocated, and encourage collaboration across disciplines. It cannot tell you whether a treatment works — but it can tell you where the scientific effort is going. ## How the Study Was Conducted The authors searched the Web of Science Core Collection (WoSCC), a large curated database of scholarly publications, on March 16, 2025. Their search combined two sets of terms: names for neurodegenerative diseases and names for AI techniques. The disease terms included variations such as "Neurodegenerative Diseases," "Degenerative Neurologic Disorders," "Nervous System Degenerative Diseases," and "Degenerative Diseases, Spinal Cord." The AI terms included "artificial intelligence," "AI," "deep learning," "machine learning," "Computer Reasoning," "Machine Intelligence," and "Computational Intelligence." They then applied filters. Only English-language publications from the year 2000 onward were kept. Only two document types were allowed: "Article" or "Review." After retrieving the records, preliminary data processing was done in CiteSpace (a software tool for visualizing scientific trends). The final dataset contained **1,402 publications** — 1,159 original research articles and 243 review articles. Three software tools did the analysis: 1. **VOSviewer v1.6.20** — builds and visualizes network maps of countries, institutions, journals, authors, and keywords. 1. **CiteSpace v6.4.R1** — specializes in tracking change over time, including "burst detection" (finding sudden surges in citations or keyword use) and dual-map overlays of journal citation paths. 1. **Bibliometrix R package v4.3.2** — handles statistical processing, citation analysis, and journal influence measures such as the H-index and G-index (both combine publication counts with citation counts to express influence). The researchers made deliberate choices to keep the data consistent. When analyzing countries, they merged administrative divisions of the same nation — England, Scotland, Northern Ireland, and Wales were all counted as the United Kingdom. For keywords, they standardized synonyms; for example, "Alzheimer disease" and "AD" were unified as "Alzheimer's disease." Two researchers independently verified the assessments. Any disagreements were re-evaluated by a third researcher, and the final decisions were reached by consensus among all three. VOSviewer was also used to map country collaborations, institutional partnerships, journal publication patterns, journal co-citation, author clusters, and high-frequency keywords. ## Publication Trends: A Slow Start, Then an Explosion The field grew in distinct phases. Before 2014, annual output stayed consistently below 10 articles per year, with small fluctuations but overall stagnation. Sustained growth began in 2014. Then, after 2017, growth became exponential — meaning each year's increase was proportionally larger than the year before. By 2024, publications reached **379 articles in a single year**. Most striking: studies published since 2023 account for more than half of the field's entire output. In total, the 1,402 publications came from 86 countries or regions and involved 8,048 scholars at 2,637 institutions. They appeared in 509 academic journals and contained 3,315 author keywords. Collectively, these works cited 71,363 references drawn from 12,374 distinct journal sources. ## Which Countries Lead the Field The United States and China dominate publication volume. The USA produced 364 articles (25.96%, about 1 in 4 publications), and China produced 338 (24.11%, also about 1 in 4). The USA also leads in total citations, with 10,223. But the United Kingdom stands out in two other ways: it has the highest average citations per publication (31.68) and the highest centrality (0.24). Centrality is a measure of how well connected a country is within the international collaboration network. India ranks fourth in publication volume (138 articles, 9.84%) but has the lowest average citations per publication at 13.62 — roughly two and a half times lower than the UK's figure of 31.68. The top 10 most productive countries and regions were: - **USA** — 364 publications (25.96%), 10,223 citations, 28.09 average citations per publication, centrality 0.17 - **China** — 338 (24.11%), 6,713 citations, 19.86 average, centrality 0.09 - **UK** — 152 (10.84%), 4,816 citations, 31.68 average, centrality 0.24 - **India** — 138 (9.84%), 1,880 citations, 13.62 average, centrality 0.11 - **Italy** — 118 (8.42%), 2,664 citations, 22.58 average, centrality 0.10 - **Germany** — 106 (7.56%), 2,557 citations, 24.12 average, centrality 0.08 - **Spain** — 82 (5.85%), 1,467 citations, 17.89 average, centrality 0.06 - **Canada** — 69 (4.92%), 1,939 citations, 28.10 average, centrality 0.15 - **Australia** — 67 (4.78%), 1,776 citations, 26.51 average, centrality 0.17 - **South Korea** — 66 (4.71%), 1,675 citations, 25.38 average, centrality 0.03 The collaboration map of the top 30 countries shows the USA, China, and the UK forming central nodes — the hubs through which much international collaboration flows. ## Leading Institutions The top 10 productive institutions were dominated by the USA, China, and the UK. University College London (UCL) published the most, while the Chinese Academy of Sciences accumulated the most citations overall (1,358). Two contrasting patterns emerged. Institutions from China, such as Sichuan University, had the most recent average publication year (2023.2), suggesting rapidly growing recent activity. The University of California San Francisco had the highest average citations per publication (42.4), suggesting its work carried unusual influence. The top 10 were: 1. **UCL** — 27 publications, 912 citations, 33.78 average, average publication year 2021.56 (USA) 1. **Chinese Academy of Sciences** — 24 publications, 1,358 citations, 56.58 average, 2020.71 (China) 1. **King's College London** — 21 publications, 780 citations, 37.14 average, 2021.52 (UK) 1. **Shanghai Jiao Tong University** — 20 publications, 388 citations, 19.40 average, 2021.45 (China) 1. **University of California San Francisco** — 20 publications, 848 citations, 42.40 average, 2022.10 (USA) 1. **Johns Hopkins University** — 16 publications, 344 citations, 21.50 average, 2022.88 (USA) 1. **University of Oxford** — 16 publications, 391 citations, 24.44 average, 2021.00 (UK) 1. **Mayo Clinic** — 15 publications, 273 citations, 18.20 average, 2022.80 (USA) 1. **Sichuan University** — 15 publications, 109 citations, 7.27 average, 2023.20 (China) 1. **University of Pennsylvania** — 15 publications, 185 citations, 12.33 average, 2022.33 (USA) The institutional map reveals distinct clusters, with UCL and the University of California San Francisco as central hubs. A second map shows a time gradient: institutions with earlier average publication years (before 2021) appear in cooler blue tones. Institutions with more recent contributions (after 2022) appear in warmer red hues — a "blue-gray-red" spectrum showing which institutions are newly active. ## Where the Research Is Published A total of 509 journals contributed to this field. Applying Bradford's Law of Scattering — a principle stating that a small core of journals publishes a large share of a field's literature — the authors identified **21 core journals**. Every one of these is ranked Q1 or Q2 in the 2024 Journal Citation Reports (JCR quartiles, where Q1 represents the top 25% of journals in a category by impact factor). That all core journals fall in the top quartiles indicates high scientific rigor in this field. The leading core journals were: - **Scientific Reports** — 47 publications, H-index 14, G-index 27, 813 citations, impact factor 3.8 (Q1) - **Frontiers in Aging Neuroscience** — 46 publications, H-index 15, G-index 25, 717 citations, IF 4.1 (Q2) - **IEEE Access** — 37 publications, H-index 12, G-index 26, 696 citations, IF 3.4 (Q2) - **Frontiers in Neuroscience** — 30 publications, H-index 12, G-index 26, 726 citations, IF 3.2 (Q2) - **International Journal of Molecular Sciences** — 29 publications, H-index 9, G-index 18, 352 citations, IF 4.9 (Q1) - **Sensors** — 26 publications, H-index 11, G-index 20, 429 citations, IF 3.4 (Q2) - **Computers in Biology and Medicine** — 23 publications, H-index 11, G-index 23, 538 citations, IF 7.0 (Q1) - **Journal of Alzheimer's Disease** — 23 publications, H-index 6, G-index 13, 185 citations, IF 3.4 (Q2) - **Diagnostics** — 22 publications, H-index 9, G-index 15, 234 citations, IF 3.0 (Q1) - **Applied Sciences-Basel** — 21 publications, H-index 10, G-index 18, 350 citations, IF 2.5 (Q1) - **Biomedical Signal Processing and Control** — 21 publications, H-index 5, G-index 10, 121 citations, IF 4.9 (Q1) - **Frontiers in Neurology** — 20 publications, H-index 7, G-index 14, 220 citations, IF 2.7 (Q2) - **NeuroImage** — 19 publications, H-index 11, G-index 19, 1,335 citations, IF 4.7 (Q1) The remaining core journals were IEEE Journal of Biomedical and Health Informatics (18 publications, 350 citations, IF 6.7). Another core journal was Computer Methods and Programs in Biomedicine (16 publications, 535 citations, IF 4.9). PLoS One was also a core journal (16 publications, 288 citations, IF 2.9). NeuroImage-Clinical was the last core journal (12 publications, 235 citations, IF 3.4). The remaining core journals also included Artificial Intelligence in Medicine (11 publications, 201 citations, IF 6.1), Bioengineering-Basel (11 publications, 73 citations, IF 3.8), Heliyon (11 publications, 104 citations, IF 3.4), and Biomedicines (10 publications, 90 citations, IF 3.9). One journal stood out sharply. **NeuroImage** achieved an exceptionally high citation rate (1,335 citations) despite publishing only 19 articles — far more citations than Scientific Reports, which published more than twice as many papers. This signals the critical role of neuroimaging data in AI-driven neurodegenerative research. In the co-citation analysis (which journals are cited together most often), 12,374 journals were identified. The three most co-cited were NeuroImage (2,159 co-citations, IF 4.7, Q1), PLoS One (1,299 co-citations, IF 2.9, Q1), and Neurology (1,228 co-citations, IF 8.4, Q1). High-impact general science journals also featured strongly: Nature (940 co-citations, IF 50.5) and Science (574 co-citations, IF 50.5). ## How Different Scientific Fields Connect The authors used a dual-map overlay, a method developed by Chen and Leydesdorff that shows how citing journals and cited journals are distributed across scientific disciplines. Four thick lines of citation flow appeared, showing the field's multidisciplinary nature: 1. A red line from **Mathematics, Systems, Mathematical** to **Molecular Biology, Genetics** (z = 2.7162597, f = 1,826) 1. A yellow line from **Molecular Biology, Immunology** to **Molecular Biology, Genetics** (z = 8.2197485, f = 5,072) 1. A yellow line from **Molecular Biology, Immunology** to **Psychology, Education, Social** (z = 1.8278345, f = 1,302) 1. A gray line from **Neurology, Sports, Ophthalmology** to **Molecular Biology, Genetics** (z = 3.514825, f = 2,297) In plain terms, computational and mathematical methods increasingly feed into genetics and molecular biology, while clinical neurology and immunology research also feed into molecular genetics. The field is not one discipline — it is several disciplines converging. ## The Most Active Researchers and Research Groups A total of 8,048 authors contributed to this field. The most prolific were Ayala, Matias-Guiu, Kovalenko, Somov, and Dickson, each with 6 articles. Dr. Shen achieved the highest total citations. The author map shows four prominent academic groups, each in a different color cluster: - **Green cluster** — Ayala, Matias-Guiu and colleagues - **Blue cluster** — Kovalenko, Somov and colleagues - **Red cluster** — Dickson and colleagues - **Yellow cluster** — Shen and colleagues These clusters show strong collaboration within each group but limited connection between groups. The authors suggest this may reflect thematic or institutional specialization — different teams are working on different problems, with less cross-talk than the field's interdisciplinary nature would ideally require. ## The Most Influential Papers Highly cited publications often mark turning points in a research field. The top three were: 1. **"The genetic architecture of Parkinson's disease"** — published in *The Lancet Neurology* in 2020, with 649 citations. Led by Blauwendraat and colleagues, this paper establishes a comprehensive genetic framework for Parkinson's disease. It identified more than 90 risk loci (specific locations in the genome linked to disease risk). The authors of the bibliometric review note that this framework enables AI-driven risk prediction, patient stratification (grouping patients by likely course or response), and precision therapeutics. The authors highlight the need for machine learning to integrate genetic data with clinical features. 1. **"Single subject prediction of brain disorders in neuroimaging: Promises and pitfalls"** — published in *NeuroImage* in 2017, with 600 citations. Led by Arbabshirani and colleagues, this is a comprehensive review of how to use AI techniques to analyze multimodal neuroimaging data for automated diagnosis, classification, and prediction of neurodegenerative diseases. The review authors say it provides important guidance for advancing AI in precision diagnosis and treatment. 1. **"Clonally expanded CD8 T cells patrol the cerebrospinal fluid in Alzheimer's disease"** — published in *Nature* in 2020, with 555 citations. This work identified antigen-specific immune signatures (immune-system patterns specific to disease targets) that can boost AI-driven biomarker discovery and treatment-target identification. It demonstrates how combining multiple data types can decode the neuroinflammation (immune-driven inflammation in the nervous system) involved in these diseases. The data types are mass cytometry, single-cell RNA sequencing, T-cell receptor sequencing, and machine learning. ## Citation Bursts: When Certain Papers Suddenly Mattered Burst analysis identifies research that received concentrated attention during a particular period. The authors listed the top 25 co-cited references with the strongest citation bursts. Three stand out: - **He et al.** — the strongest burst (strength 8.43), during 2018–2021 - **Krizhevsky et al.** — strength 8.06, during 2020–2022 - **Liu et al.** — strength 6.91, during 2018–2020 These bursts show how specific technical and biological papers became focal points at specific moments in the field's development. ## What Topics Researchers Study Most Author keywords are chosen to highlight core themes. After excluding the search terms "Artificial Intelligence" and "Neurodegenerative Diseases," the authors identified **287 keywords** appearing at least 3 times each, and built a co-occurrence network from them. The top 10 highest-frequency keywords were: 1. Alzheimer's disease 1. Parkinson's disease 1. Magnetic resonance imaging (MRI — detailed images of the brain using magnetic fields) 1. Convolutional neural network (CNN — a deep learning method for image analysis) 1. Biomarkers (measurable biological indicators of disease) 1. Dementia 1. Classification (assigning cases to categories, such as disease versus healthy) 1. Mild cognitive impairment (MCI) 1. Neuroimaging 1. Feature extraction (pulling out measurable characteristics from complex data) The authors identify the field's key hotspots as intelligent neuroimaging analysis, methodological improvements in machine learning, molecular mechanisms and drug discovery, and clinical decision support systems for early diagnosis. ## Fastest-Rising Topics Keyword emergence analysis finds words that suddenly surged in use, revealing real-time shifts in scientific focus. The top five were: - **"neural networks"** — burst strength 4.66, during 2022–2023 - **"support vector machine" (SVM — an older machine learning method for classification)** — strength 2.61, during 2019–2022 - **"deep neural network"** — strength 2.42, during 2022–2023 - **"structural MRI"** — strength 2.3, during 2019–2021 - **"medical imaging"** — strength 2.04, during 2020–2021 The timing matters. Support vector machine peaked earlier (2019–2022), while neural networks and deep neural networks surged most recently (2022–2023). That shift reflects the field moving from older statistical classification methods toward deep learning. ## What the Authors Concluded The bibliometric analysis systematically maps the research landscape of AI in neurodegenerative diseases. On publication trends, the exponential growth since 2017 underscores the field's rapid expansion, driven by advances in deep learning and the integration of multiple types of data. The journals analysis confirms high scientific standards: all 21 core journals rank Q1 or Q2 in the 2024 JCR. NeuroImage's exceptional citation rate, despite publishing relatively few papers, signals that neuroimaging data occupies a central position in AI-driven neurodegenerative research. The dual-map overlay reinforces the field's multidisciplinary integration, with strong citation pathways linking computational methods to molecular biology and clinical neurology. The authors also identify key research hotspots. These hotspots include intelligent neuroimaging analysis, machine learning methodological iterations, molecular mechanisms and drug discovery, and clinical decision support systems (CDSS — software that helps clinicians make diagnostic or treatment decisions). These hotspots also include CDSS for early diagnosis. ## Why Some Countries Publish More Than Others The dominance of the USA and China reflects their substantial investment in AI research for neurodegenerative diseases. That is a straightforward conclusion based on volume alone. But volume is not the same as influence. The UK's higher citations per publication (31.68 versus China's 19.86 and the USA's 28.09) and its highest centrality (0.24) suggest the UK makes impactful contributions through international collaboration rather than sheer output. India presents a different pattern. Its high publication output (138 articles, 9.84%) comes with the lowest average citations per publication in the top 10 (13.62). The authors suggest this may indicate a focus on quantity over quality, or limited engagement with the global research community. To address these disparities, the authors argue that developing countries need to strengthen international collaboration. ## What This Means for Patients This analysis does not test any treatment or device in patients. Still, it carries practical meaning for people living with neurodegenerative diseases and their families. First, the volume of research matters. More than half of all publications in this field appeared since 2023. A field expanding that quickly is one where new diagnostic tools and drug candidates are likely to emerge in the coming years. Second, the most-studied topics point to what is closest to clinical use. The top keywords — Alzheimer's disease, Parkinson's disease, MRI, convolutional neural networks, biomarkers, dementia, classification, mild cognitive impairment, neuroimaging, and feature extraction — cluster around **early detection and diagnosis** using brain imaging. That is where patients are most likely to encounter AI first, for example in software that helps radiologists read brain scans or flags subtle changes that might indicate early disease. Third, the shift in keyword bursts from older methods toward deep neural networks means AI tools are becoming more capable at pattern recognition. That matters for conditions like Alzheimer's, where changes in the brain begin years before symptoms appear. Fourth, the research on molecular mechanisms and drug discovery suggests AI is being used not just to diagnose but to find treatments. For instance, AI can screen compounds or identify biological targets faster than traditional lab work alone. Finally, the emphasis on clinical decision support systems indicates the goal is to give doctors better information at the point of care, not to replace their judgment. ## Limitations of This Analysis A bibliometric review describes publication patterns — it cannot prove that any AI tool works or improves patient outcomes. Several specific constraints apply. - **Single database.** All data came from the Web of Science Core Collection. Research indexed only in other databases would be missed. - **Language restriction.** Only English-language publications were included, which may disadvantage non-English-speaking countries and bias the country rankings. - **Document types.** Only "Article" and "Review" were counted, excluding conference papers, editorials, and other formats that are common in computer science. - **Citation lag.** Recently published work has had less time to accumulate citations. This can understate the influence of newer research and overstate that of older papers. - **Ranking by volume.** Publication counts and citation counts measure activity and attention, not clinical value. - **Keyword dependence.** Keyword analyses rely on what authors choose to list, which varies between journals and research traditions. The authors also note that the field's separate research clusters show strong internal collaboration but limited connection between groups. That fragmentation is a limitation of the research landscape itself, and it is one reason findings may take longer to reach patients. ## Priorities and Recommendations Going Forward The authors outline several priorities for the next phase of research. These are directions for the scientific community rather than instructions for patients. 1. **Advanced deep learning architectures.** Develop more sophisticated AI models, including transformers (a newer AI architecture originally developed for language, now applied to other data) and vision transformers (ViT — transformers adapted for image analysis). 1. **Multi-omics integration.** Combine different "omics" data layers — genomics (DNA), transcriptomics (gene activity), proteomics (proteins), and others — into unified analyses. 1. **Explainable AI systems.** Build AI that shows how it reached a conclusion, so clinicians can trust and verify its output. 1. **Digital biomarker-based early detection.** Use digital signals — for example from sensors, voice, or movement tracking — to detect disease earlier. 1. **Transformative technologies.** Pursue innovations including transformers and telemedicine (remote clinical care delivered through technology). 1. **Interdisciplinary collaboration.** Strengthen teamwork between computational scientists and clinicians so that technical advances translate into real clinical use. The authors advocate for enhanced interdisciplinary collaboration to bridge computational advances with clinical translation — in other words, to make sure laboratory and software breakthroughs actually reach patients. ## Frequently Asked Questions ### What does it mean that the United Kingdom had the highest average citations per publication (31.68)? Average citations per publication measures how often a country's papers are cited by other researchers. In this analysis, the United Kingdom's 31.68 average was higher than China's 19.86 and the United States' 28.09. This suggests the United Kingdom's contributions had unusual influence, achieved through international collaboration rather than sheer publication volume. ### What is a bibliometric analysis, and can it tell me whether an AI tool works for patients? A bibliometric analysis is a statistical study of published research itself, counting papers, tracking authors, and mapping citations. It cannot tell you whether a treatment or AI tool works or improves patient outcomes. It can only describe where scientific effort is going, such as which topics are growing and which countries or institutions are most active. ### Which topics related to Alzheimer's and Parkinson's were studied most in this AI research? The most studied topics were Alzheimer's disease, Parkinson's disease, MRI brain scans, convolutional neural networks (an AI method for analyzing images), and biomarkers (biological signals of disease). Mild cognitive impairment (memory problems worse than normal aging but not yet dementia) was also among the most studied topics. These topics cluster around early detection and diagnosis using brain imaging. ### What does the shift from support vector machines to deep neural networks mean for patients? Keyword bursts showed support vector machine peaked earlier (2019–2022), while neural networks and deep neural networks surged most recently (2022–2023). This shift means AI tools are moving from older statistical classification methods toward deep learning, which is becoming more capable at pattern recognition. That matters for conditions like Alzheimer's, where brain changes begin years before symptoms appear. ### What are the limitations of this analysis that I should keep in mind? The analysis used only one database (Web of Science Core Collection), included only English-language publications, and counted only articles and reviews, excluding conference papers. Recently published work has had less time to accumulate citations. Publication and citation counts measure activity and attention, not clinical value. The authors also noted limited connection between separate research groups. ### What did the authors recommend for the future of AI in neurodegenerative disease research? The authors recommended developing advanced deep learning architectures such as transformers and vision transformers. The authors also recommended integrating multi-omics data (genomics, transcriptomics, proteomics). The authors further recommended building explainable AI systems that show how they reached a conclusion. The authors also recommended using digital biomarkers for early detection, pursuing telemedicine, and strengthening teamwork between computational scientists and clinicians so technical advances reach patients. ### When should someone with early memory problems or a possible Alzheimer's or Parkinson's diagnosis seek a second opinion? A bibliometric review of 1,402 publications found that AI research in neurodegenerative diseases clusters around early detection and diagnosis using brain imaging. Alzheimer's disease, Parkinson's disease, MRI, convolutional neural networks, biomarkers, and mild cognitive impairment were among the most studied topics. Changes in the brain can begin years before symptoms appear. AI tools are not yet in routine patient care. A second opinion can help confirm whether imaging and biomarker findings truly support an early diagnosis. This confirmation matters before treatment decisions are made. Diagnostic Detectives Network provides independent expert second opinions. ## Source Information **Original article title:** Artificial intelligence in neurodegenerative diseases research: a bibliometric analysis since 2000. **Authors:** Yabin Zhang (co-first author), Lei Yu (co-first author), Yuting Lv, Tiantian Yang, and Qi Guo (corresponding author). **Author affiliations:** Department of Special Services, The 960th Hospital of the PLA Joint Logistics Support Force, Jinan, Shandong, China (Zhang, Yu, Guo); Campus Clinic, Shandong University of Traditional Chinese Medicine, Jinan, Shandong, China (Lv); Department of Traditional Chinese Medicine, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, Shandong, China (Yang). **Journal:** *Frontiers in Neurology*, volume 16, article 1607924. **Article type:** Review. **Publication details:** Received 08 April 2025; accepted 03 July 2025; published 16 July 2025; corrected 21 July 2025. **DOI:** 10.3389/fneur.2025.1607924 **Handling editor:** Chuanming Li, Chongqing University Central Hospital, China. **Reviewers:** Shinya Sonobe, Tohoku University, Japan; Gauri Sabherwal, Chitkara University, India. **Abbreviations used in the original article:** AI (artificial intelligence); ANN (artificial neural network); ACP (average citations per publication); APY (average publication year); CDSS (clinical decision support systems); CNN (convolutional neural network); DNN (deep neural network); MRI (magnetic resonance imaging); NP (number of publications); SCI-EXPANDED (Science Citation Index Expanded); SVM (support vector machine); TC (total citations); ViT (vision transformer); WoSCC (Web of Science Core Collection). *This patient-friendly article is based on peer-reviewed research. It is an independent plain-language description of a published bibliometric review and is not a substitute for advice from a qualified healthcare professional.* --- Publisher: Diagnostic Detectives Network (https://diagnosticdetectives.com) — independent multi-expert medical second opinions, worldwide, private-pay. Author byline: Anton Titov, MD, PhD. Contact: https://diagnosticdetectives.com/pages/contact Canonical page: https://diagnosticdetectives.com/products/how-artificial-intelligence-is-reshaping-brain-disease-research-a-25-year-scientific-map