Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

OpenTCM: A GraphRAG-Empowered LLM-based System for Traditional Chinese Medicine Knowledge Retrieval and Diagnosis

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read OpenTCM claims a knowledge-graph retrieval system can match or beat fine-tuned TCM models in expert-rated ingredient lookup and diagnosis, without any model fine-tuning.

desk verdict OpenTCM's knowledge graph is a real asset, but the headline evaluation numbers are mutually inconsistent as defined, so the central claim needs a serious rework. read the letter →

arxiv 2504.20118 v4 pith:TTZIPYR7 submitted 2025-04-28 cs.IR cs.AI

Jinglin He , Yunqi Guo , Lai Kwan Lam , Waikei Leung , Lixing He , Yuanan Jiang , Chi Chiu Wang , Guoliang Xing
show 1 more author
Hongkai Chen
This is my paper · ORCID
classification cs.IRcs.AI
keywords TraditionalChineseMedicineKnowledgeGraphRAGLargeLanguageModelsRetrieval-AugmentedGenerationClassicalTextIngredientRetrievalDiagnosticQuestionAnswering
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

OpenTCM sets out to show that a large, structured knowledge graph of classical Traditional Chinese Medicine texts, combined with graph-based retrieval, lets an LLM answer TCM ingredient and diagnosis questions as well as or better than fine-tuned TCM models. The system builds a 48,000-entity, 152,000-relationship graph from 68 ancient gynecological books using customized prompts and Chinese-oriented LLMs, then uses GraphRAG to retrieve from that graph instead of from flat text. On expert-rated tests, OpenTCM scores 4.378 of 5 for ingredient retrieval and 4.045 for diagnostic question-answering, with accuracy near 99 percent, ahead of general-purpose and TCM-specific baselines. If correct, this means classical Chinese medical knowledge can be made practically accessible without expensive domain-specific fine-tuning.

What carries the argument

The load-bearing mechanism is GraphRAG applied to a hand-verified medical knowledge graph: retrieval-augmented generation that answers a query by traversing a multi-relational graph rather than by matching text chunks. OpenTCM's graph encodes ten relation types among 48,406 entities (ingredients, diseases, symptoms, treatments, and bibliographic units like books and chapters), enabling multi-hop paths such as symptom → syndrome → treatment → ingredient. The authors construct the graph automatically with customized system prompts that direct LLMs to act as 'TCM knowledge analysis assistants' and return JSON triples, then validate a random sample of 1,795 triples with four TCM and gynecology experts, reporting precision of 98.55 percent and recall of 99.60 percent for the best prompt and LLM combination.

What would settle it

Have a second, independent panel of at least four TCM experts score the same 257 ingredient and 303 diagnostic responses under the same 5-point rubric, and compute mean scores and binary inter-rater agreement. If OpenTCM's margin over Kimi and BianCang mostly disappears on the new panel, or if the new panel's inter-rater agreement again lands near zero, then the reported superiority is specific to the original expert panel rather than the system.

Watch

Extended reading notes

Core claim

The central claim is that the quality bottleneck for TCM LLM systems is not the generator but the retrieval structure: by extracting typed triples from classical text with domain-tailored prompts and Chinese-language LLMs, then letting a graph walker retrieve along relations such as 'treats' and 'uses ingredient,' OpenTCM yields responses that four domain experts rate as highly relevant for 99.0 percent of ingredient queries and 98.8 percent of diagnostic queries. This performance is achieved without fine-tuning the underlying LLM, and the comparison suggests that the knowledge graph itself, not extra training, carries the performance. The authors further report that their customized prompt design raises triple-extraction precision from about 90 percent to 98.55 percent and recall from about 96 percent to 99.60 percent, establishing the graph as a faithful structured mirror of the classical corpus.

Load-bearing premise

The evaluation depends entirely on four experts' ratings as the ground truth for what counts as a correct answer, and those experts barely agree with each other (inter-rater agreement of 0.057 for ingredient retrieval and -0.013 for diagnostic question-answering).

Editorial extensions

If this is right

  • Because OpenTCM requires no fine-tuning, porting the system to a new classical corpus reduces to rebuilding the knowledge graph with the same prompt pipeline and expert verification.
  • The reported margin over fine-tuned baselines implies that structured external memory can substitute for much of the domain knowledge otherwise embedded in model weights, making expensive TCM-specific pre-training unnecessary.
  • Graph-grounded retrieval yields an auditable reasoning path: every answer can be traced through a chain of graph relations (e.g., symptom → treatment → ingredient), which supports clinical and educational use.
  • The large reported margin over BianCang and Kimi indicates that the knowledge graph, not the underlying LLM, is the primary source of TCM competence in this system.
  • The constructed graph itself is a reusable asset for other downstream tasks, such as exploring ingredient-disease associations or navigating the classical literature.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the graph is truly the performance driver, then deliberately corrupting or removing a random 10 percent of the relations should measurably lower expert scores, a stress test the paper does not run.
  • The near-zero inter-rater agreement reported in the paper implies the official MES and accuracy numbers are likely panel-specific; a fairer comparison would report scores from several independent expert panels or a pre-registered scoring rubric.
  • Because the approach is training-free, the same architecture could be transferred to other expert-curated classical knowledge bases, such as Ayurvedic or medieval European medical texts, suggesting a general pattern for bridging archaic specialized corpora and LLMs.
  • The paper's contrast between Chinese-oriented and general LLMs in extraction quality hints that the choice of LLM matters most at the parsing stage, while downstream reasoning may be relatively model-agnostic; a controlled test separating these stages would clarify where the gains actually come from.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper introduces OpenTCM, a system that combines a knowledge graph automatically extracted from 68 classical Chinese gynecological texts with GraphRAG to support TCM ingredient retrieval and diagnostic question-answering without model fine-tuning. The authors report that OpenTCM achieves a mean expert score (MES) of 4.378 with 99.0% accuracy for ingredient retrieval and 4.045 with 98.8% accuracy for diagnostic QA, outperforming the KIMI and BianCang baselines. The paper also reports high precision, recall, F1, and accuracy figures for the knowledge-graph construction step.

Significance. If the reported results are reliable, OpenTCM offers a substantial and practical contribution: a training-free, graph-grounded pipeline for a specialized classical-medicine domain, built on a large curated corpus and a multi-relational knowledge graph. The system's avoidance of fine-tuning and its use of Chinese-oriented LLMs for classical text extraction are sensible design choices with potential value for other historical or low-resource medical corpora. However, the evaluation as reported contains an internal inconsistency between the near-zero inter-rater agreement and the near-100% accuracy figures, and the absence of significance tests and a limited baseline set leave the central comparative claim insufficiently supported. The paper also provides a public repository link, which is a strength if the code and data are actually released, but the raw rating data needed to resolve the metric inconsistency are not provided.

major comments (5)
  1. [Section IV-C, Tables VI and VII] The reported accuracy and inter-rater agreement are mutually inconsistent under the definitions given in Section IV-C. Accuracy is defined as the proportion of expert-labeled correct responses, where a response is correct if its expert score is >=3, and IRA is Fleiss' kappa on the binarized scores. For ingredient retrieval there are 257 queries x 4 experts = 1028 ratings; an accuracy of 99.0% implies that roughly 99% of these binarized ratings are "correct", which would force near-unanimous agreement and a kappa close to 1, not 0.057. The same contradiction holds for diagnostic QA (98.8% accuracy versus kappa -0.013). The only way to reconcile these numbers is if accuracy is computed at the query level (e.g., a query counts as correct when the average or majority of expert scores is >=3), but that is not what the text states. The authors must re-specify the accuracy denominator, provide the per-expert rating matrix or per-query binarized labels, and recompute all headline metrics accordingly.
  2. [Section IV-C, Tables VI and VII] No significance testing or confidence intervals are reported for the MES differences or accuracy gaps. The claim that OpenTCM "outperforms" KIMI and BianCang rests on point estimates (4.378 versus 2.691 and 4.045 versus 3.043) without any paired test across the 257 or 303 queries or across raters. Given the near-zero IRA, the reader cannot rule out that the gap is driven by a single lenient rater or a few outlier queries. Please report per-expert MES and per-rater accuracy, and run a paired test (e.g., Wilcoxon signed-rank or permutation test) on query-level mean scores.
  3. [Section IV-B, Table V] The knowledge-graph construction evaluation reports precision, recall, F1, and accuracy of 98% or higher based on 1,795 expert-annotated triples, but no inter-rater agreement is reported for these annotations, no confidence intervals are given, and the adjudication procedure for expert disagreements is not described. Without this information, the 98-99% figures are not reproducible and the fidelity of the KG as the foundation for downstream tasks is not established.
  4. [Section IV-C] Only two baselines are evaluated in the real-world applications (KIMI and BianCang), and there is no ablation isolating the contribution of the graph structure from the corpus or the LLM. A text-RAG variant over the same classical corpus, or a GraphRAG variant without the curated KG, would let the reader attribute the MES gain to the proposed architecture rather than to the prompt design, the knowledge base, or the generation model. Such an ablation is necessary to support the claim that GraphRAG is the key enabler of the reported performance.
  5. [Section IV-C, dataset description] The origin of the 257 ingredient queries and 303 diagnostic questions is not described: the paper does not state how these queries were collected, whether they are independent of the 68 books used to build the KG, or how diverse they are. If the queries are drawn from the same corpus that was used for KG construction, the retrieval and QA results may overestimate performance on out-of-corpus generalization. Please clarify the query collection procedure and, if feasible, include an out-of-corpus test set.
minor comments (5)
  1. [Table IV caption] The caption reads "Customized Prompt Verses General Prompt"; this should be "Versus".
  2. [Throughout] Model naming is inconsistent: "Deepseek" versus "DeepSeek" and "KIMI" versus "Kimi" appear in different places. Please standardize the names.
  3. [Section III-B, Figure 2] The text states that "The whole TCM Knowledge Graph and two partial subgraphs are visualized in Figures 2", but Figure 2 appears to contain a single visualization. Please clarify the figure contents or correct the citation.
  4. [Tables VI and VII] The baseline IRA values are negative (-0.154, -0.166, -0.331, -0.275), but the paper does not interpret what negative Fleiss' kappa means in this context or whether these values are significantly different from zero. A brief discussion would help the reader assess rater reliability for the baselines as well.
  5. [Abstract and Section I] The abstract and introduction report the same headline numbers as Section IV, but the inter-rater agreement values are only mentioned in the abstract without definition. Since these values are central to the evaluation's credibility, a one-sentence explanation of IRA in the introduction would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; the reported metrics are empirical expert ratings, not quantities defined in terms of the target result.

full rationale

OpenTCM's knowledge graph is constructed from 68 classical gynecological texts via LLM extraction with expert verification, and its downstream retrieval and diagnostic QA performance is measured by expert ratings on 257 ingredient queries and 303 diagnostic questions. No parameter is fitted to those ratings, no trained model is claimed, and no self-authored benchmark is used as ground truth; the system is training-free and its outputs are generated by GraphRAG over the constructed KG. The self-citations in the paper (SHADE-AD, DrHouse, EmoMarker, MyoTrainer) appear only in related-work or future-work contexts and are not load-bearing for the central claims. The reported low inter-rater agreement (IRA 0.057 for ingredient retrieval and -0.013 for diagnostic QA) is a serious evaluation-validity concern, and the stated accuracy values appear difficult to reconcile with those IRA values under the paper's own metric definitions, but this is a measurement and reporting inconsistency, not circularity. Because no derived claim reduces by construction to its inputs, the circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No numeric free parameters are fitted because the system uses off-the-shelf LLMs and no fine-tuning. The evaluation setup does include subjective thresholds (e.g., score >= 3 deemed correct), but those are analysis choices, not model parameters. The central claim depends on domain assumptions about expert validity and the authority of the classical corpus.

assumptions (3)
  • domain assumption Expert evaluation is a valid measure of answer quality
    The entire evaluation of ingredient retrieval and diagnostic QA rests on the four experts' 5-point Likert scores. No external gold standard is used, and the near-zero inter-rater agreement raises doubt about this assumption.
  • domain assumption Classical TCM texts from the Chinese Medical Classics Database are authoritative for diagnosis and treatment
    The corpus is selected from one database and treated as ground truth for the knowledge graph without comparison to modern evidence-based medicine or independent validation.
  • domain assumption The LLM-extracted knowledge graph faithfully represents the source texts
    Only a sample of 1795 triples out of 152,754 is expert-validated; the remainder is assumed correct, and no error analysis is provided for the unvalidated portion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OpenTCM: A GraphRAG-Empowered LLM-based System for Traditional Chinese Medicine Knowledge Retrieval and Diagnosis." pith.science (2026). https://pith.science/paper/TTZIPYR7

@misc{pith2026250420118,
  author       = {Pith},
  title        = {Pith review of: OpenTCM: A GraphRAG-Empowered LLM-based System for Traditional Chinese Medicine Knowledge Retrieval and Diagnosis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TTZIPYR7}},
  note         = {Machine review of arXiv:2504.20118}
}
read the original abstract

Traditional Chinese Medicine (TCM) represents a rich repository of ancient medical knowledge that continues to play an important role in modern healthcare. Due to the complexity and breadth of the TCM literature, the integration of AI technologies is critical for its modernization and broader accessibility. However, this integration poses considerable challenges, including the interpretation of obscure classical Chinese texts and the modeling of intricate semantic relationships among TCM concepts. In this paper, we develop OpenTCM, an LLM-based system that combines a domain-specific TCM knowledge graph and Graph-based Retrieval-Augmented Generation (GraphRAG). First, we extract more than 3.73 million classical Chinese characters from 68 gynecological books in the Chinese Medical Classics Database, with the help of TCM and gynecology experts. Second, we construct a comprehensive multi-relational knowledge graph comprising more than 48,000 entities and 152,000 interrelationships, using customized prompts and Chinese-oriented LLMs such as DeepSeek and Kimi to ensure high-fidelity semantic understanding. Last, we empower OpenTCM with GraphRAG, enabling high-fidelity ingredient knowledge retrieval and diagnostic question-answering without model fine-tuning. Experimental evaluations demonstrate that OpenTCM achieves mean expert scores (MES) of 4.378 in ingredient information retrieval and 4.045 in diagnostic question-answering tasks, outperforming state-of-the-art solutions in real-world TCM use cases.

Figures

Figures reproduced from arXiv: 2504.20118 by the authors.

Figure 1
Figure 1. System architecture of OpenTCM, which consists of three main components: (1) data collection and pre-processing, (2) knowledge graph construction, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Visualization of our TCM knowledge graph. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. An example case of OpenTCM diagnostic question-answering. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Score distribution of different models for Ingredient Knowledge Retrieval across four experts. OpenTCM shows significantly higher concentration at [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Score distribution of different models for Diagnostic Question Answering across four experts. OpenTCM responses receive notably higher expert [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DeepFeature: LLM-Empowered Context-aware Feature Generation for Wearable Biosignals

    cs.AI 2025-12 conditional novelty 6.0 of 10

    An LLM-based pipeline that generates context-aware features from raw wearable biosignals and iteratively refines them via performance feedback reports AUROC gains of 4-10% over baselines.

Reference graph

Works this paper leans on

25 extracted references · 14 canonical work pages · cited by 1 Pith paper

  1. [1]

    Traditional chinese medicine,

    J.-L. Tang, B.-Y . Liu, and K.-W. Ma, “Traditional chinese medicine,” The Lancet, vol. 372, no. 9654, pp. 1938–1940, 2008

    work page 1938

  2. [2]

    Biancang: A traditional chinese medicine large language model,

    S. Wei, X. Peng, Y .-f. Wang, J. Si, W. Zhang, W. Lu, X. Wu, and Y . Wang, “Biancang: A traditional chinese medicine large language model,” arXiv preprint arXiv:2411.11027 , 2024

    arXiv 2024

  3. [3]

    Lingdan: enhancing encoding of traditional chinese medicine knowledge for clinical reasoning tasks with large language models,

    R. Hua, X. Dong, Y . Wei, Z. Shu, P. Yang, Y . Hu, S. Zhou, H. Sun, K. Yan, X. Yan et al. , “Lingdan: enhancing encoding of traditional chinese medicine knowledge for clinical reasoning tasks with large language models,” Journal of the American Medical Informatics As- sociation, vol. 31, no. 9, pp. 2019–2029, 2024

    work page 2019

  4. [4]

    Llama 2: Open foundation and fine-tuned chat models,

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale et al. , “Llama 2: Open foundation and fine-tuned chat models,” arXiv preprint arXiv:2307.09288, 2023

    arXiv 2023

  5. [5]

    Ai empowering traditional chinese medicine?

    Z. Song, G. Chen, and C. Y .-C. Chen, “Ai empowering traditional chinese medicine?” Chemical Science , vol. 15, no. 41, pp. 16 844– 16 886, 2024

    work page 2024

  6. [6]

    Ai: Bridging ancient wisdom and modern innovation in traditional chinese medicine,

    L. Lu, T. Lu, C. Tian, X. Zhang et al. , “Ai: Bridging ancient wisdom and modern innovation in traditional chinese medicine,” JMIR Medical Informatics, vol. 12, no. 1, p. e58491, 2024

    work page 2024

  7. [7]

    Medical graph rag: Towards safe medical large lan- guage model via graph retrieval-augmented generation,

    J. Wu, J. Zhu, Y . Qi, J. Chen, M. Xu, F. Menolascina, and V . Grau, “Medical graph rag: Towards safe medical large lan- guage model via graph retrieval-augmented generation,” arXiv preprint arXiv:2408.04187, 2024

    arXiv 2024

  8. [8]

    From local to global: A graph rag approach to query-focused summarization,

    D. Edge, H. Trinh, N. Cheng, J. Bradley, A. Chao, A. Mody, S. Truitt, D. Metropolitansky, R. O. Ness, and J. Larson, “From local to global: A graph rag approach to query-focused summarization,” arXiv preprint arXiv:2404.16130, 2024

    arXiv 2024

Show all 25 references
  1. [9]

    Collections of traditional chinese medical literature as resources for systematic searches,

    B. H. May, C. Lu, and C. C. Xue, “Collections of traditional chinese medical literature as resources for systematic searches,” The Journal of Alternative and Complementary Medicine , vol. 18, no. 12, pp. 1101– 1107, 2012

    2012

  2. [10]

    Deepseek-v3 technical report,

    A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruan et al. , “Deepseek-v3 technical report,” arXiv:2412.19437, 2024

    2024 arXiv

  3. [11]

    Kimi k1. 5: Scaling reinforcement learning with llms,

    K. Team, A. Du, B. Gao, B. Xing, C. Jiang, C. Chen, C. Li, C. Xiao, C. Du, C. Liao et al., “Kimi k1. 5: Scaling reinforcement learning with llms,” arXiv preprint arXiv:2501.12599 , 2025

    2025 arXiv

  4. [12]

    Traditional chinese medicine knowledge graph construction based on large language models,

    Y . Zhang and Y . Hao, “Traditional chinese medicine knowledge graph construction based on large language models,”Electronics, vol. 13, no. 7, 2024

    2024

  5. [13]

    Research on a traditional chinese medicine case-based question-answering system integrating large language models and knowledge graphs,

    Y . Duan, Q. Zhou, Y . Li, C. Qin, Z. Wang, H. Kan, and J. Hu, “Research on a traditional chinese medicine case-based question-answering system integrating large language models and knowledge graphs,” Frontiers in Medicine, vol. 11, 01 2025

    2025

  6. [14]

    SHADE-AD: An llm-based framework for synthesizing activity data of Alzheimer’s patients,

    H. Fu, H. Chen, S. Lin, and G. Xing, “SHADE-AD: An llm-based framework for synthesizing activity data of Alzheimer’s patients,” in Proceedings of the 23rd ACM Conference on Embedded Networked Sensor Systems (SenSys’25) . ACM, 2025

    2025

  7. [15]

    Qibo: A large language model for traditional chinese medicine,

    H. Zhang, X. Wang, Z. Meng, Z. Chen, P. Zhuang, Y . Jia, D. Xu, and W. Guo, “Qibo: A large language model for traditional chinese medicine,” 2024. [Online]. Available: https://arxiv.org/abs/2403.16056

    2024 arXiv

  8. [16]

    Zhongjing: Enhancing the chinese medical capabilities of large lan- guage model through expert feedback and real-world multi-turn dia- logue,

    S. Yang, H. Zhao, S. Zhu, G. Zhou, H. Xu, Y . Jia, and H. Zan, “Zhongjing: Enhancing the chinese medical capabilities of large lan- guage model through expert feedback and real-world multi-turn dia- logue,” in Proceedings of the AAAI conference on artificial intelligence , vol...

    2024

  9. [17]

    Tcmchat: A generative large language model for traditional chinese medicine,

    Y . Dai, X. Shao, J. Zhang, Y . Chen, Q. Chen, J. Liao, F. Chi, J. Zhang, and X. Fan, “Tcmchat: A generative large language model for traditional chinese medicine,” Pharmacological Research , vol. 210, p. 107530, 2024

    2024

  10. [18]

    Graph retrieval-augmented generation: A survey,

    B. Peng, Y . Zhu, Y . Liu, X. Bo, H. Shi, C. Hong, Y . Zhang, and S. Tang, “Graph retrieval-augmented generation: A survey,” arXiv preprint arXiv:2408.08921, 2024

    2024 arXiv

  11. [19]

    Retrieval-augmented generation with graphs (GraphRAG),

    H. Han, Y . Wang, H. Shomer, K. Guo, J. Ding, Y . Lei, M. Halappanavar, R. A. Rossi, S. Mukherjee, X. Tang et al. , “Retrieval-augmented generation with graphs (GraphRAG),” arXiv:2501.00309, 2024

    2024 arXiv

  12. [20]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat et al. , “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774 , 2023

    2023 arXiv

  13. [21]

    A comparative study of open-source large language models, gpt-4 and claude 2: Multiple-choice test taking in nephrology,

    S. Wu, M. Koo, L. Blum, A. Black, L. Kao, F. Scalzo, and I. Kurtz, “A comparative study of open-source large language models, gpt-4 and claude 2: Multiple-choice test taking in nephrology,” arXiv preprint arXiv:2308.04709, 2023

    2023 arXiv

  14. [22]

    Emomarker: A privacy-preserving, multi-modal sensing system for dyadic digital biomarkers of expressed emotions for patients with dementia,

    Y . Li, D. S. F. Yu, S. Chen, G. Xing, and H. Chen, “Emomarker: A privacy-preserving, multi-modal sensing system for dyadic digital biomarkers of expressed emotions for patients with dementia,” in Proceedings of the 22nd Annual International Conference on Mobile Systems, Appli...

    2024

  15. [23]

    Drhouse: An LLM-empowered diagnostic reasoning system through harnessing outcomes from sensor data and expert knowledge,

    B. Yang, S. Jiang, L. Xu, K. Liu, H. Li, G. Xing, H. Chen, X. Jiang, and Z. Yan, “Drhouse: An LLM-empowered diagnostic reasoning system through harnessing outcomes from sensor data and expert knowledge,” Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiqui- tous ...

    2024

  16. [24]

    Retcare: Towards interpretable clinical decision making through llm-driven medical knowledge retrieval,

    Z. Wang, Y . Zhu, J. Gao, X. Zheng, Y . Zeng, Y . He, B. Jiang, W. Tang, E. M. Harrison, C. Pan et al. , “Retcare: Towards interpretable clinical decision making through llm-driven medical knowledge retrieval,” in Artificial Intelligence and Data Science for Healthcare: Bridgi...

    2024

  17. [25]

    Myotrainer: Muscle-aware motion analysis and feedback system for in- home resistance training,

    Y . He, X. Wang, M. Yuan, D. Duan, D. S. F. Yu, G. Xing, and H. Chen, “Myotrainer: Muscle-aware motion analysis and feedback system for in- home resistance training,” in Proceedings of the 22nd ACM Conference on Embedded Networked Sensor Systems (SenSys’24) . ACM, 2024

    2024

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.