{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,4,22]],"date-time":"2026-04-22T21:38:55Z","timestamp":1776893935546,"version":"3.51.2"},"reference-count":71,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2025,11,6]],"date-time":"2025-11-06T00:00:00Z","timestamp":1762387200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Alzheimer\u2019s disease and related dementias (ADRD) affect nearly five million older adults in the United States, yet more than half remain undiagnosed. Speech-based natural language processing (NLP) provides a scalable approach to identify early cognitive decline by detecting subtle linguistic markers that may precede clinical diagnosis.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Objective<\/jats:title>\n                    <jats:p>This study aims to develop and evaluate a speech-based screening pipeline that integrates transformer-based embeddings with handcrafted linguistic features, incorporates synthetic augmentation using large language models (LLMs), and benchmarks unimodal and multimodal LLM classifiers. External validation was performed to assess generalizability to an MCI-only cohort.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>\n                      Transcripts were obtained from the ADReSSo 2021 benchmark dataset (\n                      <jats:italic>n<\/jats:italic>\n                      \u202f=\u202f237; derived from the Pitt Corpus, DementiaBank) and the DementiaBank Delaware corpus (\n                      <jats:italic>n<\/jats:italic>\n                      \u202f=\u202f205; clinically diagnosed mild cognitive impairment [MCI] vs. controls). Audio was automatically transcribed using Amazon Web Services Transcribe (general model). Ten transformer models were evaluated under three fine-tuning strategies. A late-fusion model combined embeddings from the best-performing transformer with 110 linguistically derived features. Five LLMs (LLaMA-8B\/70B, MedAlpaca-7B, Ministral-8B, GPT-4o) were fine-tuned to generate label-conditioned synthetic speech for data augmentation. Three multimodal LLMs (GPT-4o, Qwen-Omni, Phi-4) were tested in zero-shot and fine-tuned settings.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>On the ADReSSo dataset, the fusion model achieved an F1-score of 83.32 (AUC\u202f=\u202f89.48), outperforming both transformer-only and linguistic-only baselines. Augmentation with MedAlpaca-7B synthetic speech improved performance to F1\u202f=\u202f85.65 at 2\u202f\u00d7\u202fscale, whereas higher augmentation volumes reduced gains. Fine-tuning improved unimodal LLM classifiers (e.g., MedAlpaca-7B, F1\u202f=\u202f47.73\u202f\u2192\u202f78.69), while multimodal models demonstrated lower performance (Phi-4\u202f=\u202f71.59; GPT-4o omni\u202f=\u202f67.57). On the Delaware corpus, the pipeline generalized to an MCI-only cohort, with the fusion model plus 1\u202f\u00d7\u202fMedAlpaca-7B augmentation achieving F1\u202f=\u202f72.82 (AUC\u202f=\u202f69.57).<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusion<\/jats:title>\n                    <jats:p>\n                      Integrating transformer embeddings with handcrafted linguistic features enhances ADRD detection from speech. Distributionally aligned LLM-generated narratives provide effective but bounded augmentation, while current multimodal models remain limited. Crucially, validation on the Delaware corpus demonstrates that the proposed pipeline generalizes to early-stage impairment, supporting its potential as a scalable approach for clinically relevant early screening. All codes for LLMCARE are publicly available at:\n                      <jats:uri>GitHub<\/jats:uri>\n                      .\n                    <\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/frai.2025.1669896","type":"journal-article","created":{"date-parts":[[2025,11,6]],"date-time":"2025-11-06T12:54:35Z","timestamp":1762433675000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":4,"title":["LLMCARE: early detection of cognitive impairment via transformer models enhanced by LLM-generated synthetic data"],"prefix":"10.3389","volume":"8","author":[{"given":"Ali","family":"Zolnour","sequence":"first","affiliation":[]},{"given":"Hossein","family":"Azadmaleki","sequence":"additional","affiliation":[]},{"given":"Yasaman","family":"Haghbin","sequence":"additional","affiliation":[]},{"given":"Fatemeh","family":"Taherinezhad","sequence":"additional","affiliation":[]},{"given":"Mohamad Javad Momeni","family":"Nezhad","sequence":"additional","affiliation":[]},{"given":"Sina","family":"Rashidi","sequence":"additional","affiliation":[]},{"given":"Masoud","family":"Khani","sequence":"additional","affiliation":[]},{"given":"AmirSajjad","family":"Taleban","sequence":"additional","affiliation":[]},{"given":"Samin Mahdizadeh","family":"Sani","sequence":"additional","affiliation":[]},{"given":"Maryam","family":"Dadkhah","sequence":"additional","affiliation":[]},{"given":"James M.","family":"Noble","sequence":"additional","affiliation":[]},{"given":"Suzanne","family":"Bakken","sequence":"additional","affiliation":[]},{"given":"Yadollah","family":"Yaghoobzadeh","sequence":"additional","affiliation":[]},{"given":"Abdol-Hossein","family":"Vahabie","sequence":"additional","affiliation":[]},{"given":"Masoud","family":"Rouhizadeh","sequence":"additional","affiliation":[]},{"given":"Maryam","family":"Zolnoori","sequence":"additional","affiliation":[]}],"member":"1965","published-online":{"date-parts":[[2025,11,6]]},"reference":[{"key":"ref35","author":"Abouelenin","year":"2025"},{"key":"ref1","author":"Alsentzer","year":"2019"},{"key":"ref2","doi-asserted-by":"publisher","first-page":"208","DOI":"10.1016\/j.jalz.2013.02.003","article-title":"Alzheimer\u2019s disease facts and figures","volume":"9","year":"2013","journal-title":"Alzheimers Dement."},{"key":"ref3","doi-asserted-by":"publisher","first-page":"e0155195","DOI":"10.1371\/journal.pone.0155195","article-title":"Vocabulary size in speech may be an early indicator of cognitive impairment","volume":"11","author":"Aramaki","year":"2016","journal-title":"PLoS One"},{"key":"ref4","doi-asserted-by":"publisher","first-page":"219","DOI":"10.1016\/j.trci.2017.01.006","article-title":"Predicting mild cognitive impairment from spontaneous spoken utterances","volume":"3","author":"Asgari","year":"2017","journal-title":"Alzheimer Dementia"},{"key":"ref5","doi-asserted-by":"publisher","first-page":"1856","DOI":"10.3233\/SHTI251249","article-title":"S: harnessing multimodal innovation to transform cognitive impairment detection-insights from the National Institute on Aging Alzheimer\u2019s speech challenge","volume":"329","author":"Azadmaleki","year":"2025","journal-title":"Stud. Health Technol. Inform."},{"key":"ref6","doi-asserted-by":"publisher","first-page":"2167","DOI":"10.48550\/arXiv.2008.01551","article-title":"To BERT or not to BERT: Comparing speech and language-based approaches for Alzheimer\u2019s disease detection","author":"Balagopalan","year":"2020"},{"key":"ref7","doi-asserted-by":"publisher","first-page":"96","DOI":"10.4218\/etrij.2023-0356","article-title":"Alzheimer\u2019s disease recognition from spontaneous speech using large language models","volume":"46","author":"Bang","year":"2024","journal-title":"ETRI J."},{"key":"ref8","volume-title":"Longformer: The Long-Document Transformer arXiv. Ithaca, NY, USA","author":"Beltagy","year":"2020"},{"key":"ref9","doi-asserted-by":"publisher","first-page":"M621","DOI":"10.1093\/gerona\/59.6.M621","article-title":"Dementia assessment in primary care: results from a study in three managed care systems","volume":"59","author":"Boise","year":"2004","journal-title":"J. Gerontol. A Biol. Sci. Med. Sci."},{"key":"ref10","doi-asserted-by":"publisher","first-page":"e28244","DOI":"10.2196\/28244","article-title":"Behavioral activation and depression symptomatology: longitudinal assessment of linguistic indicators in text-based therapy sessions","volume":"23","author":"Burkhardt","year":"2021","journal-title":"J. Med. Internet Res."},{"key":"ref11","doi-asserted-by":"publisher","first-page":"101113","DOI":"10.1016\/j.csl.2020.101113","article-title":"Linguistic features and automatic classifiers for identifying mild cognitive impairment and dementia","volume":"65","author":"Calz\u00e0","year":"2021","journal-title":"Comput. Speech Lang."},{"key":"ref12","doi-asserted-by":"publisher","first-page":"743","DOI":"10.1037\/a0017579","article-title":"Language-based measures of mindfulness: initial validity and clinical utility","volume":"23","author":"Collins","year":"2009","journal-title":"Psychol. Addict. Behav."},{"key":"ref13","author":"Devlin","year":"2018"},{"key":"ref14","doi-asserted-by":"publisher","first-page":"325","DOI":"10.1007\/s10462-024-10961-6","article-title":"Speech based detection of Alzheimer\u2019s disease: a survey of AI techniques, datasets and challenges","volume":"57","author":"Ding","year":"2024","journal-title":"Artif. Intell. Rev."},{"key":"ref15","doi-asserted-by":"publisher","first-page":"407","DOI":"10.3233\/JAD-150520","article-title":"Linguistic features identify Alzheimer\u2019s disease in narrative speech","volume":"49","author":"Fraser","year":"2016","journal-title":"J Alzheimer's Dis"},{"key":"ref16","doi-asserted-by":"publisher","first-page":"388","DOI":"10.1111\/ane.13216","article-title":"Identifying epilepsy psychiatric comorbidities with machine learning","volume":"141","author":"Glauser","year":"2020","journal-title":"Acta Neurol. Scand."},{"key":"ref17","first-page":"2672","article-title":"Generative adversarial networks","volume":"3","author":"Goodfellow","year":"2014","journal-title":"Sci Robot"},{"key":"ref18","volume-title":"The Llama 3 herd of models. arXiv. Ithaca, NY, USA","author":"Grattafiori","year":"2024"},{"key":"ref19","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1038\/s41746-023-00970-0","article-title":"Large language models to identify social determinants of health in electronic health records","volume":"7","author":"Guevara","year":"2024","journal-title":"NPJ Digit. Med."},{"key":"ref20","doi-asserted-by":"publisher","first-page":"642517","DOI":"10.3389\/fcomp.2021.642517","article-title":"Crossing the \u201ccookie theft\u201d corpus chasm: applying what BERT learns from outside data to the ADReSS challenge dementia detection task","volume":"3","author":"Guo","year":"2021","journal-title":"Front Comput Sci"},{"key":"ref21","author":"Han","year":"2025"},{"key":"ref22","doi-asserted-by":"publisher","first-page":"784","DOI":"10.3233\/SHTI250947","article-title":"Optimizing entity recognition in psychiatric treatment data with large language models","volume":"329","author":"Hosseini","year":"2025","journal-title":"Stud. Health Technol. Inform."},{"key":"ref43","author":"Hurst","year":"2024"},{"key":"ref23","doi-asserted-by":"publisher","first-page":"4153","DOI":"10.1109\/JBHI.2022.3172479","article-title":"Explainable identification of dementia from transcripts using transformer networks","volume":"26","author":"Ilias","year":"2022","journal-title":"IEEE J. Biomed. Health Inform."},{"key":"ref24","volume-title":"Synthetic data generation with LLM for improved depression prediction. arXiv. Ithaca, NY, USA","author":"Kang","year":"2025"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1007\/978-1-4939-1985-7_11","article-title":"Evaluation of linguistic and prosodic features for detection of Alzheimer\u2019s disease in Turkish conversational speech","volume":"9","author":"Khodabakhsh","year":"2015","journal-title":"Eurasip J. Audio Speech Music Process."},{"key":"ref26","doi-asserted-by":"publisher","first-page":"2217","DOI":"10.21437\/Interspeech.2020-3153","article-title":"Exploiting multi-modal features from pre-trained networks for Alzheimer\u2019s dementia recognition","author":"Koo","year":"2020"},{"key":"ref27","doi-asserted-by":"publisher","first-page":"426","DOI":"10.1044\/2022_AJSLP-22-00281","article-title":"DementiaBank: theoretical rationale, protocol, and illustrative analyses","volume":"32","author":"Lanzi","year":"2023","journal-title":"Am. J. Speech Lang. Pathol."},{"key":"ref28","doi-asserted-by":"publisher","first-page":"1234","DOI":"10.1093\/bioinformatics\/btz682","article-title":"BioBERT: a pre-trained biomedical language representation model for biomedical text mining","volume":"36","author":"Lee","year":"2020","journal-title":"Bioinformatics"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2201.11838","article-title":"Clinical-Longformer and clinical-BigBird: Transformers for long clinical sequences","author":"Li","year":"2022","journal-title":"arXiv"},{"key":"ref30","doi-asserted-by":"publisher","first-page":"13887","DOI":"10.1038\/s41598-024-64438-1","article-title":"Multimodal deep learning for dementia classification using text and audio","volume":"14","author":"Lin","year":"2024","journal-title":"Sci. Rep."},{"key":"ref31","volume-title":"A robustly optimized BERT Pretraining approach. arXiv. Ithaca, NY, USA","author":"Liu","year":"2019"},{"key":"ref32","doi-asserted-by":"publisher","first-page":"620251","DOI":"10.3389\/fpsyg.2021.620251","article-title":"Ten years of research on automatic voice and speech analysis of people with Alzheimer\u2019s disease and mild cognitive impairment: a systematic review article","volume":"12","author":"Mart\u00ednez-Nicol\u00e1s","year":"2021","journal-title":"Front. Psychol."},{"key":"ref33","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1155\/2020\/4683573","article-title":"Changes in the rhythm of speech difference between people with nondegenerative mild cognitive impairment and with preclinical dementia","volume":"2020","author":"Meil\u00e1n","year":"2020","journal-title":"Behav. Neurol."},{"key":"ref34","doi-asserted-by":"publisher","first-page":"17","DOI":"10.1016\/j.cortex.2013.02.013","article-title":"Ever decreasing circles: speech production in semantic dementia","volume":"55","author":"Meteyard","year":"2014","journal-title":"Cortex"},{"key":"ref36","author":"Mistral","year":"2024"},{"key":"ref37","year":"2021"},{"key":"ref38","year":""},{"key":"ref39","doi-asserted-by":"publisher","first-page":"405","DOI":"10.1044\/jshr.2803.405","article-title":"Empty speech in Alzheimer\u2019s disease and fluent aphasia","volume":"28","author":"Nicholas","year":"1985","journal-title":"J. Speech Lang. Hear. Res."},{"key":"ref40","doi-asserted-by":"publisher","first-page":"931","DOI":"10.1111\/jgs.14716","article-title":"Impact of the REACH II and REACH VA dementia caregiver interventions on healthcare costs","volume":"65","author":"Nichols","year":"2017","journal-title":"J. Am. Geriatr. Soc."},{"key":"ref41","doi-asserted-by":"publisher","first-page":"e0251787","DOI":"10.1371\/journal.pone.0251787","article-title":"The relationship between linguistic expression in blog content and symptoms of depression, anxiety, and suicidal thoughts: a longitudinal study","volume":"16","author":"O\u2019Dea","year":"2021","journal-title":"PLoS One"},{"key":"ref42","volume-title":"GPT-4 Technical Report. San Francisco, CA, USA: OpenAI","year":"2023"},{"key":"ref44","doi-asserted-by":"publisher","first-page":"419","DOI":"10.1016\/S0010-9452(08)70257-0","article-title":"An investigation of semantic errors in unimpaired and Alzheimer\u2019s speakers of Italian","volume":"39","author":"Paganelli","year":"2003","journal-title":"Cortex"},{"key":"ref45","first-page":"3810","article-title":"Using the outputs of different automatic speech recognition paradigms for acoustic-and BERT-based Alzheimer\u2019s dementia detection through spontaneous speech","author":"Pan","year":"2021"},{"key":"ref46","author":"Papineni","year":"2002"},{"key":"ref47","first-page":"2177","volume-title":"Using state of the art speaker recognition and natural language processing technologies to detect Alzheimer\u2019s disease and assess its severity","author":"Pappagari","year":"2020"},{"key":"ref48","author":"Peng","year":"2019"},{"key":"ref49","volume-title":"A survey on speech large language models. arXiv. Ithaca, NY, USA","author":"Peng","year":"2024"},{"key":"ref50","doi-asserted-by":"publisher","first-page":"3805","DOI":"10.48550\/arXiv.2106.08689","article-title":"Alzheimer\u2019s disease detection from spontaneous speech through combining linguistic complexity and (dis)fluency features with pretrained language models","author":"Qiao","year":"2021"},{"key":"ref51","first-page":"1858","article-title":"SpeechCura: a novel speech augmentation framework to tackle data scarcity in healthcare","volume-title":"Stud health Technol inform","author":"Rashidi","year":"2025"},{"key":"ref52","volume-title":"DistilBERT, a distilled version of BERT: Smaller, faster, cheaper and lighter. arXiv. Ithaca, NY, USA","author":"Sanh","year":"2019"},{"key":"ref53","doi-asserted-by":"crossref","first-page":"101814","DOI":"10.1016\/j.csl.2025.101814","article-title":"Modality fusion using auxiliary tasks for dementia detection","volume":"95","author":"Shao","year":"2025","journal-title":"Comput. Speech Lang."},{"key":"ref54","doi-asserted-by":"publisher","first-page":"1416","DOI":"10.1044\/2020_JSLHR-19-00335","article-title":"Syntactic complexity as a linguistic marker to differentiate mild cognitive impairment from Normal aging","volume":"63","author":"Sung","year":"2020","journal-title":"J. Speech Lang. Hear. Res."},{"key":"ref55","author":"Syed","year":"2021"},{"key":"ref56","doi-asserted-by":"publisher","first-page":"100046","DOI":"10.1016\/j.nlp.2023.100046","article-title":"Context is not key: detecting Alzheimer\u2019s disease with both classical and transformer-based neural language models","volume":"6","author":"TaghiBeyglou","year":"2024","journal-title":"Nat. Lang. Proc. J."},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2509.03525","article-title":"Speech-based cognitive screening: A systematic evaluation of LLM adaptation strategies","author":"Taherinezhad","year":"2025","journal-title":"arXiv"},{"key":"ref58","doi-asserted-by":"publisher","first-page":"204","DOI":"10.1097\/00002093-199601040-00006","article-title":"Cross-sectional analysis of Alzheimer disease effects on oral discourse in a picture description task","volume":"10","author":"Tomoeda","year":"1996","journal-title":"Alzheimer Dis. Assoc. Disord."},{"key":"ref59","doi-asserted-by":"publisher","first-page":"130","DOI":"10.2174\/1567205014666171121114930","article-title":"A speech recognition-based solution for the automatic detection of mild cognitive impairment from spontaneous speech","volume":"15","author":"T\u00f3th","year":"2018","journal-title":"Curr. Alzheimer Res."},{"key":"ref60","year":"2025"},{"key":"ref61","first-page":"2579","article-title":"Visualizing Data using t-SNE","volume":"9","author":"van der Maaten","year":"2008","journal-title":"J. Mach. Learn. Res."},{"key":"ref62","doi-asserted-by":"publisher","first-page":"429","DOI":"10.48550\/arXiv.2405.06695","article-title":"Utilizing large language models to generate synthetic data to increase the performance of BERT-based neural networks","volume":"2024","author":"Woolsey","year":"2024","journal-title":"AMIA Sum. Transl. Sci. Proc."},{"key":"ref63","volume-title":"C-pack: Packaged resources to advance general Chinese embedding. Red Hook, NY, USA: ICLR 2024 conference proceedings","author":"Xiao","year":"2023"},{"key":"ref64","author":"Xu","year":"2025"},{"key":"ref65","doi-asserted-by":"publisher","first-page":"5753","DOI":"10.48550\/arXiv.1906.08237","article-title":"XLNet: generalized autoregressive Pretraining for language understanding","volume":"32","author":"Yang","year":"2019","journal-title":"Adv Neural Inf Process Syst"},{"key":"ref66","author":"Zhang","year":"2020"},{"key":"ref67","doi-asserted-by":"publisher","first-page":"1966","DOI":"10.3233\/SHTI251302","article-title":"A scoping review of large language model applications in healthcare","volume":"329","author":"Zhang","year":"2025","journal-title":"Stud. Health Technol. Inform."},{"key":"ref68","doi-asserted-by":"publisher","DOI":"10.3389\/fcomp.2021.624683","article-title":"Exploring deep transfer learning techniques for alzheimer\u2019s dementia detection","volume":"3","author":"Zhu","year":"2021","journal-title":"Front. Comput. Sci."},{"key":"ref69","doi-asserted-by":"publisher","first-page":"ooae130","DOI":"10.1093\/jamiaopen\/ooae130","article-title":"Decoding disparities: evaluating automatic speech recognition system performance in transcribing black and white patient verbal communication with nurses in home healthcare","volume":"7","author":"Zolnoori","year":"","journal-title":"JAMIA Open"},{"key":"ref70","doi-asserted-by":"publisher","first-page":"102624","DOI":"10.1016\/j.artmed.2023.102624","article-title":"ADscreen: a speech processing-based screening system for automatic identification of patients with Alzheimer\u2019s disease and related dementia","volume":"143","author":"Zolnoori","year":"2023","journal-title":"Artif. Intell. Med."},{"key":"ref71","doi-asserted-by":"publisher","first-page":"ocae300","DOI":"10.1093\/jamia\/ocae300","article-title":"Beyond electronic health record data: leveraging natural language processing and machine learning to uncover cognitive insights from patient-nurse verbal communications","volume":"32","author":"Zolnoori","year":"","journal-title":"J. Am. Med. Inform. Assoc."}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2025.1669896\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,11,6]],"date-time":"2025-11-06T12:54:39Z","timestamp":1762433679000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2025.1669896\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,11,6]]},"references-count":71,"alternative-id":["10.3389\/frai.2025.1669896"],"URL":"https:\/\/doi.org\/10.3389\/frai.2025.1669896","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,11,6]]},"article-number":"1669896"}}