AD-CARE: A Guideline-grounded, Modality-agnostic LLM Agent for Real-world Alzheimer's Disease Diagnosis with Multi-cohort Assessment, Fairness Analysis, and Reader Study
Authors: Wenlong Hou, Sheng Bi, Guangqian Yang, Lihao Liu, Ye Du, Hanxiao Xue, Juncheng Wang, Yuxiang Feng, Yue Xun, Nanxi Yu, Ning Mao, Mo Yang, Yi Wah Eva Cheung, Ling Long, Kay Chen Tan, Lequan Yu, Xiaomeng Ma, Shaozhen Yan, Shujun Wang
Organizations: Department of Biomedical Engineering, The Hong Kong Polytechnic University, Hong Kong SAR, China. · Department of Radiology and Nuclear Medicine, Xuanwu Hospital, Capital Medical University, Beijing, China. · Amazon, USA. · College of Control Science and Engineering, Zhejiang University, Hangzhou, China. · Sany AI, Hangzhou, China. · Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, China. · Department of Diagnostic Radiology, The University of Hong Kong, Hong Kong SAR, China. · Department of Geriatrics, The Third Affiliated Hospital of Sun Yat-sen University, Guangzhou, China. · Department of Data Science and Artificial Intelligence, The Hong Kong Polytechnic University, Hong Kong SAR, China. · School of Computing and Data Science, The University of Hong Kong, Hong Kong SAR, China. · Department of Neurology, Mental and Neurological Diseases Research Center, The Third Affiliated Hospital of Sun Yat-sen University, Guangzhou, China. · Research Institute for Smart Ageing, The Hong Kong Polytechnic University, Hong Kong SAR, China. · Research Institute for Sports Science and Technology, The Hong Kong Polytechnic University, Hong Kong SAR, China.
Alzheimer's disease (AD) is a growing global health challenge as populations age, and timely, accurate diagnosis is essential to reduce individual and societal burden. However, real-world AD assessment is hampered by incomplete, heterogeneous multimodal data and variability across sites and patient demographics. Although large language models (LLMs) have shown promise in biomedicine, their use in AD has largely been confined to answering narrow, disease-specific questions rather than generating comprehensive diagnostic reports that support clinical decision-making. Here we expand LLM capabilities for clinical decision support by introducing AD-CARE, a modality-agnostic agent that performs guideline-grounded diagnostic assessment from incomplete, heterogeneous inputs without imputing missing modalities. By dynamically orchestrating specialized diagnostic tools and embedding clinical guidelines into LLM-driven reasoning, AD-CARE generates transparent, report-style outputs aligned with real-world clinical workflows. Across six cohorts comprising 10,303 cases, AD-CARE achieved 84.9% diagnostic accuracy, delivering 4.2%-13.7% relative improvements over baseline methods. Despite cohort-level differences, dataset-specific accuracies remain robust (80.4%-98.8%), and the agent consistently outperforms all baselines. AD-CARE reduced performance disparities across racial and age subgroups, decreasing the average dispersion of four metrics by 21%-68% and 28%-51%, respectively. In a controlled reader study, the agent improved neurologist and radiologist accuracy by 6%-11% and more than halved decision time. The framework yielded 2.29%-10.66% absolute gains over eight backbone LLMs and converges their performance. These results show that AD-CARE is a scalable, practically deployable framework that can be integrated into routine clinical workflows for multimodal decision support in AD.
Figures & tables
Figure 1: AD-CARE Agent framework and overall strategy. (a) , Our AD-CARE for AD diagnosis was developed using multi-modal data, including individual-level demographics, imaging, neurological tests, genetic information, functional evaluation, and biospecimen results. Multi-modal data are processed through the AD-CARE with three components (reasoning engine, outcome aggregator, and specialized executors), generating multi-modal outputs (disgnosis result, confidence, diagnosis report, and visualization results). (b) , Agent workflow: Given a use query, the framework performs reasoning in four stages: (i) observation, (ii) thought, (iii) action, and (iv) aggregation. (c) , Validation on six diverse populations (n=10,303) including four public datests and two in-house cohorts: We first evaluated AD-CARE against baseline methods using four metrics. We then assessed fairness with respect to race and age. Next, we conducted reader study with agent augmentation. Finally, we benchmarked AD-CARE by using eight representative LLM backbones.
cohort (group)
Age (y), mean ± s.d.
Male, n (%)
Education (y), mean ± s.d.
CDR, mean ± s.d.
Public cohorts
ADNI
NC [n = 1,859]
74.36±6.69
826, 44.43%
16.56±2.52
0.02±0.18
MCI [n = 2,316]
73.9±7.57
1,347, 58.16%
16.10±2.74
0.45±0.22
AD [n = 1,020]
75.85±7.76
551, 54.02%
15.45±2.86
0.94±0.50
P value
<1.0×10−3
<1.0×10−3
<1.0×10−3
<1.0×10−3
Table 1: Study population and demographic characteristics (n=10,303).
Figure 2: Performance comparison of AD-CARE and baseline models across six benchmark cohorts. AD-CARE achieves superior accuracy and F1 scores on most benchmarks, with notable gains on heterogeneous cohorts (ADNI, OASIS, NACC) and robust performance on in-house cohorts (SYSUH, XWH), underscoring its potential for real-world application.
Figure 3: Fairness analysis of AD-CARE and baseline methods. (a) , Racial subgroups (Asian, Black, White). (b) , Age subgroups (<65, 65–74, 75–84, ≥ 85). Bars show subgroup performance on four metrics. Lines (right axis) show fairness dispersion (standard deviation and max–min gap across subgroups). AD-CARE delivers both high diagnostic performance and lower variability across demographic groups compared with baseline methods, indicating improved robustness and fairness across race and age.
Figure 4: AD-CARE assistance improves clinicians’ diagnostic accuracy and efficiency. (a) , Diagnostic performance of neurologists and radiologists with and without agent assistance, stratified by seniority level. Points denote mean performance estimates for doctor-only reads and doctor-plus-agent reads, and error bars indicate 95% confidence intervals obtained by bootstrap resampling. Metrics include accuracy, F1 score, sensitivity and specificity. Across both specialties and experience levels, access to the agent yields consistent gains in all metrics. (b) , Effect of AD-CARE assistance on per-case reading time. Violin plots depict the distribution of decision times for unaided clinicians (blue) and agent-assisted clinicians (orange) overall and within each subgroup and site (SYSUH neurologists, XWH radiologists). Boxes summarize median and mean times, and inset annotations report mean ± 95% CI and the corresponding efficiency gain (ratio of unaided to assisted time). AD-CARE assistance substantially shortens reading time for both neurologists and radiologists while preserving or improving diagnostic performance.
Figure 5: Benchmark comparison of AD-CARE with raw LLM backbones and cost–accuracy trade-off analysis. (a) , Accuracy of standalone LLM backbones versus the corresponding LLM-powered AD-CARE system. Numbers indicate the absolute accuracy gain achieved by AD-CARE over the raw LLM output for each backbone. Across all eight models, AD-CARE consistently improves diagnostic accuracy, demonstrating the effectiveness of the framework. (b) , AD-CARE accuracy versus overall inference cost for each instantiated backbone. Point colors denote the LLM provider. The dashed line denotes the Pareto frontier of non-dominated backbones, for which no alternative achieves higher accuracy at lower cost. Bubble diameter is proportional to the relative improvement ratio.
Figure 6: The AD-CARE web interface and workflow. (1) , Clinical data entry: users enter text-based patient information—including demographics, neuropsychological test scores, biospecimen results, APOE status, and functional assessments, and may optionally upload supporting files such as structural MRI (NIfTI) and raw genomic data (VCF). (2) , Diagnostic analysis and output: the agent executes a staged pipeline (planning → tool execution → aggregation) with transparent progress tracking and access to intermediate results. (3) , Clinical report generation and downloading: AD-CARE presents the final staged diagnosis with confidence, clinical reasoning, supporting vs. contradicting evidence and conflict resolution, diagnostic-criteria alignment, recommendations, and generated visual outputs, and enables one-click export of the structured report (PDF/Word).
Alzheimer's disease (AD) diagnosis requires integrating neuroimaging with heterogeneous clinical evidence and reasoning under established criteria, yet most multimodal models remain opaque and weakly guideline-aligned. We present AD-Reasoning, a multimodal framework that couples structural MRI with six clinical modalities and a rule-based verifier to generate structured, NIA-AA-consistent diagnoses. AD-Reasoning combines modality-specific encoders, bidirectional cross-attention fusion, and reinforcement fine-tuning with verifiable rewards that enforce output format, guideline evidence coverage, and reasoning--decision consistency. We also release AD-MultiSense, a 10,378-visit multimodal QA dataset with guideline-validated rationales built from ADNI/AIBL. On AD-MultiSense, AD-Reasoning achieves state-of-the-art diagnostic accuracy and produces structured rationales that improve transparency over recent baselines, while providing transparent rationales.
Qiuhui Chen, Yushan Deng, Xuancheng Yao +1
School of Information Science and Engineering, East China University of Science and Technology, Shanghai, China · School of Computer Science, Shanghai Jiao Tong University, Shanghai, China
Deep learning models for medical image analysis often act as black boxes, seldom aligning with clinical guidelines or explicitly linking decisions to supporting evidence. This is especially critical in Alzheimer's disease (AD), where predictions should be grounded in both anatomical and clinical findings. We present EMAD, a vision-language framework that generates structured AD diagnostic reports in which each claim is explicitly grounded in multimodal evidence. EMAD uses a hierarchical Sentence-Evidence-Anatomy (SEA) grounding mechanism: (i) sentence-to-evidence grounding links generated sentences to clinical evidence phrases, and (ii) evidence-to-anatomy grounding localizes corresponding structures on 3D brain MRI. To reduce dense annotation requirements, we propose GTX-Distill, which transfers grounding behavior from a teacher trained with limited supervision to a student operating on model-generated reports. We further introduce Executable-Rule GRPO, a reinforcement fine-tuning scheme with verifiable rewards that enforces clinical consistency, protocol adherence, and reasoning-diagnosis coherence. On the AD-MultiSense dataset, EMAD achieves state-of-the-art diagnostic accuracy and produces more transparent, anatomically faithful reports than existing methods. We will release code and grounding annotations to support future research in trustworthy medical vision-language models.
Qiuhui Chen, Xuancheng Yao, Zhenglei Zhou +2
East China University of Science and Technology · Shanghai Jiao Tong University · Tencent
Accurate diagnosis of Alzheimer's disease (AD) requires handling tabular biomarker data, yet such data are often small and incomplete, where deep learning models frequently fail to outperform classical methods. Pretrained large language models (LLMs) offer few-shot generalization, structured reasoning, and interpretable outputs, providing a powerful paradigm shift for clinical prediction. We propose TAP-GPT Tabular Alzheimer's Prediction GPT, a domain-adapted tabular LLM framework built on TableGPT2 and fine-tuned for few-shot AD classification using tabular prompts rather than plain texts. We evaluate TAP-GPT across four ADNI-derived datasets, including QT-PAD biomarkers and region-level structural MRI, amyloid PET, and tau PET for binary AD classification. Across multimodal and unimodal settings, TAP-GPT improves upon its backbone models and outperforms traditional machine learning baselines in the few-shot setting while remaining competitive with state-of-the-art general-purpose LLMs. We show that feature selection mitigates degradation in high-dimensional inputs and that TAP-GPT maintains stable performance under simulated and real-world missingness without imputation. Additionally, TAP-GPT produces structured, modality-aware reasoning aligned with established AD biology and shows greater stability under self-reflection, supporting its use in iterative multi-agent systems. To our knowledge, this is the first systematic application of a tabular-specialized LLM to multimodal biomarker-based AD prediction, demonstrating that such pretrained models can effectively address structured clinical prediction tasks and laying the foundation for tabular LLM-driven multi-agent clinical decision-support systems. The source code is publicly available on GitHub: https://github.com/sophie-kearney/TAP-GPT.
Sophie Kearney, Shu Yang, Zixuan Wen +8
University of Pennsylvania, Philadelphia, PA, USA · Stony Brook University, Stony Brook, NY, USA · University of North Carolina at Chapel Hill, Chapel Hill, NC, USA +2