Alzheimer's Disease (AD) poses a considerable global health challenge, necessitating early and accurate diagnostics. The Mini-Mental State Examination (MMSE) is widely used for initial screening, but its traditional application often underutilizes the rich multimodal data it generates, such as videos, images, and speech. Integrating these modalities with modern Large Language Models (LLMs) offers untapped potential for improved diagnostics. In this study, we propose a multimodal LLM framework fundamentally reinterprets MMSE data. Instead of relying on conventional, often limited MMSE features, proposed LLM acts as a sophisticated cognitive analyst, directly processing MMSE modalities. This deep multimodal understanding allows for the extraction of novel, high-level features that transcend traditional metrics. These are not merely visual or acoustic signals, but rich semantic representations imbued with cognitive insights gleaned by the LLM. We then construct an interpretable decision tree classifier and derive a succinct rule list, yielding transparent diagnostic pathways readily understandable by clinicians. Finally, framework integrates a counterfactual explanation module to provide individualized "what-if" analyses, illuminating how minimal feature changes could alter model outputs. Our empirical study on real-world clinical data achieves a diagnostic accuracy of approximately 6% percentage points improvements with diagnosing explanation, reinforcing the viability of our framework as a promising, interpretable, and scalable solution for early AD detection.

Deep learning (DL) has revolutionized medical image analysis (MIA), enabling early anomaly detection, precise lesion segmentation, and automated disease classification. However, its clinical integration faces two major challenges: reliance on limited, narrowly annotated datasets that inadequately capture real-world patient diversity, and the inherent “black-box” nature of DL decision-making, which complicates physician scrutiny and accountability. Eye tracking (ET) technology offers a transformative solution by capturing radiologists’ gaze patterns to generate supervisory signals. These signals enhance DL models through two key mechanisms: providing weak supervision to improve feature recognition and diagnostic accuracy, particularly when labeled data are scarce, and enabling direct comparison between machine and human attention to bridge interpretability gaps and build clinician trust. This approach also extends effectively to multimodal learning models (MLMs) and vision–language models (VLMs), supporting the alignment of machine reasoning with clinical expertise by grounding visual observations in diagnostic context, refining attention mechanisms, and validating complex decision pathways. Conducted in accordance with the PRISMA statement and registered in PROSPERO (ID: CRD42024569630), this review synthesizes state-of-the-art strategies for ET-DL integration. We further propose a unified framework in which ET innovatively serves as a data efficiency optimizer, a model interpretability validator, and a multimodal alignment supervisor. This framework paves the way for clinician-centered AI systems that prioritize verifiable reasoning, seamless workflow integration, and intelligible performance, thereby addressing key implementation barriers and outlining a path for future clinical deployment.

Background Alzheimer’s disease (AD) is a neurodegenerative disorder characterised by cognitive decline, behavioural and psychological symptoms of dementia (BPSD) and impairment of activities of daily living (ADL). Early differentiation of AD from mild cognitive impairment (MCI) is necessary. Methods A total of 458 patients newly diagnosed with AD and MCI were included. Eleven batteries were used to evaluate ADL, BPSD and cognitive function (ABC). Machine learning approaches including XGboost, classification and regression tree, Bayes, support vector machines and logical regression were used to build and verify the new tool. Results The Alzheimer’s Disease Assessment Scale (ADAS-cog) word recognition task showed the best importance in judging AD and MCI, followed by correct numbers of auditory verbal learning test delay recall and ADAS-cog orientation. We also provided a selected ABC-Scale that covered ADL, BPSD and cognitive function with an estimated completion time of 18 min. The sensitivity was improved in the four models. Conclusion The quick screen ABC-Scale covers three dimensions of ADL, BPSD and cognitive function with good efficiency in differentiating AD from MCI.

We developed and evaluated a training-free, large language model (LLM) multiagent framework, consisting of role-prompted LLM instances, that simulates hospital-style team diagnosis for Alzheimer's disease, improving interpretability and generalizability for multimodal data. We designed a self-learning and critical multiagent workflow comprising nine single-dimension agents and five senior agents that iteratively validate, retrieve experience, and refine decisions via self-evaluation and critique. Agents are augmented with an "intelligence" module and an experience-retrieval module. Participants were 1,362 Alzheimer's Disease Neuroimaging Initiative (ADNI) records spanning cognitively normal, mild cognitive impairment, and Alzheimer's disease from ADNI GO, ADNI 1, ADNI 2, ADNI 3, ADNI 4. The decision layer is not supervised on ADNI labels, and no end-to-end fine-tuning is performed on ADNI within this study. The full self-learning and critical multiagent framework achieved an F1 score = 88%, sensitivity = 88%, and precision = 90% for three-way cognitively normal/mild cognitive impairment/Alzheimer's disease classification, providing case-level rationales. Relative to early agents, sequential self-learning and critique produced large effect gains. Single-dimension agents varied widely, while the integrated framework yielded the most balanced performance and higher answer relevancy on LLM-based metrics. A label-free LLM multiagent approach can integrate heterogeneous modalities, retain diagnostic "experience," and generate transparent reasoning while delivering competitive accuracy on ADNI. The framework shows promise for real-world support of complex neuropsychological diagnosis; future work should incorporate fully multimodal LLMs for imaging and repeated, graph-driven iterations to further enhance performance. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

Background: Alzheimer’s Disease (AD) is a neurodegenerative disorder characterized by cognitive functional decline accompanied with neuropsychiatric symptoms and impairment of activities of daily living (ADL). Mild Cognitive Impairment (MCI) is considered the prodromal state of AD. Early identification of MCI is a strategy to reduce the burden of caregivers and society. Neuropsychological scales are still the most used method for identification of cognitive decline. The purpose of this study was to simplify the scales by eliminating some insensitive items based on the six years of database and machine learning methods.Method: A total of 458 patients newly diagnosed as AD or MCI were recruited in Memory Clinic, Department of Geriatrics, The First Affiliated Hospital of Chongqing Medical University from 2014 to 2019. Neuropsychological battery was used to evaluate the cognitive function, neuropsychiatric symptoms and ADL, including Mini‐mental State Examination (MMSE), Neuropsychiatric Inventory (NPI), Geriatric Depression Scale (GDS), Boston Naming Test (BNT), Digit Span (DS), Auditory Verbal Learning Test (AVLT), Trail Marking Test (TMT) A and B. Alzheimer’s Disease Assessment Scale (ADAS‐cog), Clinical Dementia Rating (CDR), Instrumental Activity of Daily Living Scale (IADL), Physical Self‐Maintenance Scale (PSMS). XGboost and CART regression tree were used to test all items contained in all scales and try to explore the importance of each item. Sensitivity was reflected by the AUC (Area Under Curve).Result: Of which 236 patients were diagnosed as AD and 222 as MCI. The mean age was 72.99±9.04 years and 51.53% were male. Among all the battery items, ADAs‐cog word recognition task showed the best importance in judging AD and MCI, followed correct numbers of AVLT delay recall, ADAs‐cog orientation, MMSE recall, TMT‐B time. The overall sensitivity for predicting AD or MCI for all items was 0.83. However, in our preliminary screening also showed the 31 items, most of which were based on the ADL scale and neuropsychiatry symptoms, contributed little to distinguishing AD from MCI. After removed these items, the sensitivity became to 0.87.Conclusion: Cognitive assessment such as recall and orientation may be a better strategy than neuropsychiatric symptoms and ADL when identifying AD from MCI.